If you haven’t subscribed, you can join 6,420 others who enjoy our deep dives and investment insights here:

Subscribe now

Wholesale investors looking to invest in the Bristlemoon Global Fund can do so via this link.

It has been a while since we’ve written publicly about a Bristlemoon short, the last writeup of which was for National Vision Holdings (EYE) back in January 2024. However, it has been for an even longer period that we’ve been itching to short U.S. department stores. While the likes of Macy’s and Kohl’s might be household names, from a business standpoint they’re melting ice cubes, with their revenues and earnings shrinking each year. However, since the inception of the Fund, we have demurred from shorting these embattled department store retailers. They always looked dangerously cheap, and we lacked a divergent view. This has now changed, and we have recently initiated shorts in Macy’s (M) and Kohl’s (KSS).

N.b.: the majority of the discussion will revolve around M, with only passing details provided on KSS.

Key takeaways

• U.S. department stores remain melting ice cubes, but for the first time in years we think the valuation and earnings setup makes them attractive shorts. Macy’s and Kohl’s have spent the last decade shrinking, yet both have recently re-rated as operating trends have improved; M now trades at 10.5x earnings versus a five-year average of 6.6x.

• We think the market is over-extrapolating Macy’s recent comp improvement. The Reimagine strategy has worked, but 200 refreshed stores already represent ~60% of the go-forward store base and ~75% of Macy’s nameplate sales, meaning much of the low-hanging fruit has already been picked just as the company begins lapping tougher comparisons.

• Macy’s headline profitability also flatters the health of the underlying retail business. Credit-card revenues equated to around three quarters of Macy’s FY25 adjusted EBIT, with recent growth benefiting from a normalization in funding costs and credit losses. That growth is already decelerating, while the card business adds another source of procyclicality if consumer credit deteriorates.

• Meanwhile, transportation inflation represents a new margin headwind that investors may be underappreciating. On our illustrative assumptions, higher inbound truckload and outbound parcel costs could create roughly $115 million of incremental expense and reduce operating earnings by ~13%, before offsets. Putting these factors together, we can get to FY27E EPS of ~$1.50 in a bear case scenario, 35% below sellside consensus; at 8x earnings this implies a ~$12 share price, more than 50% below current levels.

Melting ice cubes

Consider that Macy’s sales growth has been negative in all but two of the past 10 years. Kohl’s topline performance has not been much better.

Source: Bristlemoon Capital; Company filings

U.S. department stores, in a bygone era, used to be excellent businesses. However, the advent of e-commerce has changed this, and the probability of shoppers reverting en masse to the old (analog) way of purchasing goods rounds down to zero.

M, somewhat tragically for those with nostalgic tendencies, has shuttered 41% of its Macy’s branded store base over the last decade. KSS, somewhat defiantly, has closed only 1% of its Kohl’s stores over that period, although the average revenue per store has fallen by one fifth over that decade span. However, these sales declines only tell half of the story.

As one would expect with the sales deleverage on a fixed cost base, the margins of these department stores have also compressed, shedding roughly 4.5 percentage points over the last decade, falling from around 13% in FY15 to c.8.5% today.

Source: Bristlemoon Capital; Company filings

It’s fairly consensus that these are challenged businesses, and for years the companies traded at single-digit P/E multiples. However, more recently, both M and KSS have re-rated and are trading at a premium to where their multiples have hovered historically. M, for example, is trading at 10.5x P/E, a whopping 2.1 standard deviations above the average P/E over the last five years of 6.6x.

Source: Bristlemoon Capital; Bloomberg

So, what has led to the recent re-rate in these stocks?

Read more

If you haven’t subscribed, you can join 6,296 others who enjoy our deep dives and investment insights here:

Subscribe now

Wholesale investors looking to invest in the Bristlemoon Global Fund can do so via this link.

In recent months, the market has become increasingly concerned about the amount of AI capex being spent and the capex spenders’ ability to earn a return on their investment. The hyperscalers are on track to spend over $750 billion in capex this calendar year, and well over $1 trillion next year once sellside update their capex estimates following Q2 earnings season. Management teams continue to reiterate their qualitatively bullish expectations for ROI but remain unwilling or unable to quantify the returns they expect from these monumental investments. So, unsurprisingly, the stock prices of these capex upgraders have come under mounting pressure, and they are no longer being rewarded for earnings beats.

Source: Bloomberg consensus estimates. Note Amazon includes capex for the retail business in addition to AWS

The common refrain we see and hear is that there’s no way the hyperscalers can earn an acceptable return on $1 trillion a year of AI capex, and skeptics will point to the current pace of capex outlay and ask “well, where are the returns?!” Indeed, a naïve view of the financials supports this skepticism.

Using Alphabet (henceforth Google) as an example, the company has spent $132 billion in capex in the LTM to June 2026, while LTM operating income has increased by $26 billion. This equates to a 16% post-tax incremental ROIC – still above the company’s cost of capital, but a far cry from the asset-light growth engine it once was.

Source: Bristlemoon Capital, company filings

We would argue that the above naïve interpretation of ROIC is way off the mark. To understand why, one must realize that not all the capex outlay in a period is placed into service immediately, and obviously assets that are purchased but not yet tested, commissioned and placed into service are not generating any revenues. Therefore, while Google spent $132 billion on capex LTM, we estimate that only $82 billion was placed into service, or $57 billion net of depreciation.

Source: Bristlemoon Capital, company filings

If we use the change in average, instead of period-end, net PP&E in service, the $57 billion falls to $43 billion, which in our opinion is the correct denominator to use for calculating Google’s LTM incremental ROIC. This works out to be 49% post-tax compared to just 16% for the naïve calculation. From this perspective, it is perhaps easier to understand why the hyperscaler management teams remain so bullish about their AI capex investments.

Source: Bristlemoon Capital, company filings

While it is undeniably true that Google’s returns have declined over recent years (though note 2024 was flattered by easy comps coming out of a tough H2 2022-H1 2023), a ~50% incremental ROIC is still extremely healthy for a business of this size, and reinvesting as much as Google is. It is also notable that incremental ROIC troughed in Q4 2025 and has expanded the past two quarters even as quarterly capex has nominally increased.

Subscribe now

Finally, there is an interesting question for the finance theory nerds: would you rather own a business with 120% incremental ROIC and 10% reinvestment rate, or a business with 50% incremental ROIC but 100% reinvestment rate? Holding all other assumptions about the future constant (big if – we’ll revisit that), it should be pretty evident which business will produce the better long-term future return for owners.

On the topic of future returns…

Thus far, we’ve been looking backwards at Google’s ROIC as it ramped up its AI investments. Turning to the future, we have $123 billion in assets not yet in service sitting on the balance sheet. The not yet in service backlog has averaged around four quarters over the past three years, so we can expect that $123 billion (or ~$110 billion net of ~10% depreciation rate) will be placed into service by Q2 2027. We will use pre-tax ROIC here to make it easier to compare to Bloomberg operating profit consensus, but essentially: if we assume Google maintains the 60% incremental ROIC it achieved LTM to Q2 2026 and places all $110 billion of existing net capex into service by Q2 2027, its NTM operating income should increase by over $60 billion.

This is well ahead of current Bloomberg consensus, which has operating income increasing by $43 billion over the NTM and incremental ROIC falling to the low-40s (our estimate). More importantly, it also doesn’t include any operating income uplift from i) additional assets placed into service from NTM capex, ii) TPU hardware sales, which the CFO has said will be material in 2027, and iii) any earnings from the $920 million-per-month compute rental deal with SpaceX, which Google would not have signed if it were not extremely constrained on compute.

Source: Bristlemoon Capital, Bloomberg consensus

Now, we acknowledge that the assumption that Google can maintain 60% or even 50% incremental returns as it deploys well over $100 billion in assets over the next year is carrying a lot of weight in the above analysis. Certainly, part of the rising capex is memory price inflation, which is essentially a tax that drags on incremental returns unless GCP can pass that cost entirely onto customers. Even if we assume $10 billion of the assets not yet in service is unproductive memory inflation, it will only reduce incremental returns by mid-single-digits, and evidently management sees enough upside to continue ramping investments notwithstanding extreme hardware inflation.

The greater uncertainty, and clearly the key piece of the entire ROI puzzle, is where token demand caps out. Hyperscalers have emphatically claimed that demand for compute exceeds available supply, but we cannot infer from that alone the delta between demand and supply. If demand for frontier intelligence is truly unlimited (we don’t think it is), Google should have no trouble sustaining or even expanding its 60% incremental ROIC as it brings hundreds of billions of compute assets online.

Realistically, the relationship between token demand and supply is likely to be very fluid and highly price elastic, especially for near-frontier intelligence. As such, we have no idea where token demand might start to asymptote. Our high-level inference analysis suggests that frontier flagship model API pricing (5.6 Sol or Fable) generates gross margins in the 90s – plausibly in the high 90s – and smaller more cost-effective models are still delivering extremely attractive inference economics.

If broader compute availability puts pressure on token pricing, we believe pricing can come down significantly while still allowing for very good returns on the compute investment. For example, at 98% gross margin, API pricing can halve and GM will only fall to 96%, while the resulting increase in token demand can still drive healthy gross profit dollar growth. In such a world, incremental return on compute investment need not even come down if hardware and software efficiencies combine to enable multiple-fold increases in token throughput.

In any case, Google placed an estimated $33 billion of assets into service in Q2, so we will all find out soon enough what kind of incremental return that can earn!

Subscribe now

Disclaimer / Disclosures

The information contained in this article is not investment advice and is intended only for wholesale investors. All posts by Bristlemoon Capital are for informational purposes only. This article has been prepared without taking into account your particular circumstances, nor your investment objectives and needs. This article does not constitute personal investment advice and you should not rely on it as such. This document does not contain all of the information that may be required to evaluate an investment in any of the securities featured in the document. We recommend that you obtain independent financial advice before you make investment decisions.

Forward-looking statements are based on current information available to the author, expectations, estimates, projections and assumptions as to future matters. Forward-looking statements are subject to risks, uncertainties and other known and unknown factors and variables, which may affect the accuracy of any forward-looking statement. No guarantee is made in relation to future performance, results or other events.

We make no representation and give no warranties regarding the accuracy, reliability, completeness or suitability of the information contained in this document. To the maximum extent permitted by law, we do not have any liability for any loss or damage suffered or incurred by any person in connection with this document.

Bristlemoon Capital Pty Ltd (ABN: 22 668 652 926) is an Australian Financial Services Licensee (AFSL Number: 552045).

George Hadjia and Daniel Wu are associated with Bristlemoon Capital Pty Ltd. Bristlemoon Capital may invest in securities featured in this newsletter from time to time.

In the quarterly letter below, we reflect on the Fund’s first two years and the lessons learned from a difficult twelve-month period. In particular, we discuss the following:

• Position sizing and averaging down: We examine the risks of averaging down based on valuation alone, and explain our compounding/conviction framework, which is designed to ensure that we only size up positions when we have a well-founded view that the business’s fundamentals are positively inflecting. This framework seeks to better align our position sizing with how stocks are trading in the current market regime.

• Market structure and portfolio management: We discuss how market structure changes have contributed to greater narrative-driven volatility, with stock multiples now de-rating and re-rating more violently than in the past. This has important implications for portfolio management, and reinforces the need to be more disciplined in when we add to positions, as well as more willing to trim or exit positions when a stock has pulled forward a meaningful portion of its future returns.

• Case studies and portfolio updates: We use Timee as an example of our framework beginning to work in practice, and AppLovin as a case study in the importance of recognising when consensus expectations have caught up to our own. Finally, we provide a detailed update on ASML, where we believe the market is still underestimating the scale of the coming wafer fab equipment capex supercycle and the resulting demand for EUV tools. To substantiate this view, we include a detailed bottom-up build-up of EUV tool demand across advanced logic and DRAM, which leads us to shipment estimates that are materially ahead of sell-side consensus for 2027 and 2028. While ASML has nearly doubled since we first wrote about it in our September 2025 quarterly letter, we continue to see healthy upside from here, and the stock remains a core position in the Fund.

As always, we welcome any thoughts, questions, or pushback on the letter. For prospective investors interested in learning more about the Fund, further information is available here, or we would be pleased to hear from you at info@bristlemoon.com.

By accessing the link below, you acknowledge and agree that you are an Australian wholesale client as defined by Section 761G of the Corporations Act 2001.

Bristlemoon Global Fund June 2026 Quarterly Report Vf

905KB ∙ PDF file

Download

Download

Subscribe now

Disclaimer / Disclosures

The information contained in this article is not investment advice and is intended only for wholesale investors. All posts by Bristlemoon Capital are for informational purposes only. This article has been prepared without taking into account your particular circumstances, nor your investment objectives and needs. This article does not constitute personal investment advice and you should not rely on it as such. This document does not contain all of the information that may be required to evaluate an investment in any of the securities featured in the document. We recommend that you obtain independent financial advice before you make investment decisions.

Forward-looking statements are based on current information available to the author, expectations, estimates, projections and assumptions as to future matters. Forward-looking statements are subject to risks, uncertainties and other known and unknown factors and variables, which may affect the accuracy of any forward-looking statement. No guarantee is made in relation to future performance, results or other events.

We make no representation and give no warranties regarding the accuracy, reliability, completeness or suitability of the information contained in this document. To the maximum extent permitted by law, we do not have any liability for any loss or damage suffered or incurred by any person in connection with this document.

Bristlemoon Capital Pty Ltd (ABN: 22 668 652 926) is an Australian Financial Services Licensee (AFSL Number: 552045).

George Hadjia and Daniel Wu are associated with Bristlemoon Capital Pty Ltd. Bristlemoon Capital may invest in securities featured in this newsletter from time to time.

If you haven’t subscribed, you can join 6,190 others who enjoy our deep dives and investment insights here:

Subscribe now

Australian wholesale investors looking to invest in the Bristlemoon Global Fund can do so via this link.

Introduction

While the US trucking industry is likely considered outside the purview of investors seeking to invest in high-quality compounder businesses (it’s too cyclical and competitive), our experience is that a lot of opportunities (both long and short) tend to present when there are turning points in the US transportation market. After all, physical goods still need to move across the US economy, and shifts in transportation costs can have meaningful second-order effects on the margins of the companies that produce, distribute and sell them.

It is thus worth diving into some recent developments in the US transport and logistics industry that are causing trucking spot rates to skyrocket. Since late 2025, there has been a revenue per mile (RPM) rally, with ex-fuel van spot rates increasing by around 30% year-over-year (YoY) in May 2026, which followed increases of 11%, 23%, 20%, and 25% across January to April. This is the strongest stretch of performance since 2021.

Source: Bristlemoon Capital, DAT Solutions, Bloomberg

Notably these increases in trucking spot rates are excluding fuel surcharges (ex-fuel rates are referred to as “linehaul” rates). When we include the 60-70 cents of fuel surcharges that have been tacked onto trucking rates since March 2026, then all-in spot rates have been increasing at an almost 50% YoY clip in the recent weeks of June 2026.

For any company that needs to move goods, a 50% increase in the rate you’re paying to transport those items versus the year prior is clearly meaningful, so it’s worth us digging into what has caused this material spike in spot rates.

Some background on the trucking cycle

Before we look into what has catalyzed the current transportation market tightness, it is worth rewinding and looking at developments in the last few years that have affected the US trucking space, particularly as they relate to the level of trucking capacity.

During the pandemic years, there was a notable uptick in new active carriers in the US. There was a spike in demand for transporting goods (goods spending surged, inventories were rebuilt, and e-commerce volumes accelerated). This led to a sharp uptick in spot rates that created unusually attractive economics for owner-operators. This triggered a major wave of new carrier formation, with many new carriers popping up to take advantage of that heightened demand and elevated spot rates.

Source: Motor Carrier Management System (MCMS) Data Snapshot via FMCSA

However, it is the nature of the capacity added during that period which is noteworthy. If we look at the table below, we can observe that 65-72% of the truck carrier capacity being added in any given year during the pandemic period was from one-truck operators. And there was an 11.5 percentage point increase in the mix of new carriers added coming from one-truck operators in 2022 compared to 2019. It was a freight goldrush of sorts.

Source: Motor Carrier Management System (MCMS) Data Snapshot via FMCSA

But in 2022 the cycle started to turn. Goods demand normalized, retailers moved from restocking to destocking, and spot rates began falling. In fact, DAT dry van national average linehaul rates declined from above $2.60 per mile in January 2022 to below $1.60 in November 2022, a 40% decrease in less than a year. In other words, there was too much trucking capacity for the shrinking level of freight demand, and the spot rates swiftly deflated to reflect this.

Subscribe now

The downturn then became self-reinforcing. Spot rates fell below many small carriers’ economic cost of operating. And this was all while diesel, insurance, maintenance and financing costs remained elevated. Carriers that had bought equipment at peak prices during the pandemic frenzy were squeezed by falling revenue per mile and higher debt service. This led to trucking capacity exiting the market.

But it’s worth pointing out that while carrier capacity was quick to be added, this capacity was slow to exit, as many operators kept operating to generate cash, even at poor margins. Our belief is that this dynamic helped keep rates depressed for longer than in a normal cycle, and it was a grueling three year trucking bear market.

If we look at the Cass Freight Index that tracks the movement of goods by any mode of transportation, the index has suffered 40 consecutive months of YoY declines.

Source: Bristlemoon Capital; Cass Information Systems

This is notable, as we’ve been seeing enormous increases in trucking spot rates even with freight demand remaining weak. And this is due to some recent regulatory changes in the market that have led to further carrier exits.

Read more

If you haven’t subscribed, you can join 6,125 others who enjoy our deep dives and investment insights here:

Subscribe now

Australian wholesale investors looking to invest in the Bristlemoon Global Fund can do so via this link.

Table of Contents

1. Introduction to the business and recent financial performance

2. The power conversion challenge for AI data centers

   1. The legacy architecture and its limits

   2. Two converging architecture standards

   3. The 800V power delivery chain: from grid to transistor

3. The AI data center opportunity for Infineon

   1. Stage by stage

   2. Sizing the AI data center opportunity

   3. Competition and market share

4. Automotive: the bull case that never was

   1. The bull case that never was

   2. Should we be excited about an automotive recovery?

5. Green Industrial Power: the forgotten segment

   1. The solid state transformer opportunity

6. Is there an emerging MCU shortage?

Introduction

Infineon Technologies is not a stock that generates much excitement at dinner parties – nor at AI investor parties, it would seem – until recently. The Munich-headquartered diversified power semiconductor company lacks the headline profile of Nvidia, the Uber driver name recognition of Micron and SK Hynix, or the obscene torque of the optical bottleneck stocks. For most of the post-GFC cycle, Infineon occupied a corner of the market that can be loosely described as “European industrial cyclical with some EV exposure”, and the company’s stock price largely reflected this. Indeed, the last time we seriously looked at the company, it was coming off the back of the post-COVID automotive/industrial shortage and preparing for the global high voltage EV boom that never materialized. The stock spent most of the past four years treading water, before doubling in the space of ~8 weeks.

What has changed over the past twelve months is that the center of gravity for semiconductor demand fractured away from compute (GPUs and ASICs – too boring) and shifted across the supply chain into higher-beta areas such as memory, optical networking and packaging. The latest area to capture the attention (and imagination) of the market is power conversion and management within AI data centers. The AI DC buildout has surfaced a constraint that was obvious in retrospect: raw compute power is now scaling faster than the electrical infrastructure that delivers energy to it. Here, we refer specifically to the power infrastructure from the boundary of the DC facility to the point-of-load (POL) on the GPU itself.

Solving that constraint requires a complete architectural overhaul of data center power delivery, from the traditional 480VAC → 12VDC to 800V high voltage DC (HVDC) led by Nvidia’s 800VDC reference design and the +/- 400VDC Mt Diablo architecture developed by the hyperscalers. Regardless of which design becomes standard, the overhaul depends heavily on the wide-bandgap semiconductors, power MOSFETs and voltage regulator modules that form the core of Infineon’s product portfolio.

While the 800VDC architecture was unveiled last year, it was only in the last couple of months where it felt like everyone was talking about data center power and power semis as the next optical or memory play. The result: IFX and its power semis peers have rerated vertically since the end of March as this narrative gathered steam. Unlike some of the more speculative AI narrative pumps, we’d say this rerating is not without merit. Infineon reported Q2 FY2026 results that beat expectations, raised its full-year sales and margin guidance, confirmed its AI revenue targets for FY2026 and FY2027, and announced the early opening of its Dresden Smart Power Fab Module 4 to serve overwhelming AI power demand.

Furthermore, the 800VDC transition strikes us as a cycle that has legs. Unlike AI servers or memory where a double-digit sales growth trajectory next year could spell disaster for the relevant stocks, the 800VDC architecture is a quasi zero-to-one transition. That is to say, even if Nvidia’s Rubin and Rubin Ultra server generations sell like cold scones, the transition to 800VDC must happen as current data center power architecture simply cannot handle the power density of future AI racks.

In this piece we examine Infineon’s four main investment angles: overwhelmingly dominated by AI data centers today, but encompassing also a potential automotive recovery, industrial power and electrification, and an industry-leading microcontroller business. Ultimately, while we like Infineon as a business, the violence with which this sleepy power/analog stock has rerated to all-time highs (on both price and multiple) has likely pulled forward substantial future returns and leaves us lukewarm on IFX as an investment.

Subscribe now

Business overview

Infineon was spun out of Siemens in 1999 and has grown from a diversified industrial semiconductor company into a more focused power and embedded systems specialist through a series of transformative acquisitions: International Rectifier in 2015 ($3 billion), Cypress Semiconductor in 2020 ($9.4 billion), and most recently the Marvell automotive Ethernet business for $2.5 billion in FY2025.

The company’s competitive position rests on three pillars. First, it is the global leader in power semiconductors across silicon MOSFETs, IGBTs, and silicon carbide (SiC) MOSFETs – devices that control and convert electricity in everything from server power supplies to electric vehicle traction inverters. Second, it holds the #1 position in microcontrollers (MCUs) globally, with approximately 23.2% market share by revenue through its AURIX product family. Third, it occupies leading positions in several niche but strategically important markets including automotive radar sensors, smart card security controllers, and industrial motor drive power modules.

Source: Infineon

Through FQ2 FY2026 (March 2026 quarter), Infineon reported four segments:

• Automotive (ATV) at 50% of FY25 of revenue covers power semiconductors and MCUs for passenger and commercial vehicles;

• Power & Sensor Systems (PSS) at 29% spans power semiconductors for servers, consumer electronics, and mobile devices, plus automotive radar sensors;

• Green Industrial Power (GIP) at 11% covers power components for industrial motor drives, HVAC systems, solar inverters, and grid infrastructure; and

• Connected Secure Systems (CSS) at 10% covers smart cards, security controllers, and IoT connectivity.

Effective July 1, 2026, Infineon will reorganize into three divisions: Automotive (~50% of revenue), Power Systems (~30%), and Edge Systems (~20%). The restructuring consolidates AI data center and industrial power exposure into a single Power Systems reporting unit rather than split across two segments, making the AI opportunity more visible to investors.

Source: Infineon

Recent financial performance

Q2 FY2026 results came in ahead of expectations on the critical metrics, indicating the protracted cyclical trough is firmly in the rearview mirror. Indeed, management’s confident tone on the Q2 call marks a stark contrast to the cautious language that was still being used as recently as January 2026.

“A broader upswing across many end markets is clearly on the horizon. We are seeing rising demand in several key markets. While geopolitical conflicts continue to weigh on people and markets, our business indicators such as order intake, delivery times, cancellation rates, and inventory levels are showing a significantly improved picture. […] The market development in industrial applications is being supported by rising demand for energy infrastructure. In the automotive sector, order intake is rising as customers begin to replenish their low inventory levels. However, electromobility remains in difficult waters, while we are seeing a positive global trend in software-defined vehicles. Overall, demand in our end markets is improving significantly. We are preparing for a broad-based upswing.” – CEO, Q2 FY2026 analyst call

Compared to a quarter ago:

“Near-term indicators such as order intake or lead times are now starting to align in a positive way, and visibility is improving. That said, we feel our view of a gradual and uneven recovery remains appropriate for the time being. […] Automotive and industrial markets are beyond the cyclical trough, but demand has yet to really inflect. Consumer, compute ex-AI, and communication markets are in the early innings of the upcycle. AI remains the clear standout, continuing its extraordinary growth propelled by sustained massive investment in data centers and related infrastructure.” – CEO, Q1 FY2026

Revenue was €3.8 billion, up 4% QoQ and 6% YoY, with a segment result margin of 17.1%. The positive surprise was driven by stronger-than-expected AI power demand and a meaningful recovery in the GIP segment (+15% QoQ). Infineon raised its full-year FY2026 revenue guidance to above €16 billion with a segment result margin of approximately 20%, a step up from the high-teens margin guided earlier in the fiscal year and an upgrade to sellside consensus. The company simultaneously confirmed its dedicated AI power revenue targets of €1.5 billion in FY2026 and €2.5 billion in FY2027. The AI business is, per management’s own characterization, in allocation through FY2026 as demand exceeds available supply.

“Currently, our AI business is in allocation and we are converting capacities from other areas as well as ramping new ones as quickly as possible. Given the unabated momentum, we are reconfirming our guidance of dedicated AI power revenues of EUR1.5 billion for ‘26 as well as our indication of EUR2.5 billion for ‘27, irrespective of a weaker US dollar.” – CEO, Q2 FY2026 analyst call

“As we said carefully in our prepared script, the longer the allocation lasts on AI and adjacent markets, the more positive dynamic for pricing could be also visible still in the remaining quarters.” – CFO, Q2 FY2026 analyst call

The most significant near-term operational event is therefore the opening of Dresden Smart Power Fab Module 4 on July 2, 2026, earlier than the originally planned autumn timeline. Dresden Module 4 represents Infineon’s largest single investment at approximately €5 billion and is the primary production site for AI silicon MOSFETs. Its ramp through H2 FY2026 and into FY2027 is the central production lever underpinning management’s €2.5 billion FY2027 AI revenue target. The company is also converting underutilized silicon IGBT capacity (for EV and PV inverters) into AI silicon MOSFETs and further expanding power-related analog and sensor capacity by pulling forward an additional €500 million of capex.

“And for the time being, as I mentioned also in my summary, we are less hopeful for the XEV market. In fact, we believe that the development will be showing less growth than many market participants anticipate. And in the IGBT domain, for XEV as much as for certain applications in photovoltaics, we believe that it’s better to use this capacity, as I outlined in my intro, to convert it to leading-edge silicon MOSFETs and use it for the fast-growing, margin-accretive AI business.” – CEO, FY2025 analyst call

“These investments, including the conversion of existing IGBT capacities, are capital efficient and will underpin our market leadership. A large part of them will be used for a faster-than-originally-planned ramp-up of our new power and analog/mixed-signal fab module in Dresden. Fortunately, we are able to accelerate the opening of the facility to this summer with the additional capacity we project to achieve AI-related revenues of around EUR2.5 billion in our 2027 fiscal year.” – CEO, Q1 FY2026 analyst call

“The customers are very demanding. Technically, it’s very demanding. But we will be able also to supply in-house manufacturing Dresden Module 4, also what comes more and more that particularly our US customers ask for non-China, non-Taiwan supply chains which we can offer. So, plenty of aspects which speak in favor of Infineon.” – CEO, Q2 FY2026 analyst call

The power conversion challenge for AI data centers

Understanding Infineon’s AI opportunity requires some basic knowledge of the power delivery chain inside a modern data center – a topic that has received surprisingly little investor attention until recently compared to power generation and grid transmission. This is definitely rudimentary to anyone who has studied electrical engineering 101 (we didn’t) but we will keep it simple to not lose anyone. For more details on the power architecture of AI data centers, numerous excellent Substack articles have been published on the topic in the past two months, coinciding with (or perhaps triggering) the sharp rally in power semis-related stock prices. See here, here and here for example.

Subscribe now

The legacy architecture and its limits

Traditional data centers receive power as high-voltage alternating current (HVAC, typically 13.8kV at the campus boundary), step it down through successive transformers to 480VAC, protect it through uninterruptible power supply (UPS) systems, distribute it through power distribution units (PDUs), rectify it to 12V direct current through server power supply units (PSUs), and finally step it down again through on-board voltage regulator modules (VRMs) to the 0.7-1.2V required by the processor die. This multi-stage cascade was engineered when a fully-loaded server rack drew 5-10kW. It delivers around 70-85% of grid electricity to the compute die, with the balance dissipated as heat in power conversion components and current paths. For a 1GW AI data center, even 85% power conversion efficiency means 150MW of power dissipated as heat during voltage conversion.

Source: Nvidia technical blog

While burning off 100,000+ homes worth of electricity every hour as heat would be immensely regrettable, the more fundamental physical constraint forcing the transition to 800VDC is current. A fundamental identity of electricity is Power equals Voltage times Current (P = V * I). At a fixed voltage (e.g. 48V), higher power requires proportionally higher current. This wasn’t an issue when server racks would draw 6-12kW of power, requiring up to 250A of current which could be handled by a copper wire the width of your pinky. At 140kW per rack, the GB300 NVL72 starts to really test the limits of the 48V rack architecture. By the time we get to the Rubin Ultra generation, a rack is expected to draw 600kW+ of power, which exceeds 12,500A of current at 48V. Based on Ohm’s Law P_loss = I² * R, this 50x increase in current leads to a 2,500x increase in conduction loss assuming resistance R is constant. We would need a 2 meter long solid copper busbar with the cross-sectional area of an iPhone to deliver the required power to the GPUs.

Source: Vik’s Newsletter

This is obviously not physically or economically viable at scale, so what is the solution? For a given power envelope, power loss scales quadratically with current and linearly with resistance. Halving the current by doubling the voltage reduces conduction loss by a factor of 4, while halving the resistance by doubling the cross-sectional area of the copper busbar only reduces conduction loss by half. Hence the industry’s solution has been to shift to 800VDC rack power architecture: this reduces rack current by 16.7x compared to 48V, which reduces conduction loss by 278x assuming resistance remains constant. Of course, resistance will increase as the size of the busbar shrinks from the area of an iPhone, but the overall net reduction in power loss will still be many multiples of the prevailing 48V architecture.

NVIDIA’s GPU thermal design power (TDP) trajectory is familiar to most technology investors, but its implications for power delivery infrastructure are less widely appreciated. The A100 at 400W was manageable within the 480V → 48V → 12V architecture. The H100 at 700W was manageable with upgraded cabling. The Blackwell GB200 NVL72 rack draws approximately 120kW across 72 GPUs and 36 CPUs. The forthcoming Vera Rubin and Rubin Ultra rack designs will ramp power consumption from 200kW to 600kW, and by the time the Feynman generation rolls around, single racks will draw over 1MW of power.

Two converging architecture standards

The industry has coalesced around higher DC bus voltages as the solution to the current density problem. Two standards have emerged in parallel. Mt. Diablo, backed initially by Microsoft and Meta and subsequently joined by Google, proposes ±400VDC (effectively 800V differential) delivered from standalone power cabinets to IT racks, published through the Open Compute Project. Its commercial deployments are expected from 2026. The standard retains the existing 480VAC backbone at the facility level and inserts a DC conversion shelf at the rack boundary. This is a more conservative approach that lowers adoption friction for hyperscalers with large existing AC infrastructure estates (which can be retrofitted), and leverages the established EV battery pack supply chain that has validated ~400V components at volume for over a decade.

Nvidia’s 800V HVDC architecture, unveiled at GTC 2025, is more radical: it replaces the 480VAC backbone entirely, delivering 800V monopolar DC from the utility interface throughout the data center floor and directly into racks. The architecture explicitly targets 1MW+ rack power densities consistent with Rubin Ultra and beyond. It is the more aggressive standard as it requires native 800VDC data center designs (existing LVAC data centers can’t be retrofitted), but also the one best suited to efficient high density power delivery as the industry converges on megawatt-scale racks.

The practical differences between the two standards matter less for the semiconductor content opportunity than their common element: both converge on roughly 800V delivered to the rack, and both require wide-bandgap power semiconductors at every conversion stage. However, the Mt. Diablo standard preserves more of the 480VAC architecture and thus incumbent power supply vendors, while Nvidia’s 800V reference design is more favorable to high-voltage, wide bandgap power semis suppliers.

Subscribe now

The 800V power delivery chain: from grid to transistor

In the 800V HVDC architecture, power delivery from the utility interconnect to the compute die collapses from seven stages to three or four. A solid-state transformer (SST) converts 13.8kV AC directly to 800VDC at up to 99% efficiency, replacing the traditional multi-stage transformer-plus-UPS combination. An 800VDC bus runs throughout the data center floor: at 140kW per GB300 NVL72 rack, this requires only 175 amps which is manageable with a standard industrial bus bar and reportedly reduces copper requirements by approximately 45% versus an equivalent 54V architecture. A board-level HV DC-DC converter steps 800V to 12V (likely using SiC MOSFETs on the primary side and GaN on the secondary), achieving 97-98.5% efficiency at switching frequencies measured in hundreds of kilohertz. Finally, point-of-load VRMs adjacent to the GPU package step 6V or 12V down to the 0.7V core voltage required by the GPU.

Source: Nvidia technical blog

The cumulative efficiency of this architecture can exceed 90%, against 70-85% for the legacy AC chain. For a 1GW data center, the improvement can avoid multiple tens of MWs in conversion losses compared to best-in-class 480VAC hyperscale architectures, which can then be used to power more compute (and as we know, compute = $$$).

For semiconductor companies, the shift to 800VDC changes the technical requirements at every stage of the power chain. Low-voltage silicon MOSFETs in simple buck topologies are supplanted by wide-bandgap devices – SiC for high-voltage switching, GaN for intermediate bus and POL – alongside isolated gate drivers, reinforced isolation components, and power management controllers. The bill of materials is richer and more technically demanding at every node, which is why companies with broad power portfolios like Infineon stand to capture disproportionate content per dollar of AI infrastructure spend.

The AI data center opportunity for Infineon

Infineon participates at virtually every stage of the AI data center power chain, which in our opinion makes the opportunity more interesting, albeit at the cost of maximum torque to any particular power stage. The company is present in the LV AC-DC (480VAC → 54/48VDC) and intermediate bus conversion (54/48V → 12/6V) stages through silicon MOSFETs and GaN devices used in PSU reference designs, and more critically, in the second stage VRM market through its OptiMOS TDM quad-phase modules that enable vertical power delivery. Management has most explicitly tied the AI revenue trajectory to the VRM and GaN stages, which is where Infineon’s technical differentiation is most defensible. Additionally, given Infineon’s strong position in SiC established over the years in the EV industry, we believe the company is also well positioned for the Solid State Transformer opportunity as that market develops towards the end of the decade.

Source: Infineon

Stage by stage

Source: Infineon

At the PSU stage (rack-level AC-to-DC conversion), Infineon supplies CoolMOS silicon power MOSFETs and GaN devices to PSU manufacturers including Delta, Murata, Bel, and Acbel. This is an indirect market where Infineon is designed in by the PSU maker rather than by the accelerator platform owner, and the silicon content at this stage is subject to competitive substitution at design-in cycles of 12-18 months. Texas Instruments, STMicroelectronics, onsemi, and Navitas amongst others compete with Infineon for PSU content. Content per unit is meaningful but not highly differentiated, so the value tends to be lower than for stages closer to the GPU itself.

The intermediate bus conversion (IBC) stage is likely to see the most change during the transition to 800VDC architecture. The IBC stage itself basically exists as a compromise between distribution efficiency and POL feasibility as 54/48V is better for moving power across the server board (lower current), while 12/6V makes the final POL conversion down to ~0.7V easier. In the transitional 800V architecture, the 480VAC → 54/48VDC stage is replaced by an 800V → 54/48V DC-DC stage, with the LV IBC stage retained to preserve as much of the existing rack power infrastructure as possible to ease the transition. But the ultimate end state is to bring 800V straight into the server blade and perform a 67:1 conversion down to 12V near the GPU itself, thus minimizing the distance of the high-current low-voltage path. This would do away with the conventional LV IBC but require a much more demanding 800V → 12V high voltage isolated DC-DC converter.

We believe Infineon is well positioned for this shift because the HV DC-DC conversion stage moves the value pool towards the power semis components and module design where Infineon is strongest: high-voltage switching using a combination of Si, SiC and GaN, isolation, gate driving, sensing, protection, and system-level power control. The company forecasts the IBC SAM to grow at over 50% CAGR from 2025 to 2030, driven mainly by HV IBC, which supports our view that this stage will pull in a significant amount of high value power semis content from earlier stages.

In March 2026, the company introduced two HV IBC reference designs for 800V → 50V and 800V → 12V architectures using its CoolGaN switches. We understand that Infineon is currently the only company globally with scaled 300mm GaN wafer production versus the industry standard of 150mm or 200mm, which should enable materially lower die costs and tighter process control. This is not to say that Infineon will dominate this stage, as competitors like Texas Instruments and STMicro have both published their own 800V → 12V converter reference designs. But these kinds of major system-level transitions tend to reward vendors who can deliver a complete, integrated power semis portfolio.

Finally, the most strategically important stage is the second-stage voltage regulator module. These are the power modules that sit adjacent to or directly underneath the GPU package on the server board, converting the intermediate bus voltage to the GPU core voltage (typically 0.7-1V). This stage is technically demanding because it feeds voltage directly into the GPU power rails, must handle extremely high current (a 2,300W GPU package with 0.7V core voltage requires 3,300A of current), and is responsible for much of the power regulation required for the GPU’s very dynamic loads. Furthermore, as should be evident from the power delivery discussion above, this massive aggregate current at the chip level i) cannot be lowered unlike during earlier power stages, and ii) is where the largest efficiency losses occur. Therefore, VRM and power module vendors are increasingly turning to vertical power delivery (VPD) as a solution to the challenges presented by multi-kilowatt-rated AI processors.

Source: Infineon

Infineon’s OptiMOS TDM2454xx quad-phase modules, announced in March 2025, deliver 2A per mm² of current density and the company is working with Delta Electronics, the primary Nvidia power supply vendor, to enable VPD for AI processors. Management has committed to improving this to 3A/mm² by 2027, with substrate-integrated voltage regulation (SiVR) as the five-year horizon technology. By shortening the current path from centimeters as is the case with lateral power delivery where VRMs surround the GPU package, to millimeters through the compute board itself with VPD, conduction losses can be reduced by an order of magnitude (resistance scales linearly with the length of the conductor). Infineon also claims that its VPD modules enable higher power density, better thermals due to less power loss, and space savings on the compute board allowing for denser system designs.

Source: Infineon

The module business carries meaningfully higher gross margins than discrete MOSFETs because it integrates the gate driver, controller interface, and packaging in a form factor that reduces board space and design complexity for the customer, often supporting an ASP several times higher than the equivalent discrete component. Management has confirmed that AI power revenue is margin-accretive relative to the group’s high-teens segment result margin average. And on VPD specifically, management has also stated that these modules land at the high end of the power semis content range per installed kilowatt ($100-$250, more on this below).

“And for us, of course, the second stage part is giving us particular growth momentum. So we expect the share of the second stage in the overall numbers to increase over the quarters to come. As Andreas pointed out that vertical power deliveries, the modules have a much higher value. And again we are here teaming up with all GPU makers, or ASICS makers, whatever you want to call them (including in China).” – CEO, Q1 FY2026 analyst call

Sizing the AI data center opportunity

For paid subscribers, we discuss the size of Infineon’s AI power opportunity and the competitive landscape. Additionally, we dig into the positioning and outlook for the Automotive segment, the emerging AI-related opportunity within the Green Industrial Power segment, and consider whether an MCU shortage may emerge as the next opportunity for Infineon.

Read more

Feel free to subscribe, and join 6,074 others who get our investment memos directly to their inboxes.

Subscribe now

In my recent travels to the U.S., I had a lot of conversations with other investors who are grappling how to adapt to this market environment. Market environment is a bit of a nebulous descriptor, so it’s worth elaborating.

With multi-strategy funds (a.k.a. multi-managers or pods) capturing a disproportionate share of hedge fund AUM growth over the past decade, they have become a more important part of how liquid securities trade. Given how much these pods lever up their equity capital and the high level of portfolio turnover, increasingly the pods are the marginal buyer (or seller) of liquid securities. Layer on top of that the rise of passive and retail participating more actively in pockets of the equity market, and it’s not hard to see how these developments have changed the manner in which stocks trade. This is the change in market environment I am referring to.

As we can see below, multi-managers account for around 10% of the total hedge fund industry AUM. Incredibly, the multi-strategy funds went from managing $91 billion of assets in 2010 to roughly $428 billion in 2025, a 4.7x increase in assets over the last 15 years.

However, data from the Goldman Sachs prime desk showed that multi-strategy firms accounted for around one-third of the gross market value managed by hedge funds in equities[1] (i.e., due to the leverage employed).

J.P. Morgan’s data for the month of November 2025 revealed that multi-strategy funds had leverage at 444.3%. In other words, for every $100 of capital from investors, the multi-strategy fund had almost $450 in long and short positions[2]. That’s a lot of leverage!

It’s thus no surprise that the pods operate within very tight risk parameters. They run market neutral (although some pods have flexibility to tilt their net exposure beyond strictly neutral), which basically means they need to net out their long exposure with an equivalent short exposure. With that much leverage and highly restrictive risk parameters, there is minimal scope for drawdowns.

There is no shortage of stories of pods being wound down if a manager blows through their risk limits. So, what this means is that the pods have minimal duration and have a strong tendency to be long the stocks that are “working” at any given time. The corollary of this is that the pods have an inherent bias towards being long momentum (bottlenecks, accelerating growth profiles, and the like).

If we think about the current market, the area of the tech sphere that is working (and in a gargantuan way) is semiconductors. The SOX (The Philadelphia Semiconductor Index) is up around 67% year-to-date, and some of the earnings beats during Q1 were simply eye-watering. If you’re a tech pod, you’re getting up and dancing while the music plays. In other words, you’re probably buying semis due to strongly positively inflecting growth profiles, good visibility into demand persisting (and there being little to derail near-term earnings beats), as well as just strong momentum in these stocks.

But, if you’re running market neutral like the pods do, you can’t simply buy semiconductor stocks and call it a day. You have to fund that exposure, and have an offsetting short book. Software and consumer internet names have been prime funding source candidates. We’ve previously noted that there’s been a lack of narrative nuance, with the tech landscape bifurcating into perceived AI winners and perceived AI losers. The way in which the multi-managers execute their strategies has likely exacerbated this dynamic.

Just take a look at the spread between the IGV and SOX. While the SOX is up c.67% year-to-date, the IGV (a software sector ETF) is down around 12%. That’s a 79 percentage point outperformance of semiconductors versus this software basket in less than six months.

This has split the fortunes of fund managers. If you’d owned a lot (or even just a little) of semi and AI-beneficiary companies, your Fund is probably shooting the lights out. If you had been caught on the wrong side of this trade, it’s likely a completely different story, with a different set of conversations with clients. I have never witnessed a period in markets where capital has rotated from one sector into another at such a scale and in such a short period of time. It feels truly unprecedented.

We admittedly have had too much of what was in the IGV, and too little of what was in the SOX. And that’s hurt our year-to-date performance. I’m not writing this to lament this soft run of performance, and I’m also not blaming the market (as they say, “no crying in the casino”). Rather, I thought this was a good opportunity to relay the thoughts and discussions that flowed from conversations I had with investment managers who I met during my travels to the U.S., many of which were grappling with similar challenges in navigating this market.

Subscribe now

How are managers adapting?

One of the biggest challenges is that in order to keep up in this hot market, it has required most fundamental managers to pivot dramatically from their process which has worked historically (e.g., deep fundamental work, valuation sensitivity, buying and holding quality businesses that compound over the long-term, etc). The current market is faster moving, more narrative driven, and if you’re constrained by fundamentals, you probably won’t get there on many of the higher torque AI-related names that are further out on the risk curve.

This is certainly not to say that all AI-related stocks are expensive. But many of the stocks that are going vertical require a continuation of heady demand years into the future, such that these stocks can grow into their lofty valuations. Investors dabbling in the more speculative part of the semis complex are increasingly underwriting demand that extends into 2030 and beyond. That demand might transpire, but these stocks become increasingly risky to the extent that investors are treating that growth as a sure thing.

There was broad acknowledgement from managers I spoke to that these stocks are expensive. But in terms of how those stocks are being traded and the market structure changes highlighted above, the absolute valuations of these stocks have been relegated to a secondary concern. Rather, the focus is squarely on whether these companies are accelerating their topline and producing earnings beats. So long as that is happening, seemingly expensive stocks can march higher. It’s the rate of change of the fundamentals in the short term that has mattered most for these stocks, and more and more non-pod investors are recognizing this and changing the manner in which they invest.

This approach of playing near-term accelerations/decelerations is anathema to fundamental investors who typically underwrite businesses based on the earnings and cash flows they will produce over long stretches of time. This approach acts as an anchor that makes it hard for fundamental managers to simply pivot into the stocks that are working.

That discipline helps avoid overpaying for speculative stocks, but it also makes it difficult to perform well in markets where investors are willing to pay up for outer year cash flows that rely on a chain of assumptions around ever increasing demand, restrained capacity additions, tech roadmaps that do not upend favorable supply/demand dynamics, and accommodating capital markets. In some of these AI stocks, the path to getting paid appears exceptionally narrow and deceptively multiplicative.

It does feel like there’s a huge spread in the willingness of the market to give credit to outer year cash flows for AI winners versus AI losers, which speaks to the prevailing confidence in the AI trade persisting.

The difficulty in adapting to this market is exacerbated by many managers feeling the need to adhere to what they’ve marketed to their LPs (i.e., sticking to what’s on the tin of the fund strategy). This dynamic is more acute for managers with more institutional mandates that will give far greater scrutiny to any process changes.

I heard numerous stories of fund managers who can invest on their personal account chalking up massive year-to-date returns, but being down in their fund. This likely reflects the tendency of a manager’s approach to investing to become sclerotic if their LPs are likely to penalize them for process changes (even if those process changes are necessary and a sensible course of action). I think the impetus to change is always greatest during periods of drawdown/underperformance but it was a topic in virtually all the conversations I was having.

Semis is the FOMO trade

Semis is now firmly the FOMO trade. Regardless of fundamentals, every manager is feeling the pull of the seemingly infinite money glitch of owning semis. Lots of generalists I’ve spoken to during my trip have sharpened their pencils on the space, with some shifting meaningful portions of their book into select hardware names.

So, while no doubt pods are in semis and using internet/software as funding sources, the rotations of capital out of software and into semis are to an extent also being directed by long-only funds that are obviously very conscious of lagging their benchmarks.

Interestingly, the generalists venturing into semis are unconstrained by the knowledge/experience of how low quality some of these names were considered previously, or how cyclical the fundamentals of these stocks have been historically. Rather, it’s easier to come to the space with a blank slate and, for example, form a view around the sustainability of the supply/demand imbalance in memory persisting.

We have not done a great job at playing the AI boom. We own NVIDIA and ASML as they are quality businesses and we can get comfortable with the valuations of these stocks. But they have much lower torque to the AI trade compared to say memory prices going parabolic, and they won’t have the same insane earnings beats. We have been waiting for a pullback which very much has not occurred in some of the semi stocks we’d be comfortable owning.

Semis tourists make the trade more fragile

The problem with so many generalists chasing semis is that it makes the AI trade more fragile. Having spoken to a number of investors in recent weeks who have recently initiated positions in semis, they have openly admitted that they’re not across the tech details and acknowledged that they could be blindsided by tech changes that are negative to fundamentals.

This is perfectly fine when there’s upside momentum in the stocks. But the risk is that an increasing portion of the shareholder bases of these hot semis stocks lack true conviction in the fundamentals, which increases the risk of an unwind to the extent that there are any unforeseen developments that cast doubt on the fundamentals (or perhaps even just if these stocks stop working, and momentum reverses).

Lots of managers are entering these stocks simply because they’re the stocks that are working and the fear of underperforming/career risk is forcing them to dance while the music plays. But because of this, they’re also much more likely to pull the plug once these stocks stop working. So while the fundamentals for many semi stocks are currently amazing (monstrous earnings beats and accelerating growth profiles), some of the valuations require conviction around very strong fundamentals extending into 2030+. We have been looking at names in the space and have struggled to build that conviction.

We manage money on behalf of high net worth individuals, family offices, a charity, and we have our own money, and our families’ money in the Fund. For this reason we do not feel compelled to chase the popular stocks in a hot market, and unnecessarily risking the capital we’ve been entrusted with managing. If we do not assess the risk/reward to be attractive, and we lack the conviction in that stock, we will pass, even if others are making a lot of money. These are just not risks we think are sensible to take.

It will be interesting to see how it all plays out, and as always, we welcome constructive pushback on any of the above!

[1] https://www.reuters.com/business/finance/hedge-funds-double-down-using-near-record-leverage-quest-boost-returns-2025-12-03/

[2] https://www.reuters.com/business/finance/hedge-funds-double-down-using-near-record-leverage-quest-boost-returns-2025-12-03/

Disclaimer / Disclosures

The information contained in this article is not investment advice and is intended only for wholesale investors. All posts by Bristlemoon Capital are for informational purposes only. This article has been prepared without taking into account your particular circumstances, nor your investment objectives and needs. This article does not constitute personal investment advice and you should not rely on it as such. This document does not contain all of the information that may be required to evaluate an investment in any of the securities featured in the document. We recommend that you obtain independent financial advice before you make investment decisions.

Forward-looking statements are based on current information available to the author, expectations, estimates, projections and assumptions as to future matters. Forward-looking statements are subject to risks, uncertainties and other known and unknown factors and variables, which may affect the accuracy of any forward-looking statement. No guarantee is made in relation to future performance, results or other events.

We make no representation and give no warranties regarding the accuracy, reliability, completeness or suitability of the information contained in this document. To the maximum extent permitted by law, we do not have any liability for any loss or damage suffered or incurred by any person in connection with this document.

Bristlemoon Capital Pty Ltd (ABN: 22 668 652 926) is an Australian Financial Services Licensee (AFSL Number: 552045).

George Hadjia and Daniel Wu are associated with Bristlemoon Capital Pty Ltd. Bristlemoon Capital may invest in securities featured in this newsletter from time to time.

If you haven’t subscribed, you can join 5,954 others who enjoy our deep dives and investment insights here:

Subscribe now

Australian wholesale investors looking to invest in the Bristlemoon Global Fund can do so via this link.

Business Overview

Today we’re taking a quick look at Montana Aerospace (AERO SW), the Swiss aerospace company. AERO is one of the fastest growing aerospace suppliers, having grown its Aerostructures revenue at an 11.5% annual rate between FY22 and FY25. This has primarily been driven by market share gains and ramping OEM build rates. The aim of this writeup is to jot down our thoughts on the business and whether it’s interesting enough to be considered for the Bristlemoon portfolio.

AERO is a supplier of complex, mission-critical components for aviation and space applications. The company’s product portfolio includes structural components for fuselage (think of this as the structural long beams that comprise the skeleton of an aircraft’s fuselage), wings, and landing gear, critical engine parts exposed to extreme thermal and mechanical stresses, as well as large-format extrusion products.

Source: Company filings

The company’s aerospace portfolio is split as follows:

1. Universal Alloy Corporation (UAC) – manufacturer of aerospace structural components and assemblies.

2. ASCO Industries – designer and manufacturer of high-lift structures, complex mechanical assemblies, and critical functional components for the aerospace industry.

Montana’s products are on all major commercial aircraft platforms (historically the company has had a roughly 50:50 split between Boeing and Airbus). Back in 2023 the company had around €350k of average shipset content across its platforms, with this expected to ramp up to €450-500k in the years following, driven by new content share wins and further OEM outsourcing[1].

The way AERO works as a business is that they win content on various aircraft platforms. Think of this as Boeing contracting with Montana as a supplier of say, a landing gear product. In reality, there are many different parts that Montana will supply into an aircraft platform (note that AERO offers over 100k SKUs in aerostructures).

This all ties together into a contracted shipset value – that is, the total dollar value of all the components that Montana is contracted to provide for one single aircraft of a specific model. And recall that AERO has significant content on all Boeing and Airbus aircraft types.

Subscribe now

So, in this sense you can think of AERO’s sales as being driven by the shipset value multiplied by the number of aircraft that Boeing and Airbus will produce in the future. And how much certainty we can get over Montana’s future sales depends on how long Montana remains as the supplier to that aircraft platform, and how the economics of supplying those components change over time.

There’s reason to believe that AERO will be able to keep selling its products into these aircraft platforms for many years because most of the company’s contracts are “single source” – that is, AERO is the exclusive supplier of that product, creating strong visibility over future revenues and providing barriers to competition (given the time and cost of qualifying alternate suppliers). However, we found it more difficult to get confidence around the pricing power and economics of AERO supplying into those platforms (as we discuss further below).

Vertical integration

AERO is vertically integrated – it covers the aerospace value chain from processing raw materials to the assembly of entire components. This removes some of the issues that result from multiple third-party suppliers, whereby there might be delays that are capable of disrupting the entire production process. In contrast, a non-integrated value chain means that the final outcome of the component for the OEM is multiplicative – that is, it requires execution from multiple suppliers and if there’s a hiccup with one then it stalls the entire process (basically one of the suppliers being a “zero” in the multiplicative string of events). AERO can mitigate these risks by having greater control over the value chain.

Source: Company filings

Read more

This was the most severe drawdown the Fund has experienced since inception. Importantly, it was not driven by a deterioration of the underlying earnings power of the businesses owned by the Fund, but rather by a sharp and rapid repricing of what investors were willing to pay for those earnings. In this letter we will explain what contributed to the weak performance and how we plan to address this with our compounding/conviction position sizing framework, discuss how we have adjusted the Fund’s positioning in light of developments from the Iran War, and provide an update on the Fund’s AppLovin, FICO, Hemnet, and PAR positions.

We believe the current opportunity set is the most compelling we have seen since the Fund was launched. For those contemplating an investment in the Fund, we are seeking to partner with long-term, aligned investors, and we would welcome the opportunity to discuss our investing approach in more detail. For new investors looking to enter the Fund, more information can be found here or by emailing info@bristlemoon.com.

By accessing the link below, you acknowledge and agree that you are an Australian wholesale client as defined by Section 761G of the Corporations Act 2001.

Bristlemoon Global Fund March 2026 Quarterly Report Vf

792KB ∙ PDF file

Download

Download

U.S. Trip

Our Chief Investment Officer, George Hadjia, will be in Los Angeles on 28-29th April, in Omaha for the Berkshire Hathaway weekend, and in New York on 4-12th May. For investors and allocators located in these cities, please reach out at info@bristlemoon.com if you would like to catch up. We are always looking to connect with likeminded investors to exchange stock ideas, as well as allocators who align with our investing approach.

Subscribe now

Disclaimer / Disclosures

The information contained in this article is not investment advice and is intended only for wholesale investors. All posts by Bristlemoon Capital are for informational purposes only. This article has been prepared without taking into account your particular circumstances, nor your investment objectives and needs. This article does not constitute personal investment advice and you should not rely on it as such. This document does not contain all of the information that may be required to evaluate an investment in any of the securities featured in the document. We recommend that you obtain independent financial advice before you make investment decisions.

Forward-looking statements are based on current information available to the author, expectations, estimates, projections and assumptions as to future matters. Forward-looking statements are subject to risks, uncertainties and other known and unknown factors and variables, which may affect the accuracy of any forward-looking statement. No guarantee is made in relation to future performance, results or other events.

We make no representation and give no warranties regarding the accuracy, reliability, completeness or suitability of the information contained in this document. To the maximum extent permitted by law, we do not have any liability for any loss or damage suffered or incurred by any person in connection with this document.

Bristlemoon Capital Pty Ltd (ABN: 22 668 652 926) is an Australian Financial Services Licensee (AFSL Number: 552045).

George Hadjia and Daniel Wu are associated with Bristlemoon Capital Pty Ltd. Bristlemoon Capital may invest in securities featured in this newsletter from time to time.

If you haven’t subscribed, you can join 5,698 others who enjoy our deep dives and investment insights here:

Subscribe now

Australian wholesale investors looking to invest in the Bristlemoon Global Fund can do so via this link.

Introduction

“The airline industry’s demand for capital ever since that first flight has been insatiable. Investors have poured money into a bottomless pit.” – Warren Buffett, 2008 Annual Letter to Berkshire Hathaway Shareholders

Yes, we are indeed starting this missive with a Buffett quote. But while Buffett in his usual folksy manner has consistently deplored the subpar economics of the airline industry, we would posit that the quote above (and the many other Buffett airline quotes) are reductive and fail to account for the distribution of economic fortunes of the various airline industry participants.

Let’s dig into this idea. As we can see in the chart below, there is a distribution of value creation amongst various industries. But in the smaller sub-charts, we can also observe that within any given industry there is a distribution of value creation. In other words, the average level of value creation (or destruction) for an industry masks the fact that beneath the surface there will be some companies creating value, and others destroying value.

Source: Credit Suisse

(N.B. the blue bars measure the spread between the cash flow return on investment (CFROI) and the cost of capital across 68 global industries, classified under MSCI’s GICS framework, using 5,500 publicly listed companies)

Admittedly, when we first started looking at AerCap (AER) in mid-2024, our thinking was prejudiced by the bad reputation of the airline industry as a cash incinerator. You see, AER is the world’s largest aircraft lessor, and we initially struggled to fathom how this would be a business worth investing in for Bristlemoon. As we did more work on AER, this uneducated view very much changed.

As we can see from the industry map below, the airline industry is obviously a lot broader than just airlines.

Source: Credit Suisse

So how does an aircraft lessor stack up? Based on Michael Mauboussin’s profit pool chart below, we can observe that lessors have generated a healthy amount of economic value (measured as a function of the economic profitability of the industry – that is, the CFROI less the discount rate – and the aggregate invested capital).

Source: International Air Transport Association, “Vision 2050”, February 12, 2011; via Credit Suisse.

(N.B. Returns data is from 2002-2009, and investment data is as of 2009. CRS = computer reservations systems; ANSP = air navigation service provider)

A recap on AerCap

While this is a high level, rather academic approach to industry analysis, it’s worth reiterating what attracted us to invest in AerCap back in 2024. We would point readers to our original write up on AER back in July 2024 which can be found here. We have removed the paywall to that article so all readers can view the writeup in full (notably the majority of points in the article continue to provide tailwinds for AER today).

The stock is up more than 60% since July 2024, having compounded at 33% per annum. However, despite this stock price increase, there are still reasons why we continue to own the stock in the Bristlemoon Global Fund.

Source: Bristlemoon Capital; Bloomberg

As a recap, AER is a multiple arbitrage story. The company is selling planes from their fleet at 2x book value, and buying back stock at around 1x book. This is, of course, financial engineering. But it is also wonderful value generation mechanism given the very high degree of confidence we were able to build around the aircraft undersupply situation persisting.

Source: Bristlemoon Capital; Company filings

AER CEO Aengus Kelly has done a spectacular job of creating value for AER shareholders (it’s no wonder – his interests are quite aligned given that he owns more than $770 million of AER stock). For example, AER repurchased 22.1 million shares in 2025, shrinking its diluted shares outstanding by around 11% year-over-year. In fact, the number of shares outstanding for AER is now 31% lower than it was just three years ago.

In this sense, owners of AER over that period now have a 45% greater claim on the business due to these buybacks. This is remarkable, and there are very few examples in public markets where companies retire almost one-third of their share count in such a condensed period.

Subscribe now

These buybacks have led to the company’s book value per share (BVPS) compounding at a healthy clip (and note that this BVPS is understated for a variety of reasons we state in our original writeup).

Source: Bristlemoon Capital; Company filings

Sandbagged Guidance

Earlier this month the company reported its Q4 2025 earnings result and gave guidance for FY26 EPS of $12 to $13. With the stock trading around the $150 mark, that guidance sounds wholly underwhelming when we consider that AER has historically traded at an 8x P/E (implying a share price range of $144 to $152 per share). This misses a few things, and we will mount a case why we believe that the guidance given is highly conservative.

Firstly, we would note that the $12 to $13 EPS range does not include any further gains on sale from selling aircraft. Thought of another way, the guidance assumes that AER simply stops selling planes for high prices and realizing material gains on sale that benefit its EPS. We would round the probability of this occurring down to zero, particularly given the company is already assuming $2.0 to $3.0 billion in aircraft sales in FY26 (the company disclosed this when it disaggregated the various P&L projections that support the company’s FY26 EPS guidance).

It is worth noting that in FY25 AerCap sold $3.9 billion worth of aircraft and generated $3.95 per share from gains on sale of these owned aircraft (i.e., almost 26% of FY25 adjusted EPS of $15.37 came from gains on sale). While these gains from asset sales are difficult to underwrite and will be lumpy by quarter, perhaps we can bake in an additional $3-4 of EPS from gains on sale in FY26. This bumps up the FY26 EPS range to $15-$17.

It is also worth noting that even outside of any gains on sale, AER has a tendency to be conservative when guiding the base business EPS. For example, in Q4 2024, the company guided FY25 EPS to a range of $8.50 to $9.50. As we progressed through 2025, the company continued to upgrade this EPS guidance. Excluding gains on sale, AER achieved $11.42 of base business EPS in FY25, a whole 20% to 34% higher than the initial FY25 guidance range given in Q4 2024.

Our expectation is for this FY26 EPS guidance to be upgraded as we move through the year, and this is based not so much on precedent but on some of the conservative assumptions underpinning the guide relative to what we think the future holds for AER.

Below we can see the various assumptions the company is using to arrive at its $12 to $13 of FY26 EPS.

Source: Company presentation

What immediately jumped out at us was the projection for $6.7 billion of basic lease rents in FY26. This is mostly because AER generated $6.7 billion of basic lease rents in FY25. So, in other words, in one of the tightest aircraft markets we’ve seen in decades, the company is assuming that basic lease rents will be flat in FY26.

Subscribe now

For example, AER’s owned aircraft average utilization rate has been 99% for the last two consecutive years. In FY25, AER extended 87% of leased aircraft, up from 79% in FY24, highlighting very strong customer demand for leased assets. This is all putting upward pressure on lease rates which should be supportive of basic lease rent growth.

However, it should also be noted that the company is selling planes from its fleet, which naturally has a deleterious effect on lease revenues. Nevertheless, basic lease rents increased by 5% and 2% year-over-year in FY25 and FY24, respectively. These were years when there were a heavy number of asset sales. We would also point out that AER continues to replenish its fleet via aircraft purchases, which helps offset the impact on basic lease rents from aircraft sales.

As a final point, 12% of AER’s current fleet is comprised of COVID-era leases, which will roll off linearly out to 2031-2032 (i.e., around 2% of AER’s fleet will reprice each year out to the early 2030s). This is an important point due to the magnitude of repricing. During AER’s May 2024 Investor Day, management disclosed that AerCap extended an aircraft for an existing operator in 2020. It was a three year extension struck at a sub-$120,000 rate. The aircraft came up for re-lease and was extended again at a lease rate above $165,000. In other words, AER was able to achieve a 40% increase in lease rates even though the aircraft was three years older.

We suspect that given lease rates have continued to increase since then that AER could achieve a very healthy pricing uplift on the 12% of its lease book that will roll off in the coming years. For example, if 2% of the lease book were to reprice with a 50% lease rate uplift, then this should boost basic lease rents growth by 1 percentage point.

It’s possible that another $2 of EPS could emanate from management upgrading its base business guidance. Tying all of this together, we believe a more reasonable EPS range for FY26 is $18-19 per share. The next part in framing the return shareholders are playing for is to determine what multiple AER might trade at in the future.

What multiple is fair?

Historically, AER has roughly traded at between 0.6x to 1.2x of its book value.

Source: Bloomberg

The stock has also traded at a c.7.5x P/E on average over the last decade, struggling to sustain above 10x.

Source: Bloomberg

The fact that AER’s multiple is at the top of its historical trading band, sitting at around 9.5x is not a great starting point. And there’s clearly downside to the extent to which the multiple de-rates back to its historical levels (this is a key reason why the Fund’s position in AER is sized much smaller today).

It is, however, also worth turning our minds to future opportunities that could cause the multiple to rerate. AER management was asked the following question on the Q4 2025 earnings call:

“In the industry, another owner of engine assets announced they’d be looking to convert engines to power turbines to service data centers. Is that something you guys are exploring?”

CEO Aengus Kelly’s answer highlighted AER’s willingness to participate in converting aeroderivative opportunity – that is, converting aerospace engines such as the CFM56 to turbines that power data centers:

“And of course, we’re looking at this. And if it turns out that the demand for what are currently commercial aerospace engines, if that demand to convert them into ground-based power generation for data centers is very durable and is long lived, then of course, we will participate in that, either directly or indirectly, directly by converting the engine into ground-based power generation or indirectly by taking advantage of this surge in demand. But at the moment, we want to make sure that this demand is a durable demand and it’s not fleeting, and that’s our focus at the moment is sizing the market. And from there, then we will participate, as I said, one way or the other.

In terms of the quantum of engines that we have ourselves, if you look at what’s installed on our aircraft, take the CFM56 model, we are the largest owner of CFM56s in the world. And I would remind you that all of our engines are serviceable and have to be returned for the most part in full life conditions, which is very different to a portfolio of engines that is half life or run out engines, and you’re swapping modules to make a serviceable engine. So, when we look at the engine portfolio that we have…on any of those metrics, we would have more engines than anyone in the world.”

The recent pop in the share price of FTAI Aviation (+55% YTD) highlights the immense uplift in value the market is currently willing to ascribe to companies hitching their cart to the AI data center buildout. FTAI recently announced FTAI Power, an initiative to convert CFM56 engines into 25-megawatt turbine units that can generate electricity to power data centers. The stock went vertical.

Source: Bristlemoon Capital; Bloomberg

Our belief is that energy is one of the key bottlenecks in the AI infrastructure buildout. Existing grid capacity is insufficient to power these energy-hungry data centers, which is why we’re already seeing an increasing prevalence of behind-the-meter (BTM) power solutions (i.e., energy systems installed on the customer’s side of the electricity meter rather than on the utility/grid side). The demand for energy is likely to be so great that these BTM energy systems such as aeroderivative turbines could become permanent fixtures to providing power, as opposed to a stop-gap solution that will temporarily bridge these energy needs while the grid plays catch up.

Given the multi-year tailwinds around data center energy demand that we think there is decent visibility over, combined with Aengus Kelly’s openness to participating in the aeroderivative opportunity, there is a scenario where AER significantly re-rates above its past historical multiple range if the company did start selling CFM56 engines to FTAI or if it embarked on converting the engines itself. We believe the former is much more likely, given that FTAI Power required a year of investment to operationalize and it was built on FTAI’s Maintenance, Repair & Exchange (MRE) platform (which is not something AER has).

If the market were to catch wind that AER is now selling engines to FTAI which are then being converted for data center power generation use cases, then this could lead to a re-rating. However, this is a rather speculative prospect to be banking on for further multiple expansion (again, which is why the Fund’s AER position is currently small) and we suspect that Aengus Kelly would prefer that there was not a significant re-rating in AER’s multiple. Rather, we’re almost certain he’d prefer to surreptitiously profit from the aeroderivative opportunity and then use those profits to repurchase large amounts of AER stock at a depressed multiple – the same playbook he’s been using for years.

This isn’t a terrible setup. Management has indicated that the aircraft shortage should persist into the 2030s, meaning that they should be able to continue selling aircraft assets at a healthy premium to book and buying back stock at a lower multiple. But if investors were to suddenly become giddy about AER being an indirect beneficiary of the data center buildout and subsequently bid up the stock, then AER stockholders obviously do well in that scenario too.

AER is one of those rare opportunities where you have very good visibility over the value drivers, but with the stock having re-rated somewhat and many other stocks in this market having significantly de-rated, we are currently holding AER as a smaller sized position in the Bristlemoon Global Fund.

Subscribe now

Disclaimer / Disclosures

The information contained in this article is not investment advice and is intended only for wholesale investors. All posts by Bristlemoon Capital are for informational purposes only. This article has been prepared without taking into account your particular circumstances, nor your investment objectives and needs. This article does not constitute personal investment advice and you should not rely on it as such. This document does not contain all of the information that may be required to evaluate an investment in any of the securities featured in the document. We recommend that you obtain independent financial advice before you make investment decisions.

Forward-looking statements are based on current information available to the author, expectations, estimates, projections and assumptions as to future matters. Forward-looking statements are subject to risks, uncertainties and other known and unknown factors and variables, which may affect the accuracy of any forward-looking statement. No guarantee is made in relation to future performance, results or other events.

We make no representation and give no warranties regarding the accuracy, reliability, completeness or suitability of the information contained in this document. To the maximum extent permitted by law, we do not have any liability for any loss or damage suffered or incurred by any person in connection with this document.

Bristlemoon Capital Pty Ltd (ABN: 22 668 652 926) is an Australian Financial Services Licensee (AFSL Number: 552045).

George Hadjia and Daniel Wu are associated with Bristlemoon Capital Pty Ltd. Bristlemoon Capital may invest in securities featured in this newsletter from time to time.

If you haven’t subscribed, you can join 5,600 others who enjoy our deep dives and investment insights here:

Subscribe now

Free subscribers will only receive a partial preview of our reports. The remainder of our reports, which contain the deeper analysis, are reserved for paid subscribers. Consider becoming a paid subscriber for full access to our reports.

A reminder that you can register for our upcoming fireside chat with Timee CEO Ryo Ogawa via the following link.

Australian wholesale investors looking to invest in the Bristlemoon Global Fund can do so via this link.

Introduction

Towards the end of last year, we started sifting through what we thought was the rubble of the software sector looking for attractive opportunities. Software sector multiples as measured by the IGV had fallen to the lows of the 2022 bear market on fears that AI would “kill SaaS” for a multitude of reasons: some sensationalist (vibe coding replacements for systems of record) and some real (seat loss, margin compression and/or more competition). Evidently, we were still too early as the rubble has subsequently been nuked into glass.

Source: Bloomberg. Note non-GAAP EPS

One stock that looked derisked enough for us to initiate a starter position is Adobe Inc (ADBE), which had derated to sub-20x forward GAAP P/E compared to a post-SaaS transition history spent almost entirely north of 40x. We laid out some initial high-level thoughts in our December quarterly letter as to why the AI disruption narrative for Adobe might be overstated. However, as we spent the past month or so drafting this article (while watching the market speedrun the diSaaSter —> catSaaStrophe —> SaaSmageddon trade), we’ve refined our framework for software stocks we want to own and formed a much more nuanced view on Adobe, which could reasonably be summarized by Superman:

This is not an article where we pontificate our views on why the SaaSpocalypse narrative is overblown, at least for some SaaS companies. There is already so much public discourse arguing both sides that we don’t have much incremental to add; but more importantly, the market doesn’t give a damn about what you or I think and anyone who resolutely believes they know what the world will look like in two years (let alone the 20+ years of software terminal value) is the real irrational actor.

This is also not a deep dive into the history and nitty-gritty details of Adobe’s business segments and products, because that’s not where we think the alpha lies in the present environment. As such, for those who are unfamiliar with the company, we’ve enlisted ChatGPT to provide an overview of Adobe’s product suite and use cases.

Source: Bristlemoon Capital, ChatGPT, Adobe

With this preamble out of the way, we can move onto the real question.

Is Adobe an AI loser?

To us, this question is ultimately about whether the value proposition of Adobe’s creative applications lies in “creation” or “orchestration/workflow”. In a generative AI world, do high-powered professional grade content creation/manipulation tools need to exist? No serious customer is going to vibe code an Adobe product replacement regardless of how frustrated they are with Adobe’s pricing. But might existing and prospective customers decide they need less Adobe in their lives?

To determine whether Adobe stock is a generational buying opportunity or a melting ice cube (or value trap at best), we believe it is critical to answer the following questions:

1. What does being an AI loser look like for Adobe?

2. Are Adobe’s products susceptible to being disrupted by AI?

3. Is Adobe’s business model or economics susceptible to being disrupted by AI?

4. Can Adobe self-disrupt using AI before an AI competitor or startup disrupts it?

5. If AI disruption does occur, where will it materialize?

In seeking to answer these questions, we focus mainly on Adobe’s Digital Media segment, which consists of Adobe’s Creative Cloud suite of applications plus Acrobat and is central to the stock’s terminal value. We believe that any decision to own Adobe stock must be underpinned by the view that the Digital Media business is insulated from AI disruption, as the Digital Experience (DX) segment is very unlikely to offset a deterioration of the core creative business.

We flag upfront that the remainder of this article expends no words contemplating the positioning of the DX business. At a high level, we think the DX products are more insulated from AI disruption because of consumption pricing and potentially higher usage if AI agents proliferate the internet. It is, in a sense, a narrow System of Record for the marketing organization of an enterprise. However, this segment is only 25% of total revenue (unchanged since FY19, so it hasn’t increased in importance), and likely a lower share of earnings given its low-70% gross margin compared to 95% for Digital Media. So, if Digital Media is disrupted by AI, DX is not going to save the day.

What does Adobe being an AI loser look like?

As one of the pioneers of the perpetual license to SaaS transition, Adobe’s core business model is underpinned by monthly Price * Quantity. If AI was disrupting Adobe’s digital media business, we would expect to see any of the following:

• Seat count reduction

• Pricing power erosion

• Margin compression

It has been more than three years since the ChatGPT moment, and Adobe’s financial performance is not representative of a company that is an AI loser.

Source: Bristlemoon Capital, company filings

Adobe does not report seats, churn or net retention, but we can infer from the ARR growth above and price increases in recent years that seat count has not declined post-ChatGPT, nor has pricing power been eroded. The net increase in Digital Media ARR looks healthy, and margins have remained stable at a high level. By all indications, this does not look like the financial profile of a business that is being disrupted by AI, and the first-order conclusion is that the current narrative-driven derating has completely dislocated from fundamentals. But, seeing as the market price is the consensus NPV of all future expectations (rational or otherwise), we need to turn our minds from past performance to the future and think about how things are evolving.

Subscribe now

Are Adobe’s products susceptible to being disrupted by AI?

Before answering this question, let’s take a step back and briefly look at the broad categories of software that exist and what characterizes them as more or less susceptible to AI disruption. There is plenty of existing material on the Internet written about this topic and we are not trying to reinvent the wheel here, so we enlist ChatGPT again to provide a high-level overview.

Source: Bristlemoon Capital, ChatGPT

As we can see, Adobe’s core creative software falls within the most vulnerable category of software – productivity-based application software. This explains why Adobe stock has experienced one of the more extreme de-ratings within the software sector and currently trades below 15x forward GAAP P/E and ~1x PEG ratio. It is this objectively, rather than subjectively, cheap multiple that attracted our attention initially.

To answer whether Adobe’s products are susceptible to AI disruption, it is helpful to separate Adobe’s “job to be done” into creation and workflow. Within the creative fields, some software tools are used mainly for the creation of digital assets (e.g. Autodesk Maya, Blender, Adobe Photoshop in some cases) and other tools are used mainly for workflow orchestration or getting a project from ideation through to finished product.

If workflow is the most important job to be done by Adobe software, the company should ultimately prove to be a beneficiary of generative content creation. Generative image and video models that enable much faster and cheaper content creation should drive an explosion of content, some of which will require software like Adobe Photoshop or Premiere Pro to polish into a professional finished product (i.e. TAM expansion).

If creating content is the most important job, the value of Adobe’s software is at risk of erosion because an alternative tool (an image or video model) can do it more efficiently and at lower cost. A high-powered, hard-to-master application like Photoshop gets relegated to the utility workflow layer where minor edits are made to already-generated content. Even if the Adobe workflow remains essential, pricing power may be impaired if users and/or expense approvers perceive greater value in the creation stage.

In simple terms:

• Creation is highly susceptible to AI disruption and is already happening

• Workflow orchestration is less susceptible to AI disruption and is not happening (yet)

Much of the bear case for Adobe’s creative software rests on people (whether amateur or professional) being able to create images, illustrations, videos etc. using just natural language prompts instead of spending years mastering Adobe’s feature-rich but complex user interface (UI). Thus, even as demand for content creation explodes, Adobe’s seat-based business model may shrink.

We think this perspective largely misunderstands Adobe’s role in the content production process. Even before the proliferation of generative AI, “creation” – as in something from nothing – rarely started within Adobe’s product suite. For example, take Photoshop and image editing. The image most likely originated as pixels on a digital image sensor, which was then uploaded into Photoshop for the editing process. If the intended output is a composite image, Photoshop’s role as the orchestration layer is even more important, allowing the user to combine and manipulate disparate visual elements within one image.

The same can be said about software like Premiere Pro (and competitors such as Avid Media Composer or DaVinci Resolve) for video post-production, i.e. they are used for the editing stage once the raw footage has already been produced. Sure, Nano Banana Pro or Kling 3 might be able to one-shot a simple creative prompt, but natural language edits using these models can be clunky and imprecise (essentially re-prompting the model to generate the output slightly differently). In these cases, having precise human-controlled editing tools may even be more time and cost efficient.

Of course, this will depend on the creative field as well, e.g. illustrators or digital artists who start with a blank canvas and a flexible creative brief will be most disrupted by “good enough” AI image generators. So, we don’t dispute the assertion that some Adobe products will face some disruption, but AI’s raw generation capability is likely overstated as a risk.

Subscribe now

Having said all this, the one reason we find it so hard to disprove a steelman AI disruption thesis is that the AI we have today is the worst it is ever going to be. Therefore, dismissing the disruption risk because AI image gen can’t one-shot a complicated scene with perfect fidelity or AI video still occupies the uncanny valley is likely to be shortsighted. Likewise, a strongly held belief that Adobe will retain its central role as workflow orchestrator is a bet that AI development won’t advance to a stage where precise compositions and edits are possible via text prompt or that Adobe tools won’t be relegated to the backend of an agentic natural language UI.

Source: X

Ultimately, all tools are a means to an end. For the commercial creative field, the end is a piece of content that adheres to some specification. Adobe is one of multiple tools used to achieve that end, and AI that can directly (now, or in future) create the output poses a threat to the incumbent tools and workflows used to create that same output. For this reason alone, it would be difficult to size up ADBE as a high conviction investment.

The remainder of this article is reserved for paid subscribers, where we discuss whether Adobe’s business model and economics are susceptible to AI-disruption, what Adobe can do to mitigate any pressure on seats and pricing, and turn our minds to how AI disruption might materialize in Adobe’s business in the future. A big thank you to our paid subscribers!

Read more

We’re pleased to share that Bristlemoon Capital will be hosting a live webcast discussion with Ryo Ogawa, Founder & CEO of Timee.

Timee is Japan’s leading marketplace for spot-work labour, enabling real-time matching between flexible workers and employers to address the acute labour shortages that exist in Japan. Ryo-san, who launched Timee in 2018 when he was just 21, has scaled the platform to 12.7 million registered workers, reshaping how short-term labour is matched within Japan.

The conversation will be led by George Hadjia, Chief Investment Officer of Bristlemoon Capital, and will explore Timee’s business model, long-term strategy, and the structural forces shaping the flexible labour market in Japan.

For background reading, you may wish to review our most recent December 2025 quarterly letter, which includes our high level investment thesis on Timee and reasons for holding it in the Bristlemoon Global Fund.

Zoom registration link: https://us06web.zoom.us/webinar/register/WN_8GFWs9ImTUKWz8spxMtAgQ

Webcast details

• Date: Thursday February 12th

• Time:

  • 8am JST

  • 10am AEST

  • 6pm ET (Wednesday February 11th)

  • 3pm PT (Wednesday February 11th)

• Format: Zoom webcast

• Duration: ~45–60 minutes discussion, followed by audience Q&A

• Language: Japanese with English interpretation

Discussion topics will include

• Timee’s origin story and founding philosophy

• Product-market fit and user engagement dynamics

• Unit economics, monetisation, and scalability

• Competitive landscape and barriers to entry

• Management’s long-term vision and capital allocation priorities

George will be asking Ryo-san a prepared set of questions, followed by time for audience Q&A.

A replay will be made available to registered attendees following the event.

We hope you can join us for what should be a thoughtful and wide-ranging discussion.

Subscribe now

Disclaimer / Disclosures

The information contained in this article is not investment advice and is intended only for wholesale investors. All posts by Bristlemoon Capital are for informational purposes only. This article has been prepared without taking into account your particular circumstances, nor your investment objectives and needs. This article does not constitute personal investment advice and you should not rely on it as such. This document does not contain all of the information that may be required to evaluate an investment in any of the securities featured in the document. We recommend that you obtain independent financial advice before you make investment decisions.

Forward-looking statements are based on current information available to the author, expectations, estimates, projections and assumptions as to future matters. Forward-looking statements are subject to risks, uncertainties and other known and unknown factors and variables, which may affect the accuracy of any forward-looking statement. No guarantee is made in relation to future performance, results or other events.

We make no representation and give no warranties regarding the accuracy, reliability, completeness or suitability of the information contained in this document. To the maximum extent permitted by law, we do not have any liability for any loss or damage suffered or incurred by any person in connection with this document.

Bristlemoon Capital Pty Ltd (ABN: 22 668 652 926) is an Australian Financial Services Licensee (AFSL Number: 552045).

George Hadjia and Daniel Wu are associated with Bristlemoon Capital Pty Ltd. Bristlemoon Capital may invest in securities featured in this newsletter from time to time.

Key performance contributors in the month of December included AppLovin, Synopsys, and PAR Technology Corporation. Notable performance detractors over the same period included Timee, Fair Isaac Corporation, and Uber Technologies. Notable positive contributions from the Fund’s short book in December included National Vision Holdings, Nike, and Starbucks.

In the letter (accessible via the link below) we discuss some of the new positions in the Fund – Adobe, Baltic Classifieds Group, and Timee – as well as give a more detailed investment thesis around our recently initiated position in Floor & Decor Holdings.

By accessing the link below, you acknowledge and agree that you are an Australian wholesale client as defined by Section 761G of the Corporations Act 2001.

Bristlemoon Global Fund December 2025 Quarterly Report Vf

1.29MB ∙ PDF file

Download

Download

Investing in the Bristlemoon Global Fund

For new investors looking to enter the Fund, more information can be found here.

We appreciate your interest in Bristlemoon Capital and hope this letter provides a clear view of our investment philosophy, current portfolio positioning, and ongoing efforts to compound capital prudently over time. If our investing approach resonates with you, we encourage you to connect with us to explore an investment in the Fund.

Subscribe now

Disclaimer / Disclosures

The information contained in this article is not investment advice and is intended only for wholesale investors. All posts by Bristlemoon Capital are for informational purposes only. This article has been prepared without taking into account your particular circumstances, nor your investment objectives and needs. This article does not constitute personal investment advice and you should not rely on it as such. This document does not contain all of the information that may be required to evaluate an investment in any of the securities featured in the document. We recommend that you obtain independent financial advice before you make investment decisions.

Forward-looking statements are based on current information available to the author, expectations, estimates, projections and assumptions as to future matters. Forward-looking statements are subject to risks, uncertainties and other known and unknown factors and variables, which may affect the accuracy of any forward-looking statement. No guarantee is made in relation to future performance, results or other events.

We make no representation and give no warranties regarding the accuracy, reliability, completeness or suitability of the information contained in this document. To the maximum extent permitted by law, we do not have any liability for any loss or damage suffered or incurred by any person in connection with this document.

Bristlemoon Capital Pty Ltd (ABN: 22 668 652 926) is an Australian Financial Services Licensee (AFSL Number: 552045).

George Hadjia and Daniel Wu are associated with Bristlemoon Capital Pty Ltd. Bristlemoon Capital may invest in securities featured in this newsletter from time to time.

If you haven’t subscribed, you can join 5,449 others who enjoy our deep dives and investment insights here:

Subscribe now

Free subscribers will only receive a partial preview of our reports. The remainder of our reports, which contain the deeper analysis, are reserved for paid subscribers. Consider becoming a paid subscriber for full access to our reports.

Bristlemoon readers can also enjoy a free trial of the Tegus expert call library via this link.

Australian wholesale investors looking to invest in the Bristlemoon Global Fund can do so via this link.

Introduction

Private equity in public markets – it’s a phrase often used by fund managers, and one that borders on being trite. But it’s no wonder that fund managers tend to overuse the expression – private equity as an asset class has performed decently well. The Cambridge Associates US Private Equity Index achieved a pooled annual net return of 12.09% over the last 25 years, compared to the annualized returns of 8.46% and 9.38% for the Russell 2000 and the S&P 500 indices, respectively[1].

There is, however, a publicly listed company that has implemented the “private equity in public markets” concept better than anyone else. That company is TransDigm (TDG), and its financial and stock price performance trounces that of even the most respectable private equity outfits.

TransDigm was formed in 1993 with initial equity of $25 million, after which not a dollar of additional equity was contributed to the company. Today, TDG’s equity value is $72 billion. In other words, TransDigm has returned 2,880x its initial invested equity.

Since its 2006 IPO, the stock has increased from $21 per share to $1,270 per share, a 23.1% annualized return for almost 20 years! This has been supported by revenues and earnings growing at 17% and 23%, respectively, over the last 20 years. TransDigm, with these numbers, is one of the all-time great stock market compounding stories.

We will look at what makes TransDigm a phenomenal business, some of the risks inherent in the company’s approach to achieving such stellar results, and what the future could hold for TDG.

Business Overview

TransDigm is a designer, producer, and supplier of highly engineered aircraft components. In fact, the company sells its components into nearly every commercial and military aircraft worldwide. If you’ve ever boarded a plane, it’s a virtual certainty that you’ve interacted with one of their many products.

TDG’s products are numerous and varied, with the company supplying components such as aircraft passenger seatbelts, cockpit security systems, actuators, and the list goes on. Below is an overview of TDG’s products.

Source: 2024 TransDigm Analyst Day

TDG’s business is organized into three segments:

1. Power & Control – includes operations that make components and systems to provide power to or control the power of the aircraft using electronic, fluid, power and mechanical motion control technologies. This includes products such as mechanical actuators, ignition systems, specialized pumps and valves, batteries, etc.

2. Airframe – includes operations that make products used in airframe applications. This could be latching and locking devices, connectors, cockpit displays, lavatory components, seat belts, thermal protection and insulation. In addition to systems and components, TDG also has some services based businesses such as wind tunnel and jet engine testing services, and equipment testing solutions.

3. Non-aviation - includes seat belts for ground transportation, actuators and controls for space applications, parts for gas turbines, and refueling systems for heavy equipment used in the mining and construction industries.

TDG’s operating units sell components into the original equipment and aftermarket channels. Their customers include OEMs such as Boeing and Airbus, suppliers that sell to the OEMs, airlines, third party maintenance suppliers, and military buying agencies.

Below is a snapshot of TDG’s revenue mix by end market for FY24.

Source: Bristlemoon Capital; company filings

It is worth noting that TransDigm is a holding company that owns a collection of mostly aerospace component suppliers. Each business operates separately under its own brand, as opposed to going to market under the TransDigm name (we touch on this highly decentralized approach below). Importantly, all of these businesses have been acquired.

Subscribe now

Private equity approach

TransDigm is essentially a private equity firm that specializes in the aerospace industry. Since its founding in 1993, TransDigm has acquired more than 90 businesses. Today, these acquired businesses are split into 51 independently run operating units that collectively produce parts for nearly every commercial and military aircraft platform.

Below is a history of the numerous acquisitions TDG has made both as a private company as well as after it went public.

Source: 2024 TransDigm Analyst Day

The company employs significant leverage to enact its acquisitive growth strategy, running net debt / EBITDA at almost 6x on average for the last 10 years. Usually this significant level of leverage would concern us and rule out the company as an investment prospect.

We’d point to two things that give us comfort: 1) TDG derives a significant portion of its earnings from the aftermarket (c.3/4 of adjusted EBITDA comes from the aftermarket), which is a much more stable earnings stream; and 2) the business navigated and survived two of the most severe exogenous shocks (the GFC and COVID), testing their highly levered strategy and showing that their model can withstand material demand shocks and credit crunches. Aside from a -6% EBITDA decline in FY20 and -4% in FY21, TDG has never experienced a full-year EBITDA decline, hinting at a well-managed, resilient business that is capable of supporting elevated amounts of leverage.

OEM and aftermarket

TDG sells its products into original equipment manufacturer (OEM) markets as well as the aftermarket. TDG supplies components to OEMs which include the major global aircraft manufacturers (e.g., commercial aircraft manufacturers such as Boeing and Airbus, as well as private aircraft manufacturers like Gulfstream, Bombardier and Cessna) and defense contractors (e.g., Lockheed Martin and Northrop Grumman). Often, however, TDG is selling products to other suppliers, who then sell to companies such as Boeing and Airbus.

The OEM market is defined as the sale of components to the aircraft manufacturer (or related suppliers that sell to that aircraft manufacturer) for the production and initial delivery of new aircraft.

The aftermarket, on the other hand, refers to the supply of replacement parts for aircraft that are already in operation. It is a much more stable stream of revenues that is correlated to the distance flown by aircraft; revenue passenger kilometers (RPKs) have grown at 5-6% for decades, providing a nice demand tailwind. In other words, more flight hours lead to more wear on an aircraft’s components, necessitating periodic replacement of those parts.

It is worth noting that the aerospace industry follows a two tier pricing model. The OEMs (e.g., Airbus and Boeing, or a supplier selling into these firms) pay TDG one price for the original part and additional warranty parts. The OEMs exert their duopolistic leverage over the aerospace supply chain to push down the prices they pay for these parts so that they can improve the margin they make on selling planes.

These planes are then sold to airlines, who are then responsible for purchasing aftermarket replacement parts at a higher price than what the OEMs initially paid. Sometimes there are long-term agreements (LTAs) that specify the aftermarket pricing that TDG can charge, but typically there’s flexibility for TDG to exercise its pricing power in the aftermarket.

The graphic below gives a snapshot of the aerospace value chain and highlights where TDG derives the bulk of its earnings from.

Source: Compounding Labs

In essence, the OEM market is where TDG gets its proprietary parts designed into a platform, securing an annuity-like revenue stream for the decades-long aftermarket phase (which is much higher margin than selling to the OEM market). In this way, TransDigm is implementing a razor razor-blade model: sell the parts cheaply upfront and make the margin on the long tail of replacement demand for that part (note that TDG’s philosophy is that you don’t have to lose money on the upfront sale to the OEMs, even if many other suppliers do).

In the original 2003 investment memo from Berkshire Partners when they were looking to acquire the then-private TransDigm, it was estimated that the company will usually achieve $100 of aftermarket sales over the lifespan of an aircraft from each $1 of OEM sales[2]. This long-tail of high margin replacement demand is one of the most attractive aspects of TransDigm’s business.

Why TransDigm is a Great Business

There are reasons why TransDigm’s aftermarket-focused strategy has worked spectacularly well. Essentially, the businesses TDG has acquired have unique characteristics that have enabled the company to raise the prices of its products over long stretches of time without customers switching to other suppliers. We will dig into each of these characteristics in turn.

Read more

If you haven’t subscribed, you can join 5,421 others who enjoy our deep dives and investment insights here:

Subscribe now

Free subscribers will only receive a partial preview of our reports. The remainder of our reports, which contain the deeper analysis, are reserved for paid subscribers. Consider becoming a paid subscriber for full access to our reports.

Bristlemoon readers can also enjoy a free trial of the Tegus expert call library via this link.

Australian wholesale investors looking to invest in the Bristlemoon Global Fund can do so via this link.

Introduction

As a recap, Hemnet is the monopolist for online real estate listings in Sweden with a c.90% share of listings. Think of it as the Swedish-equivalent of Australia’s realestate.com.au. The growth in the business had been spectacular: Hemnet’s property seller revenue increased by 6x between FY18 and FY24 (a 34.8% annual growth rate), and consolidated operating profit increased by 8.6x over the same period (a 43.3% annual growth rate). We wrote about the stock in our March 2025 quarterly letter, with a bullish thesis that mapped out the potential for a continuation of this strong performance. We also discussed Hemnet’s pricing actions here in a subsequent article.

What has happened in the interim has been a deterioration in the growth rate of Hemnet and a steep decline in the share price. For context, Hemnet’s share price fell from SEK 420 in February 2025 to around SEK 160 in recent weeks. We wanted to explain in greater detail what has happened to cause such a precipitous share price decline, whether the bear case has merit, and what the future could look like for Hemnet from here.

Is Hemnet’s moat broken and its future growth prospects permanently impaired, or is this a monopoly asset whose price has been buffeted by erroneous narratives, with the company’s shares now trading at an abnormally low 22x forward EPS?

Read more

If you haven’t subscribed, you can join 5,305 others who enjoy our deep dives and investment insights here:

Subscribe now

Free subscribers will only receive a partial preview of our reports. The remainder of our reports, which contain the deeper analysis, are reserved for paid subscribers. Consider becoming a paid subscriber for full access to our reports.

Bristlemoon readers can also enjoy a free trial of the Tegus expert call library via this link.

Australian wholesale investors looking to invest in the Bristlemoon Global Fund can do so via this link.

Table of contents

• Introduction

• Are we in an AI bubble?

  • Demand

  • Willingness to pay

  • Unit economics

  • Supply

• TSMC N2 capex

• ASML EUV demand

• Nvidia revenue outlook

Introduction

Over the past few weeks, we decided to take stock of the ongoing AI investment cycle in an attempt to gauge where we are in the cycle and how much further there is for fundamentals to run. Essentially, we wanted to understand the magnitude of AI capex baked into current share prices and whether there could be substantially more upside or downside to estimates and thus prices.

Unfortunately, Jensen Huang stole some of this post’s thunder with his revelation at GTC last week of $500 billion of booked Blackwell and Rubin orders to ship in CY 2025 and 2026. We arrive at similar magnitudes of revenue and explore what we think this means for a subset of the AI supply chain. We have limited exposure in the Fund to AI capex beneficiaries and instead have preferred to own early and obvious AI application (verb) winners. As such, this discussion is limited to the supply chain stocks that we are interested in (ASML, NVDA and TSMC), but the implications might be applicable to the AI cycle participants more broadly. As always, readers should do their own due diligence and draw their own conclusions.

The chart that pushed Nvidia to $5 trillion market cap

This exercise is also motivated by a quote from Eowyn, Shieldmaiden of Rohan: “The women of this land learned long ago that those without swords can still die upon them.”

In other words: if the AI capex beneficiaries leading the markets higher were to falter, we believe they could bring the entire market down with them; being diversified or underweight AI exposure is no consolation on the way up or down. It should not be lost on anyone that AI related stocks have accounted for 75% of S&P 500 returns, 80% of earnings growth and 90% of capex growth since ChatGPT launched in November 2022[1]. We also can’t ignore the development that in 1H 2025, AI capex contributed more to US GDP growth than the US consumer[2].

Are we in an AI bubble?

We don’t think hyperbole around the “AI bubble” (on both sides) is helpful or instructive for rational decision making, so we endeavored to approach this analysis from both a top down and bottoms up perspective while checking any preconceptions at the door.

Starting with the top down view, we compare the available data points on demand and supply for AI.

Demand

Consumer demand for AI remains robust. OpenAI reported 800 million weekly users in September, up from 500 million in March. Google just disclosed 650 million Gemini app MAU on its Q3 2025 call, up from 400 million at Google I/O in May. Over 2 billion monthly users benefit from AI Overviews, while the initial rollout of AI Mode reached 100 million users in July. Meta likewise has disclosed some big user numbers – over 1 billion Meta AI MAUs on its Q3 call.

Source: NBER “How people use ChatGPT”

We know that WAUs and MAUs are metrics that don’t necessarily align with actual usage. For that, token demand is the better measure. Google has sporadically disclosed monthly tokens processed, and that number has grown exponentially. The company also disclosed that the Gemini API processes 7 billion tokens per minute, while OpenAI’s API processes 6 billion per minute – good for 300 trillion and 260 trillion tokens per month, respectively. Yes, tokens processed have been inflated by reasoning models, but we don’t have a breakdown of reasoning vs query growth. In any case, we believe higher quality output from small reasoning models vs non-reasoning models should drive greater AI usage.

Source: Bristlemoon Capital, Google

While OpenAI hasn’t consistently disclosed tokens processed, the “How people use ChatGPT” research paper by NBER does include data showing daily messages per WAU is increasing for all cohorts.

Source: NBER “How people use ChatGPT”

Earlier this month, there was some data from Apptopia (a third-party app intelligence firm) showing OpenAI app DAUs have plateaued since September, while average sessions and time spent have also declined. This period coincided with both the botched rollout of GPT-5 and Google’s launch of Nano Banana which pushed Gemini to the top of the app download charts, so we would be hesitant to exclaim “Aha! OpenAI has peaked!” on this data alone. Based on the most recent OpenAI and hyperscaler commentary, it appears that demand is still well in excess of supply and expected to remain so into 2026, despite the rapid AI server capacity additions.

Subscribe now

Enterprise AI demand has, unsurprisingly, lagged consumer adoption. One report recently making waves is a July study by MIT NANDA canvassing 1H 2025 that found 95% of enterprises were getting zero return on $30-40 billion of AI investment. To borrow the 20 second executive summary:

“Tools like ChatGPT and Copilot are widely adopted. Over 80 percent of organizations have explored or piloted them, and nearly 40 percent report deployment. But these tools primarily enhance individual productivity, not P&L performance. Meanwhile, enterprise grade systems, custom or vendor-sold, are being quietly rejected. Sixty percent of organizations evaluated such tools, but only 20 percent reached pilot stage and just 5 percent reached production. Most fail due to brittle workflows, lack of contextual learning, and misalignment with day-to-day operations.”[3]

Source: MIT NANDA

Our interpretation of the findings is that 1) enterprises are generally not ready to successfully implement AI, whether that be lack of sponsorship, data cleanliness, or resistance to change; and 2) the models – particularly narrow domain agentic models – are just not good enough yet. We don’t have a strong opinion on what enterprises can do to overcome the former limitations as each organization is different, but on the latter, we would just remind readers that the AI in deployment today is the worst that AI will ever be.

Yet even so, one very compelling enterprising use case for AI has already emerged – the AI coding agent. Many tech companies have revealed a high and growing percentage of their code base being initially written by AI coding agents (whether true or not, we can’t verify). Anthropic is the leader in this area and has just recently projected $20-26 billion in annualized revenue in 2026, up from $9 billion projected exiting 2025, and $1 billion in January 2025[4]. We believe a substantial majority of that revenue comes from enterprise, and specifically Claude Code. So, it seems that if an AI use case achieves strong product market fit, enterprises are willing to pay up for measurable improvements in productivity.

As an aside – AI bubble doomers often cite de minimis AI revenue as the key reason why the bubble must pop. Yet this MIT report analyzing 300 public GenAI implementations captured $30-40 billion of GenAI investments. Sure, a majority might be spent on headcount and consultants, but some AI service provider must be receiving some of this as revenue…

Willingness to pay

Having insatiable user or token demand is great, but means nothing and is unsustainable if given away for free. When we consider AI demand from a willingness to pay angle, the picture is perhaps somewhat less rosy. We think there is a strong case to be made that enterprises will pay if they see value in a product or service (basically the entire premise of the SaaS market). However, consumers evidently haven’t shown a strong willingness to pay for access to more capable models and higher rate limits.

It’s been reported[5] that of ChatGPT’s 800 million WAUs, only 5%, or ~40 million, are paying users. Subscriptions account for 70% of OpenAI’s $13 billion ARR, with API access contributing the remainder. This basically says to us that the 800 million users, taken as a whole, find ChatGPT no more valuable than a freemium mobile game, where a single digit percentage of “whales” account for substantially all a game’s revenue.

We would argue that ChatGPT provides more value than Candy Crush, but the difficulty is clearly in getting consumers to pay for the premium experience when a small, non-reasoning model is probably enough to satisfy the average user’s requirements. To that end, it is clear that OpenAI is preparing a shift towards an advertising model for ChatGPT, having hired Fidji Simo and several other Meta product alumni to build out the “applications” (i.e., non-research/AI) organization.

It remains to be seen whether OpenAI can build an advertising business rivaling the scale of Meta or Google, but the precedent of supporting $100+ billion annual capex buildouts using ad dollars is a strong one. Obviously, ad dollars don’t just materialize out of thin air, so unless AI adoption is driving incremental GDP growth, OpenAI will need to take share of ad budgets from somewhere.

Subscribe now

Unit economics

Taking a step back, a lack of consumer willingness to pay should not be conflated with weak unit economics. The “AI is a bubble” justification we take the most issue with is this misconception (often magnified by the media) that serving AI models to users is a loss-making endeavor. This is most often misrepresented as spending $2 or $3 dollars to generate $1 of revenue, with the implication being that these AI labs will never be profitable.

A rudimentary understanding of accounting is all that is needed to disabuse oneself of this misrepresentation. The Information reported that OpenAI generated an operating loss of $7.8 billion in H1 2025 on $4.3 billion in revenue, good for $3 in costs to generate $1 of revenue[6]. However, buried at the end of the report is a line stating that cost of revenue was only $2.5 billion, meaning gross profit was $1.8 billion for a 42% gross margin. This indicates that serving current models to users carries positive unit economics, with substantially all the operating loss attributable to training the next generation of models. This means that in time and at scale, frontier model providers will be capable of generating positive net profit.

To understand just how profitable model inference can be, we ran our own back-of-the-envelope tokenomics analysis. Note there are a lot of assumptions that go into this – we are aiming for ballpark estimates here. In summary, depending on model size and API pricing or paying user mix, inference gross margins might range from ~50% to as high as ~80%.

Source: Bristlemoon Capital

For the purposes of the analysis, we used DeepSeek V3 to approximate a medium-sized Mixture of Experts model that might be used in production. We also assume DeepSeek V3 API pricing, which is extremely low compared to the closed source frontier model pricing (e.g. GPT-5 is $1.25/$10.00 input/output and GPT-5 mini $0.25/$2.00). We haven’t found any reliable information on the sizes of the various GPT-5 models or which models are used for ChatGPT, so we’ve gone for the most conservative assumptions that generate the most conservative revenues.

As we can see, even with some of the cheapest API pricing in the market, served on the most expensive (on a GPU/hour basis) AI servers currently available, inference via API still commands a respectable gross margin for the model provider. This of course assumes the provider also owns the servers (e.g. Google or Microsoft) – if they rent the servers from a CSP, the gross margin will be lower.

We can also see that for a subscription model where 5% of users are paying $20 a month, the service provider can still generate attractive gross margins depending on the level of utilization and other optimizations they can achieve. Apptopia data indicates US ChatGPT users spend 20-25 minutes a day in the app, so one could even argue that 10 minutes of GPU active time is excessive, especially for non-thinking chats (i.e., the majority of free ChatGPT usage). Of course, these are averages, and the many free users who might use ChatGPT once a week may be more than offset by the power users on the $200 pro tier.

One final thing to note for OpenAI specifically is that it has a revenue share with Microsoft of just under 20%, in addition to renting server capacity from Azure. Yet even with these incremental contra revenues and COGS, the company is still achieving a 42% gross margin. As that revenue share declines through 2030 and OpenAI starts owning more of its servers (the 16GW it has signed directly with Nvidia and AMD), we expect gross margin to expand, even before considering additional monetization avenues such as advertising.

Subscribe now

Supply

Finally, we get to token supply, or more accurately, the demand for AI infrastructure. We see two key debates here: 1) how much (more) will the hyperscalers spend in capex, and 2) what kind of ROI are they earning on this capex?

The debate over how much AI capex can or will be spent is, in our opinion, a straightforward one to settle. The hyperscalers – mostly comprised of the world’s biggest cash flow printing presses – see the pursuit of AGI as existential. Some are more fanatical (Meta, Google, Oracle) than others (Microsoft, Amazon) but all have said they would rather have too much compute than too little. The ones pursuing AGI in particular are caught in a prisoners’ dilemma where the Nash equilibrium is for all involved to invest as aggressively as possible in that pursuit.

These hyperscalers have significantly more capacity to spend. If we compare current Bloomberg sellside consensus operating cash flows to capex for the five hyperscalers, we see that capex is “just” ~2/3^rd of operating cash flow for the group. Only Oracle is currently, and is projected to continue, spending more than its cash flow on capex. If the group were to spend up to 100% of their operating cash flows on capex (i.e. free cash flow = zero), current sellside capex estimates would need to increase by as much as 50%. If these companies were to collectively lever up to 1x EBITDA to invest in AI, their capex spend would be 3x higher than current estimates!

Note this doesn’t include any OpenAI investment in owned infrastructure (to the extent they can afford to pay for it…) nor any neocloud capex.

Source: Bristlemoon Capital, Bloomberg

How much the hyperscalers decide to invest in AI capex will ultimately come down to what return on investment they expect to generate. The ROI side of the debate is undoubtedly more subjective/biased because no one can predict with any degree of confidence what revenues might (or might not) be unlocked in future years as a result of ongoing AI investment. AI bulls will point to the acceleration of revenues at the hyperscalers (short term), plus historic precedent of the value unlocked by massive capex cycles (railroads, fiber etc; long term), as evidence of at least an acceptable level of ROI. Bears will point to the massive and growing operating losses reported by the frontier labs and say there is no ROI on AI investment.

Our opinion is that the returns will (for now) fall somewhere in the middle. We’ve established above that unit economics of serving inference are quite attractive, with the hurdle being a relatively low willingness to pay for today’s models. Outside of the standalone frontier labs (e.g. OpenAI, Anthropic), it is difficult to precisely ringfence how much revenue the hyperscalers are generating specifically from their AI investments. It is clear that Meta, Google, Azure and AWS are reporting accelerating revenue growth; what’s not clear is the extent to which AI is driving this acceleration.

We think it would be unfair to ascribe zero benefit to these companies’ AI investments, particularly for the likes of Meta and Google where AI-based recommender and ad models are clearly driving results. Likewise, it would be intellectually dishonest to claim that none of the AI capex is wasted and that these companies couldn’t have achieved similar results with some fraction of their current spend. For example, we know that Meta allocates some portion of its total AI server capacity to “core AI” (the ads business) and some portion to “gen AI” (the frontier superintelligence stuff), and that mix changes over time depending on the opportunity set.

The simplest way we can think about ROI is to again look at sellside estimates. We know that the sellside capex forecasts broadly reflect the continuation of the current AI investment cycle (even though they tend to underestimate the capex). We strongly believe that, collectively, sellside are not explicitly modeling specific AI revenue streams for these hyperscalers. Rather, they are typically modelling company guidance and extrapolating growth rates into the future. As such, any AI monetization breakthroughs are not captured in sellside earnings estimates, while the capex burden is. Even so, we can see that the incremental ROI on the hyperscalers’ total capex is still expected to be decent, and certainly above their cost of capital (while returns for individual hyperscalers may vary).

Source: Bristlemoon Capital, Bloomberg

We think the market has gotten accustomed to these companies generating outlandishly strong ROIC year-in year-out with their increasing returns to scale and dominance of their massive end markets. We note the above ROI is incremental, meaning their strong historical ROIC is simply being diluted during this capex investment cycle, rather than being imminently reset downward to a mid-teens level. In any case, a mid-teens post-tax ROIC with nearly 100% reinvestment rate at the scale these companies are operating at would be a textbook example of an attractive “compounder”.

Should the sellside be underestimating the monetization potential of all this AI investment, the realized incremental returns will be even more attractive. But should this mostly cashflow-funded investment really be a waste of capital, these companies can reduce their capex by 20%, 30% or even 50%, without necessarily hurting the performance nor the fundamental attractiveness of their core businesses.

Now that we have established our position on AI demand and supply, we discuss in the paid section what we think this means for TSMC’s capex, the demand for ASML EUV tools, and implications for Nvidia’s revenue. Consider becoming a paid subscriber for full access to our analysis.

TSMC’s N2 capex

When TSMC reported its Q3 2025 earnings last month, the company narrowed its 2025 capex guide to $40-42 billion from $38-42 billion but did not comment on the 2026 capex outlook. Sellside consensus for 2026 has been raised by ~5% to $46 billion, but we think this may still be too low.

The first thing to note is that TSMC typically spends ~70% of its capex on advanced nodes (N7 and below), 10-20% on specialty mature nodes, and 10-20% on advanced packaging and other backend processes. Within its budget for advanced nodes, TSMC is typically investing across multiple nodes in any given year. The majority is allocated to the current leading edge node being ramped (N2), with the remainder spent on both further expanding capacity in the sub-leading node (N3) and laying the physical groundwork (land, fab shell construction) for the next node (A14).

TSMC management has consistently said that they expect N2 to be a bigger node than both N3 and N5, based on the number of N2 tape outs so far. Sellside estimates of N3 and N5 capacity are all over the place, but we have seen estimated peak capacity of 150-170kwpm for each node. If N2 is to be TSMC’s biggest node, it would suggest peak capacity of at least 170kwpm, possibly even 180kwpm. In addition to the volume of N2 tape outs to date, HPC and smartphone volumes are also expected to converge at N2.

Historically, smartphone customers (namely Apple) would lead the ramp of each new node, with HPC – both CPU and GPU/ASIC – lagging by one or two generations. This meant that as Apple prepared to move to the next node, HPC customers would step in and fill the capacity that Apple vacated in the prior node. With N2, however, AMD is expected to lead the high-volume ramp alongside Apple and MediaTek. Nvidia will be using N3 for its Rubin generation of GPUs after using N4 for both Hopper and Blackwell, and is expected to jump to N2 or even straight to A16 with its Feynman generation in 2028.

Noting the above, we believe TSMC may need to spend as much as $85 billion in capex to build out 180kwpm capacity for N2, spread over ~3 years starting in 2025. This translates to ~$4.7 billion capex per 10kwpm, which is materially higher than some of the early sellside N2 capex estimates we heard in the ~$3 billion per 10kwpm range.

We believe the reason for this is twofold: 1) sellside likely significantly overestimates ASML’s EUV tool productivity – which as we will show below may be only a third of nameplate throughput in practice; and 2) for N3, TSMC had the benefit of situating those fabs alongside its existing N5 fabs at its Tainan Gigafab, allowing for tool reuse and conversion of spare N5 capacity into N3. TSMC’s N2 fabs are in Hsinchu, located on the opposite end of Taiwan island, thus limiting capacity conversion of N3 into N2. The Arizona Gigafab situates N5, N3 and N2 fabs together, but given the strategic importance of TSMC’s leading edge capacity and lack of alternatives in the US, we do not believe TSMC will be doing node conversions in the US any time soon.

We arrive at our $85 billion estimate using a bottom-up approach starting with ASML EUV tools given we have the specs for these machines and a rough idea of EUV layer count. We then apply some rules of thumb to scale up our estimated EUV capex into lithography capex, total WFE capex and finally total fab capex. For context, this is likely to be more than the capex for N3 even though EUV layer count remains flat (the most expensive process step), for the reasons we described above.

Source: Bristlemoon Capital, ASML filings

ASML EUV demand

It may be apparent now that although we are discussing TSMC capex, we are actually using this to frame demand for ASML’s EUV tools. Our estimates suggest as many as 80 EUV tools are required for 180kwpm of N2 capacity assuming no tool reuse, again likely spread over a 3-year ramp. This means even if TSMC purchases 20 EUV tools this year, it will still need to materially step up its purchases over the next two years compared to its previous peak of 2021-22 when it was ramping up the N3 node.

We also make an attempt to estimate HBM and DRAM demand for EUV, considering the memory vendors collectively are now the second largest source of EUV demand after TSMC. We would note that this is even more of an approximation than our estimate of TSMC’s EUV demand, considering 1) estimates of HBM demand over the next few years could be wildly off target, and 2) memory vendors engage in even more tool reuse than TSMC as the former typically do not maintain production capacity for more than two trailing nodes.

Source: Bristlemoon Capital, broker estimates

The memory vendors are shaping up for two big years of ASML equipment purchases in 2024 and 2025, so we believe 2026 may be a slower year considering the lack of cleanroom space and a general reluctance by the big 3 to invest in greenfield capacity expansion when they are currently enjoying unprecedented pricing power. However, by 2027 we expect EUV demand to pick up again as the industry will be forced to expand capacity by HBM customers (unless the AI bubble has burst by then…), and lest they start losing share in commodity DRAM to CXMT.

Overall, we see more upside to ASML’s EUV shipments in 2026 and 2027 than the 51 and 61 units being modelled by the sellside. We think TSMC alone could account for more than half of these units, along with a strong rebound in memory in 2027. TSMC’s N2 price – rumored to be $30,000 per wafer – is now said to be a 10-20% increase from N3 compared to a previously rumored 50% jump, but only because N3 prices have now risen from ~$20,000 to above $25,000 each. On Monday, TSMC unveiled an unprecedented four-year plan to raise advanced node pricing (N5, N4, N3 and N2) by as much as 10% a year, citing cost pressures and tight capacity. It is evident that the industry needs an alternate source of leading edge foundry capacity, for both geopolitical and competitive reasons. A resurgent Intel (or even Samsung Foundry) could drive further upside to ASML’s numbers, especially as they are far less efficient with their tools than TSMC.

Nvidia revenue outlook

The last piece of the puzzle for us is reconciling TSMC’s capacity estimates with Nvidia and the broader AI accelerator industry’s wafer demand. Our objective is twofold: 1) to gauge if TSMC needs to further expand its N3 capacity (and the implications for ASML), and 2) to get a sense of Nvidia revenues based on TSMC capacity.

On 1), we have recently seen sellside analysts revise up their TSMC CoWoS capacity estimates to the ~100kwpm range exiting 2026, with ~50% allocated to Nvidia. As CoWoS has historically been the bottleneck for AI accelerator output at TSMC, this is a good gauge of what the upper limit to Nvidia’s revenue could be. 50kwpm CoWoS capacity translates to approximately 12 million reticle-limit GPUs.

On the N3 capacity front, 12 million reticle-sized GPUs at mature yields would require in the vicinity of 430k wafers to produce, or nearly 36kwpm. We estimate that the CPU and networking-related silicon take up half the GPU die area or another 18kwpm. We further assume that Nvidia is 65% of total AI accelerator wafer demand, based on its 50% share of CoWoS and reticle-sized dies compared to smaller custom ASIC dies or AMD’s chiplet architecture. This gets us to over 80kwpm of AI accelerator wafer demand for N3, which would be more than half of N3 capacity assuming 160kwpm peak.

Source: Bristlemoon Capital

This doesn’t quantitatively tell us whether TSMC needs to build or convert additional N3 capacity. However, looking at past nodes, when flagship smartphone customers stepped up to the next gen node, HPC (mainly CPU) and non-flagship smartphone customers would backfill the vacant capacity and then some. Those nodes certainly did not have to contend with tens of thousands of incremental wafer demand from AI customers, which is largely a new dynamic for N4 and especially N3, since both Rubin and TPU v7 are moving to N3P. So, at least qualitatively, this says to us that TSMC will be more constrained on N3 than it has been in sub-leading nodes past. Whether this means entirely new EUV tools are required or just upgrade kits to older converted tools, ASML should stand to benefit next year.

As it relates to revenue for Nvidia, 12 million Rubin GPUs in the NVL144 configuration would be something like $440 billion in revenue, assuming unchanged $/W pricing vs Blackwell. Once we layer networking revenues on top, we get potential VR200 generation data center revenue of as much as $500 billion, likely spread across 5-6 quarters. Note that this is not like-for-like with what Jensen presented at GTC, which was $500+ billion in 2025 and 2026, most of which will be GB200 and GB300, plus the initial ramp of VR200 in late 2026. Our estimates for Rubin are for FY28/CY27, with some overflow into late 2026 and early 2028. We have also not seen established pricing for the VR200 NVL144 yet, so our estimate of $5.3 million per rack based on the GB200’s $/W could be too high. Having said that, we’d be surprised if Rubin GPUs were not priced at some material premium to Blackwell, considering Rubin is a full node jump and also packs more HBM than GB200 (288GB HBM4 vs 192GB HBM3E).

Source: Bristlemoon Capital

We are not suggesting that there will be 30% upside to Nvidia’s consensus revenue estimates in FY27 based on our napkin math. Rather, we simply sought to identify the potential magnitude of delta between consensus and what could be achieved based on Nvidia’s capacity allocation at TSMC. Also, assuming Nvidia’s $500+ billion in AI server revenue across 2025-26 implies over $300 billion in 2026 (FY27), our Rubin estimates suggest 30%+ server revenue growth in FY28 is possible compared to consensus Data Center revenue growth of 20% off a smaller base. Tying this back to our earlier discussion of supply and demand, the hyperscalers alone have sufficient capex capacity between remaining FCF and modest leverage to close the gap between consensus and our estimates, assuming they see a path to an acceptable ROI.

Ultimately, our (perhaps unsatisfying) conclusion is that whether we are in an AI bubble or not is largely irrelevant to making sound investment decisions. What we care about is whether there is a rational basis for further upside to the forecasts of those companies involved in various aspects of the AI infrastructure buildout. Based on our work, we believe there is scope for the buildout to continue and for estimates to move higher through 2026 at the very least, backed by the hyperscalers’ capacity to pay, before we even get into enterprise and sovereign demand.

One final point, which we have not discussed here but did allude to briefly in this note, is that there are bottlenecks and constraints across the AI supply chain, from generation to transmission to chips, which militate against fervent overbuilding and stockpiling – behavior that typically triggers a cyclical downturn or in extreme cases, the popping of a bubble.

Disclaimer / Disclosures

The information contained in this article is not investment advice and is intended only for wholesale investors. All posts by Bristlemoon Capital are for informational purposes only. This article has been prepared without taking into account your particular circumstances, nor your investment objectives and needs. This article does not constitute personal investment advice and you should not rely on it as such. This document does not contain all of the information that may be required to evaluate an investment in any of the securities featured in the document. We recommend that you obtain independent financial advice before you make investment decisions.

Forward-looking statements are based on current information available to the author, expectations, estimates, projections and assumptions as to future matters. Forward-looking statements are subject to risks, uncertainties and other known and unknown factors and variables, which may affect the accuracy of any forward-looking statement. No guarantee is made in relation to future performance, results or other events.

We make no representation and give no warranties regarding the accuracy, reliability, completeness or suitability of the information contained in this document. To the maximum extent permitted by law, we do not have any liability for any loss or damage suffered or incurred by any person in connection with this document.

Bristlemoon Capital Pty Ltd (ABN: 22 668 652 926) is an Australian Financial Services Licensee (AFSL Number: 552045).

George Hadjia and Daniel Wu are associated with Bristlemoon Capital Pty Ltd. Bristlemoon Capital may invest in securities featured in this newsletter from time to time.

[1] https://assets.jpmprivatebank.com/content/dam/jpm-pb-aem/global/en/documents/eotm/the-blob.pdf

[2] https://am.jpmorgan.com/us/en/asset-management/adv/insights/market-insights/market-updates/on-the-minds-of-investors/is-ai-already-driving-us-growth/

[3] https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf

[4] https://www.reuters.com/business/retail-consumer/anthropic-aims-nearly-triple-annualized-revenue-2026-sources-say-2025-10-15/

[5] https://www.ft.com/content/a169703c-c4df-46d6-a2d3-4184c74bbaf7

[6] https://www.theinformation.com/articles/openais-first-half-results-4-3-billion-sales-2-5-billion-cash-burn?rc=oc3j7k

Key performance contributors in the month of September included AppLovin, ASML, and Alphabet. Notable performance detractors over the same period included PAR Technology Corporation, Salesforce, and Hemnet.

In the letter (accessible via the link below) we talk about PAR Technology Corporation, a performance detractor during the quarter, and explain why we still like the stock. We also outline our investment theses for several investments that performed well in the quarter, including ASML, Alphabet and Synopsys.

By accessing the link below, you acknowledge and agree that you are an Australian wholesale client as defined by Section 761G of the Corporations Act 2001.

Bristlemoon Global Fund September 2025 Quarterly Report Vf

750KB ∙ PDF file

Download

Download

Investing in the Bristlemoon Global Fund

For new investors looking to enter the Fund, more information can be found here.

We appreciate your interest in Bristlemoon Capital and hope this letter provides a clear view of our investment philosophy, current portfolio positioning, and ongoing efforts to compound capital prudently over time. If our investing approach resonates with you, we encourage you to connect with us to explore an investment in the Fund.

Subscribe now

Disclaimer / Disclosures

The information contained in this article is not investment advice and is intended only for wholesale investors. All posts by Bristlemoon Capital are for informational purposes only. This article has been prepared without taking into account your particular circumstances, nor your investment objectives and needs. This article does not constitute personal investment advice and you should not rely on it as such. This document does not contain all of the information that may be required to evaluate an investment in any of the securities featured in the document. We recommend that you obtain independent financial advice before you make investment decisions.

Forward-looking statements are based on current information available to the author, expectations, estimates, projections and assumptions as to future matters. Forward-looking statements are subject to risks, uncertainties and other known and unknown factors and variables, which may affect the accuracy of any forward-looking statement. No guarantee is made in relation to future performance, results or other events.

We make no representation and give no warranties regarding the accuracy, reliability, completeness or suitability of the information contained in this document. To the maximum extent permitted by law, we do not have any liability for any loss or damage suffered or incurred by any person in connection with this document.

Bristlemoon Capital Pty Ltd (ABN: 22 668 652 926) is an Australian Financial Services Licensee (AFSL Number: 552045).

George Hadjia and Daniel Wu are associated with Bristlemoon Capital Pty Ltd. Bristlemoon Capital may invest in securities featured in this newsletter from time to time.

Subscribe now

Free subscribers will only receive a partial preview of our reports. The remainder of our reports, which contain the deeper analysis, are reserved for paid subscribers. Consider becoming a paid subscriber for full access to our reports.

Bristlemoon readers can also enjoy a free trial of the Tegus expert call library via this link.

Australian wholesale investors looking to invest in the Bristlemoon Global Fund can do so via this link.

Lessons from launching a hedge fund

Last year, George Hadjia discussed the journey of launching Bristlemoon Capital, following an X (Twitter) post that garnered interest from the FinTwit community. One year following the launch of the Fund, George details further lessons from building Bristlemoon.

It is worth opening this post with some of the more memorable comments I received from people before launching Bristlemoon Capital.

• "How are you going to raise any money?"

• "If you don't partner with someone [a capital partner] I can tell you right now it's going to be a miserable failure".

• "You should go to another fund first and get a track record".

• "You're not smart enough to work at a hedge fund".

Starting a fund requires grit. Most people will be encouraging and supportive when you divulge your fund launch plans. Others won’t be. Having full conviction in what you’re trying to build from the outset, and the tenacity to see it through, is critical. Without a clear vision, any negative comments along the way might dissuade you from taking the fund launch plunge.

At the risk of stating the obvious, the odds of successfully scaling up a fund are low. An investment fund as a business requires a lot of AUM to become self-funding. With this in mind, it maybe sounds foolhardy, even crazy, to have launched Bristlemoon Capital last year with no major investor commitments, instead bootstrapping the business from my own savings.

Launching Bristlemoon has been the most challenging, but also the most fulfilling thing I’ve ever done. In this post I will detail some lessons I’ve picked up in the first year since launch that may be relevant for those contemplating starting their own fund, or starting a business in general. Launching any sort of business can be a difficult process – it helps to have a good reason for starting a fund, and to know what you’re signing up for.

Why start a fund?

Many investment analysts, whether secretly or overtly, harbor ambitions to one day start their own fund. I was no exception. During an interview for a hedge fund role in 2015, I was asked “where do you see yourself in 10 years?” My response, somewhat naively and perhaps far too candidly, was “I want to be running my own fund”.

To best frame the rationale for starting your own fund, it’s instructive to think through when it makes sense to not start a fund.

There are broadly two reasons for staying at an existing fund as an analyst: 1) you are learning a lot from those around you; and/or 2) you are being paid well for your contributions.

If you are achieving both, and are given research autonomy by your portfolio manager, then this can make for a satisfying job. If you are achieving neither, then the impetus to leave and start your own fund will be higher.

While it makes sense to hone your craft at an established fund early in your career, the problem is that your learning curve as an investment analyst will inevitably begin to flatten. In other words, the rate at which you are accumulating new investing knowledge from those around you will naturally slow down over time. This dynamic is especially noticeable as you get more senior in your career, given you have already built up a base of expertise.

As the competence gap between yourself and your portfolio manager narrows, the potential for pushback on your portfolio manager’s thoughts and ideas increases. This pushback often serves as the fuel for wanting to start your own fund. It can become more difficult to maintain job satisfaction if your stock and business suggestions are being overruled by a portfolio manager you increasingly view as an equal.

Subscribe now

Another factor that feeds into the fund launch calculus is the way that a fund’s economics are split, with your superiors typically taking a larger slice of the pie.

Starting a fund may seem like a neat way to kill two birds with one stone, enabling you to: 1) capture full decision-making autonomy; and 2) benefit from a greater share of the fund’s economics.

Sure, these are possible benefits from starting your own fund. But you are giving something up for them; don’t forget that.

When you are working at a fund as an employee, you will only capture a mere sliver of the value you are generating. But in return, you enjoy a level of stability and safety. You will receive a salary, knowing precisely how much will be deposited in your bank account each month. Many of the strategic and operational decisions will be borne by somebody else, enabling you to spend the majority of your time researching stocks. You will likely receive mentorship and there will often be a clear, pre-defined path for career progression.

Launching a fund means abandoning these benefits that stem from being an employee, and not an owner. You will be out on the open seas and it’s incumbent upon you to navigate the route forward when managing your own fund.

You have to acknowledge and be prepared for everything that comes with the fund launch route. It is far more than just picking stocks and reaping the associated rewards. You will be running a business – the operational, compliance, regulatory, tax, and marketing burdens will need to be considered. You will be running a team as the fund scales, and you must be equipped to deal with a multitude of people issues. You are responsible for building your fund’s brand, letting the world know that you’re open for business and are worth investing in.

Much of your mental bandwidth will be consumed by the venture, so you must accept that simply shutting off is not really an option. There is no convenient “off switch” when you have been entrusted with managing the capital of others.

If all of this sounds like an acceptable trade-off, then it’s worth thinking through the various ways in which you can approach a fund launch.

The different ways to start a fund

The number of fund launch options available to you will heavily depend on your personal circumstances. The key considerations are as follows:

• Your experience and whether you have a marketable track record;

• The amount of personal capital you have saved;

• Your relationships with individuals and organizations that would be comfortable backing you; and

• Your risk appetite.

We can plot the different fund launch archetypes along a spectrum. Notably there’s often an inverse relationship between the level of risk taken in the fund launch approach and the level of fund economics you will capture (but not always).

Launching an internal fund within an established fund manager

This route is the least risky from a financial sense. You will likely be underwritten with a salary, and the fund manager bears the expenses of the fund. The trade off is that you will keep a much smaller share of the fees – often just some portion of the performance fee for the capital you have discretion over. While these arrangements vary, at best you might be able to capture half of the performance fee.

This is potentially still very lucrative, particularly if the fund has a good reputation and established distribution capabilities to scale up your internal fund.

Partnering with a capital partner

In this arrangement, you will ostensibly be running your own fund and will be able to build a brand. You will split the economics with the capital provider on the basis that they will do the heavy lifting on raising money for the fund. This can work very well. It can also be problematic to the extent that often these capital partners are representing multiple investment managers, with potential conflicts of interest to the extent there are other managers in the same asset class with a similar strategy that are also being marketed to allocators.

Furthermore, you won’t have control over the distribution personnel, which are interfacing with end clients and representing your brand. This is fine if the capital partner has professional and competent personnel. It can be an enormous distraction to the extent that they don’t, and any decision to separate from a capital partner could be a messy undertaking given the sharing of economics.

An alternative is to launch the fund independently, build up a track record, and then retain the option to partner with a capital partner if the fund’s AUM does not reach sufficient scale.

Subscribe now

Launching a fund independently

Within this option there are a variety of flavors. Launching out of a high-pedigree, name-brand fund with your own track record will translate into a very different AUM scaling path compared to launching from a no-name fund with no track record. While success in the latter scenario is not impossible, and there are cases of under-the-radar managers launching with $1-2 million of AUM with no pedigree and no track record, who then scale to hundreds of millions of dollars of AUM in a matter of years, it is a much harder path.

The trick is properly appreciating how the AUM ramp up might look for the type of fund launch you choose and then appropriately budgeting for that. A common reason emerging managers fail is that they run short of personal capital before they can get enough traction. In order for your fund to be able to scale, you need the financial runway to stay in the game long enough to prove yourself to investors.

It’s critical to be realistic about how long it takes to properly scale a fund. It is an almost certainty that it will take longer than what you think.

The advice I was given was to haircut your AUM launch expectations by 90%. This was precisely our experience – to the dollar, in fact. Virtually every fund manager we have spoken to launched with significantly less capital than what they were expecting. That’s just how it goes. But it does make the journey harder, and it reinforces the need to be very sure that you do indeed want to start a fund.

The personal side to starting a fund

Much of the fund launch decision will hinge on your personal circumstances. Here are some questions you should consider:

• Do you have the personal capital to give yourself sufficient runway to build a track record and raise capital? At this point I have gone more than two years without a salary. It is an uncomfortable adjustment, but a necessary part of the journey of building something great.

• Are you willing to drastically lower your personal cash burn in order to extend this runway? This means significantly curtailing discretionary expenditures. For some, the associated hit to one’s lifestyle might be wholly unpalatable, particularly if a spouse is not okay with making the same sacrifice.

• Do you have a spouse, and are they onboard with you taking on this monumental challenge? Absent this support, the demands of running the fund, in terms of time, energy, and money, could put strain on your relationship.

• Do you have relationships with wealthy individuals who will give you money to manage? The reality is that no matter how good of an investor you are, if you don’t have a network of people that trust you enough to invest with you, then you will struggle to convert your investing acumen into a viable business. There is a sales component to building a funds management business that can’t be ignored. Be honest with yourself around your ability to network and build relationships with wealthy individuals. Any shortcomings here will make the scaling process much more difficult.

Launching a fund too early can be detrimental. If you are too young, allocators (most of which are likely to be older than you) will not take you seriously.

However, launching a fund too late can also be problematic. One’s personal cash burn tends to increase over time due to mortgages, additional dependents such as a spouse and children, schooling fees, etc.

There is likely a sweet spot for launching a fund where you have sufficient experience, but do not have a personal life that has become too financially encumbered. You will need to factor all of this into the level of personal capital you have when starting the fund.

Launching a fund is launching a business

When you launch a fund, it can be difficult to retain your investing purism. For example, if you were managing only your personal capital, you might invest with a certain desired approach. However, when running a fund your capital is co-mingled with the capital of external investors. Their expectations and goals might not perfectly align with yours. In fact, it is almost certain that their expectations and goals won’t perfectly align with yours. However, it is the fees from these clients that will underpin your business, so their expectations and goals must be duly reflected in the fund’s strategy.

At the end of the day, a funds management business isn't so much managing a portfolio as it is creating an investment product that is saleable and well-received by the target market. This requires product market fit. So, the investment strategy must be carefully designed so as to be attractive to potential investors. It is sometimes a delicate balance in reconciling how you might want to manage the portfolio, and how capital allocators want the portfolio managed.

If you are too much of a purist, you will trade AUM scalability for job satisfaction. If you are too commercial and design the product so that it has maximum appeal to the target market, you risk running a strategy that is poorly-aligned with how you want to manage capital, and in the worst case, is impaired in its ability to generate good returns.

Subscribe now

Some of the fund parameters that are important to consider are when structuring an investment strategy are as follows:

• Gross and net exposure ranges and limits for a long/short fund

• Maximum cash weighting for a long only fund

• Maximum position size

• Sector concentration limits

• Sector or geographic specialization

• Quality bias (or not)

• Portfolio turnover

• Level of volatility

• Fee structure

Be very careful with how you craft your fund’s strategy. Often there is a push and pull between retaining enough flexibility in your investment mandate to generate good investment returns and having a narrow enough focus to carve out a niche that makes you stand out to allocators.

Many allocators will already have investments in a variety of funds – if yours doesn’t stand out, it might be hard to get a look in. This is where a fund with a more narrowly defined mandate can help open doors with allocators. Specialization in a niche with a compelling argument for why that niche is ripe with investment opportunities will resonate more with some capital providers than simply saying you’re a good stock picker and will find great opportunities wherever they present.

For example, a fund that only invests in SMID cap technology stocks in Southeast Asia will probably generate more initial interest from allocators than a fund that has an open investment mandate. The lack of constraints in the latter fund makes it harder for allocators to mentally pigeonhole that strategy. This simply makes it more difficult for an allocator to determine how that fund will blend with the remainder of their manager investments.

The other consideration with a niche strategy is that there will be periods where the market will be favorable for that strategy. Performance will be good, making it easier to get traction with capital allocators. However, there will also be periods when the market environment for that targeted niche becomes unattractive, perhaps even hostile. It then becomes increasingly hard both to make money and attract investor interest.

By committing to a more narrowly focused strategy, particularly if investors are allocating to that strategy to get a certain exposure, you are then locked into that approach. Launching a fund with a narrow mandate means you are basically rolling the dice, hoping for a market environment that is kind to the investing approach you’ve adopted. This is clearly problematic to the extent that you launch into an environment that is punishing stocks in your targeted domain. This can create genuine business risk.

The timing of when you start a fund matters, an incredible amount actually. You could be a savvy tech investor, but if you had launched a U.S. tech fund at the beginning of 2022, there were few places to hide in that sector from the carnage that ensued over the course of that year.

The narrower the approach, the more you will need luck on your side in the early years of the fund to help you build a solid track record. Absent this, and if bad luck strikes early in the fund’s life, then it is likely to elongate the AUM scaling process as it will take time to get the fund’s performance back on track.

Launching solo or with a business partner/team

I launched Bristlemoon as a solo operator. My thinking was to get the business up-and-running and build a track record with my own capital in a structure that could later onboard external capital. It was a leap of faith, and frankly I didn’t know for sure whether other investors would enter the fund at such a nascent stage.

Thankfully I was able to secure capital from some early backers, and our AUM has grown healthily since then, allowing me to bring on my close friend and former colleague, Daniel Wu, as a business partner.

Launching a fund as a solo operator has its benefits. You have total autonomy. You also get to intimately understand the various aspects of your own business (versus outsourcing the various functions to a team). The problem, however, is it is more difficult to scale as a solo operator, and unfettered autonomy doesn’t always translate into the best decisions.

Numerous investors we’ve spoken to also expressed more comfort in having more than one person running the fund.

Bringing Daniel into Bristlemoon has been a gamechanger for the business. We are able to share the workload and he’s served as a great sounding board to pressure test ideas. All of this translates into greater rigor in the decision making process.

Managing a portfolio with someone else requires a huge amount of trust and knowing how to calibrate the other person’s judgment. For example, Daniel saying “I think this stock is good” is probably the equivalent of another investor banging the table and saying “this is a generational buying opportunity where we need to back up the truck”.

People express their views and opinions differently; having familiarity with how that person communicates their stock calls is an important part of a solid investment partnership. Often there’s no substitute to achieve this level of understanding outside of having countless detailed investing discussions and knowing that person for a long period of time to understand the quirks of their personality.

Subscribe now

Before starting Bristlemoon I spoke to many who'd started their own investment firms, but had later decided to wind down their funds. I found the stories of why some of those firms had subsequently shuttered to be illuminating. Many cited an irreconcilable dispute with their co-founder that led to them parting ways. An investment partnership is akin to a marriage; choose your partner wisely.

Prior to Bristlemoon, I had known Daniel for 14 years, and we had been talking about launching a fund together for a decade. We were housemates in Sydney and New York, and worked together for years. Becoming business partners was a pretty straightforward decision for both of us.

But even with a shared history, it is still vital to figure out roles and responsibilities, as well as economics before you set out. Figure out exactly who will do what, and formalize a shareholders agreement before you set out on the journey. The absolute last thing you want to be doing is figuring out how to share the pie mid-journey. You particularly don’t want to be negotiating the economic split on an ongoing basis. That’s a recipe for contention and disagreement. Just agree on how the economics will be divided between yourselves and then stick to it.

Markers of legitimacy / building credibility

Step into the shoes of a prospective client who’s contemplating investing money in your fund. Even if they’ve met you multiple times, there’s not a lot from the outside they have to go off to get comfort – after all, most funds are a bit of a black box. Maybe they can see your website and research you’ve written. Regardless, there’s often a trust and credibility gap that typically needs to be bridged.

For this reason, many prospective investors rely on heuristics, typically seeking out markers of legitimacy. This manifests in prospective investors asking us similar questions:

• How big is the team?

• How much AUM do you have? (Rarely do investors want to be the first to invest in a new fund. AUM has a high signaling function and more of it gives investors confidence that others have confidence in you)

• Where’s your office? (Working from home, while perfectly fine, is far less comforting to some investors than having a city address)

• Where do you live? (I get this question all the time, believe it or not. Prospective investors, rightly or wrongly, draw comfort if you live in a good suburb rather than a rough part of town. It shouldn’t matter one iota to your ability to generate investment outperformance, but it hints at the mental shortcuts prospective investors take to try to assess whether you’re credible.).

To the extent that you can tick these boxes, you give a prospective investor fewer reasons to say no to investing in your fund. In fact, every strategic business decision you make should be in service of reducing the number of reasons a prospective investor might be inclined to say no. Many of these frictions reduce over time. Others will depend on structuring decisions that are made at the outset of the fund launch journey.

No one cares about your fund more than you

The funds management space is crowded and competitive, with limited vectors across which you can truly differentiate yourself. There is no shortage of fund managers saying the same things: “We take a long-term view, buying quality stocks that are mispriced”. This certainly makes it more difficult to stand out, and without something to make you stand out, you will likely be reduced to a commodity product in the eyes of an allocator.

The truth is no one will care about your fund more than you do. The wealthy individuals you will be pitching to have their own priorities, and investing in your fund is probably not super high on their list. They are probably not thinking about how critical it is to promptly invest in your fund when they’re out vacationing in Saint-Tropez. Now, this doesn’t mean that they’re not interested in your fund whatsoever, but it does mean that there’s typically a healthy dose of persistence needed to chase up prospective investors to build those relationships. It usually requires a balance between being politely persistent, but not so persistent that you become a pest.

I hope you enjoyed these thoughts and reflections. It’s an honor to get to share these learnings and I hope they’re of use for anyone who has contemplated starting their own fund. If these thoughts resonated with you, or if you have been on a similar journey, I’d love to hear from you and can be reached at info@bristlemoon.com.

Subscribe now

Disclaimer / Disclosures

The information contained in this article is not investment advice and is intended only for wholesale investors. All posts by Bristlemoon Capital are for informational purposes only. This article has been prepared without taking into account your particular circumstances, nor your investment objectives and needs. This article does not constitute personal investment advice and you should not rely on it as such. This document does not contain all of the information that may be required to evaluate an investment in any of the securities featured in the document. We recommend that you obtain independent financial advice before you make investment decisions.

Forward-looking statements are based on current information available to the author, expectations, estimates, projections and assumptions as to future matters. Forward-looking statements are subject to risks, uncertainties and other known and unknown factors and variables, which may affect the accuracy of any forward-looking statement. No guarantee is made in relation to future performance, results or other events.

We make no representation and give no warranties regarding the accuracy, reliability, completeness or suitability of the information contained in this document. To the maximum extent permitted by law, we do not have any liability for any loss or damage suffered or incurred by any person in connection with this document.

Bristlemoon Capital Pty Ltd (ABN: 22 668 652 926) is an Australian Financial Services Licensee (AFSL Number: 552045).

George Hadjia and Daniel Wu are associated with Bristlemoon Capital Pty Ltd. Bristlemoon Capital may invest in securities featured in this newsletter from time to time.

Subscribe now

Free subscribers will only receive a partial preview of our reports. The remainder of our reports, which contain the deeper analysis, are reserved for paid subscribers. Consider becoming a paid subscriber for full access to our reports.

Bristlemoon readers can also enjoy a free trial of the Tegus expert call library via this link.

Australian wholesale investors looking to invest in the Bristlemoon Global Fund can do so via this link.

Introduction

Investors love businesses with pricing power. The ability to raise prices without losing customers is one of the clearest indications of a business moat. Fair Isaac Corporation (Nasdaq: FICO) is one of the more notable examples of a public company exerting its pricing power in recent times. The company dramatically raised the prices of its eponymous credit score in a very short span of time.

For some context, the company didn’t raise its credit score royalty rates in real terms for nearly 30 years. But following an intense process of contract renegotiation that permitted FICO to begin raising prices, which we’ll explore, the price of a FICO Score went from around 50-60 cents in 2022 to $4.95 per score in 2025. In other words, FICO raised its prices by 800% in just three years! Clearly, FICO has been making up for lost time on the pricing front.

The company argues that there remains a demonstrable price-to-value gap with its FICO Score that it is seeking to narrow (i.e., via further price increases). Regulators, however, have taken notice, and have sought to break FICO’s monopoly by making changes that attempt to foster competition in the scores market. The key question is the extent to which these changes will erode FICO’s monopoly as well as the company’s ability to continue its heady price increases.

This has caused investors to fret over whether FICO still has pricing power. This tends to be problematic when a company’s stock trades at over 60x earnings, as FICO did earlier this year, and where the future growth is almost wholly reliant on price increases. Consequently, the regulatory rumblings have brought about an enormous destruction of value in FICO’s equity, with the share price declining by 45% from the all-time high of c.$2,380 in November 2024, to c.$1,300 in August 2025.

Source: Bloomberg; Bristlemoon Capital

This report is intended to serve as a primer so that you can understand how FICO’s business works, from the nuances of its Scores business, its complicated relationship with the credit bureaus, as well as some financial analysis to better understand the economics of FICO’s growth story. We also unpack the key debates surrounding the stock, and analyze the potential ways in which FICO’s business might develop in the future.

Key Takeaways

• FICO is not only a monopolist in the credit score market, but its credit score is used as the industry standard for consumer credit risk in the U.S. The company has aggressively increased prices, raising them by 800% over the last three years. This growth has led to 95% incremental operating margins for the Scores business, and operating margins that are near 90%.

• The FICO Score is incredibly entrenched in the US mortgage ecosystem, and it is infused into multiple systems, workflows and regulatory requirements across the point of loan origination, as well as the secondary market securitization value chain. The FICO Score very much is not just a number referenced by lenders, but is integral to the plumbing of the U.S. mortgage system in a way that makes the score very sticky.

• FICO is used at different stages of a mortgage application, and we estimate that there are roughly 10-12 credit pulls in a typical mortgage application, resulting in $49.50 to $59.40 of royalties for FICO. FICO’s royalties comprise a small part of total mortgage closing costs – in the vicinity of 90 basis points for a typical mortgage loan. FICO, however, has attracted regulatory scrutiny from the FHFA, despite the fact that the bulk of incremental credit report inflation can be attributed to the credit bureaus and other third parties layering on fees and marking up the cost of the FICO Score when a tri-merge credit report is generated.

• FICO has a somewhat complicated relationship with the credit bureaus – one that we would categorize as “frenemies”. While the three major credit bureaus compete with FICO via their VantageScore JV, the bureaus also make more revenue from marking up the FICO Score than what FICO makes from its scores. Both FICO and the credit bureaus are well-incentivized for FICO to continue its aggressive pricing campaign, and we believe there is an exceedingly low chance that any traction from VantageScore would result in a race to the bottom on prices.

• The securitization market, in particular the standardization of the to-be-announced market where more than 90% of agency MBS are traded, acts as a gating factor on VantageScore getting traction. Heterogenous loan pools – that is, those that use both FICO and VantageScore – are harder to price and we believe that MBS investors will demand a discount for loan pools scored with VantageScore. There is also the risk of loan putbacks that create a disincentive for lenders to deliver loans to the GSEs with an alternate score such as VantageScore, in the case that those loans default and the GSEs force the lender to buyback the loan at par.

• We estimate that for every basis point of VantageScore risk premium over FICO, a borrower will end up paying over $900 more over the life of a 30 year, $400,000 mortgage underwritten using VantageScore. We estimate that the value of the savings over the term of a mortgage are potentially 100x greater than the $50 in royalties FICO might receive in a dual borrower mortgage origination.

  Subscribe now

Table of Contents

• Business Overview

  • Software – a brief mention

  • Scores – the key value driver for FICO

• The FICO Score is the Industry Standard

  • Creating an industry standard

  • Moving away from an industry standard

  • Are there examples of an industry standard being displaced?

• Business History

• FICO’s Special Price Increases

• The Real Cost of Price Increases

  • The CRAs have contributed more to credit report cost inflation than FICO

• Regulatory Woes

• Increased Competition in the Credit Score Market?

  • FICO 10 T vs VantageScore 4.0

  • VantageScore is more inclusive?

• FICO’s Relationship with the Credit Bureaus

  • The quirks of the VantageScore JV

  • Do the credit bureaus make more from a FICO Score than from VantageScore?

• The GSEs and the Securitization Market

  • FICO Score the industry standard for mortgage securitization

  • The To-Be-Announced (TBA) Market

  • Risk of loan putbacks

• Why are lenders complaining about the FICO Score if the cost is passed through to consumers?

• What is the Value of the FICO Score?

• Concluding Comments and Valuation

Business Overview

FICO’s business is comprised of two segments:

1. Scores (FY24: 54% of revenues and 75% of operating income)

2. Software (FY24: 46% of revenues and 25% of operating income)

The Scores segment is the key value driver for FICO as a business. Not only does it comprise three-quarters of the company’s profits (it is incredibly high margin, with 88% segment operating margins in FY24), but it also contributes the bulk of incremental profit dollars, given that the Scores segment is growing relatively faster and at much higher incremental margins than the Software business (Scores revenues have come in at 95% incremental margins on average over the last six years).

Software – a brief mention

FICO’s Software business covers six categories: fraud detection, customer management, originations, financial crimes compliance, customer engagement and marketing. The company typically sells its software as multi-year subscriptions, with payments based on usage metrics such as the number of accounts, transactions, or decisioning use cases.

FICO’s Falcon Fraud Manager dominates the fraud detection market. It can determine whether a transaction is fraudulent or not in 15 milliseconds, and it protects two-thirds of credit card transactions worldwide. FICO TRIAD Customer Manager, which is a decision management software suite used by lenders to optimize and automate credit decisions, is similarly dominant in its market segment. We could spend pages talking about FICO’s Software business but it’s the Scores segment that is going to drive the value of the stock, so we will spend the remainder of the report on Scores. We would just lastly note that FICO management have expressed openness to one day selling the Software business, and the Software and Scores segments operate independently.

Scores – the key value driver for FICO

FICO is an applied analytics business, with the company best known for its flagship FICO Score product. The FICO Score – a three-digit number ranging between 300 and 850 – is the standard measure of consumer credit risk in the U.S. This metric is used by over 90% of the top U.S. lenders when making consumer lending decisions across mortgages, auto loans, credit cards, and other areas[1]. Typically, the higher your FICO Score, the lower your credit risk, and thus the greater the likelihood that lenders will write you a loan.

The FICO Score is used to rank-order individuals against a certain variable, such as the likelihood of making a credit payment on time. It utilizes five major factors in determining the score, weighting each of those factors based on predictive importance:

1. Payment history (35%)

2. Amounts owed (30%)

3. Length of credit history (15%)

4. New credit (10%)

5. Credit mix (10%)

FICO uses predictive modelling to uncover patterns in historical data, which enables classifications about future credit outcomes to be represented by a single FICO Score number (although notably there are over 50 variations of FICO Score for various markets, including FICO periodically releasing new generations of FICO Scores which refine how the algorithm interprets credit bureau data, or incorporate the ability to process additional datasets).

FICO is really a licensing business. The company has a set of intellectual property that is used in the underwriting process to assess credit risk. FICO licenses out this intellectual property to the three U.S. consumer reporting agencies (CRAs) – Experian, TransUnion, and Equifax – and in return receives royalties from the CRAs. It is worth noting that the three major credit bureaus mentioned above accounted for 45% of FICO’s revenues in FY24, and 84% of the company’s Scores revenue.

Subscribe now

Below is a visual depiction of the process to generate a credit score:

Source: Jefferies

FICO is not actually the entity that produces the FICO Score. It is the CRAs that apply FICO’s proprietary analytic algorithms to their respective credit data in order to generate a FICO Score. The CRAs then sell the resulting FICO Score to lenders and other end-users (typically via resellers). In this sense, the credit bureaus act as distributors, and are an intermediary between FICO and the ultimate user of the scores product. Importantly, FICO doesn’t house or handle any of this sensitive consumer credit data.

The end users of FICO’s products are mainly financial institutions, with the company’s products being utilized by three-quarters of the 100 largest financial institutions in the U.S. It is then perhaps unsurprising that FICO’s largest market segment is financial services, comprising 92% of FY24 revenue.

The Scores business is further broken down into business-to-business (B2B) Scores (77% of total Scores revenue) and business-to-consumer Scores (23% of Scores revenue). Lastly, FICO is primarily concentrated in the Americas region, with this geography representing 84% of FY24 revenue.

The FICO Score is the Industry Standard

It is worth spending a moment to discuss what industry standards are, why it’s valuable to be an industry standard, and how FICO’s deep integration into the U.S. lending markets has made it the industry standard for assessing and monitoring credit risk.

An industry standard refers to a technical specification, practice, or product that is widely used across an industry as a baseline for compatibility, quality, and interoperability. Essentially, an industry standard is what joins different parties together in a way that allows them to unlock value.

In the context of FICO, credit data on a single borrower is not particularly useful. Data in isolation is simply far less valuable than datasets that have been joined together. If the credit data of millions of borrowers can be joined via a singular representation of credit risk, then that data becomes exponentially more valuable. This is what the FICO Score represents: a “join key” that links these millions of credit datasets in a way that unlocks value in the form of not only more predictive loan underwriting, but a common language for participants in the lending industry to communicate credit risk.

In a great article on Data-As-A-Service, Auren Hoffman made the following insightful observations[2]:

“One of the big ways that data becomes useful is when it is tied to other data. The more data can be joined, the more useful it is. The reason for this is simple: data is only as useful as the questions it can help answer. Joining, linking, and graphing datasets together allows one to ask more and different kinds of questions…the more join keys (and joined data sets) you can find, the more valuable those data become…As you keep joining data, the number of questions you can ask grows exponentially.”

There are two things to bear in mind for industry standards: 1) it is incredibly hard to make something an industry standard, and it often requires a unique set of goldilocks circumstances; and 2) once a market builds around an industry standard, it is enormously challenging to move off that industry standard.

Creating an industry standard

An industry standard can arise from voluntary consensus (e.g., agreement on a standard by an industry consortium), widespread market adoption (i.e., one company gets so much market share that it becomes the de facto standard), or it can be mandated by regulators.

In the case of FICO, it became the de facto standard due to widespread lender adoption. U.S. lending institutions adopted the FICO Score for two key reasons: 1) it allowed them to write better loans, given the FICO Score did a better job of predicting default risk than those banks’ internal methods at the time; and 2) a number of regulatory changes, such as the Equal Credit Opportunity Act in 1974, supercharged the adoption of the FICO Score, given that algorithmic credit scoring was an ideal way to assuage regulators that a given lending decision was not discriminatory.

The adoption of the FICO Score as the industry standard came from a set of irreplicable circumstances – namely being the first mover in creating an algorithmic scoring system at a period in time where regulatory changes made it highly convenient for lenders to adopt said scoring system.

Subscribe now

Moving away from an industry standard

Shifting to a new industry standard is very difficult. It is a bit of a chicken-and-egg problem. Industry participants might be interested in utilizing a new standard but will be reluctant to adopt that new standard until others are already using it. So, this means that either a very large industry participant, or a group of participants need to switch to the new standard en masse to drag along the other players. In an industry as fragmented as that of lending in the U.S., we believe that there is an exceedingly low likelihood of the industry shifting away from the FICO Score. Industry standards, due to the immense switching costs when moving to a different system, often have significant longevity.

A good example of an industry standard is the QWERTY keyboard. It is a subpar arrangement of keys in terms of typing speed and ergonomics, yet it’s hard to imagine switching from this standard, given it’s the keyboard arrangement virtually everybody has become accustomed to, as well as how integrated it is into computer hardware. This is the power of industry standards: once they’re entrenched, they are really difficult to move away from. Regardless of any incremental ergonomic benefits from a redesigned keyboard, it would be a tall order to convince consumers to switch to a different keyboard design (e.g., Dvorak or Colemak) when virtually no one else is using those alternate designs. And because of these frictions consumers face when switching, keyboard manufacturers have very little incentive to push for change (particularly given they do not bear the cost of a slightly less ergonomic keyboard).

It's possible to map the value of an industry standard using Metcalfe’s Law. Metcalfe’s law states that the value of a network is proportional to the square of the number of connected users. In other words, the more industry participants that use the FICO Score as the industry standard, the more useful that standard becomes. This is particularly so when you have multiple types of industry participants all rely on the industry standard – in the case of the FICO Score, it is relied upon by borrowers, lenders, investors, and regulators.

The FICO Score is infused into multiple systems, workflows, and regulatory requirements across the point of loan origination, as well as the secondary market securitization value chain. The FICO Score very much is not just a number referenced by lenders, but is integral to the plumbing of the U.S. mortgage system in a way that makes the score very sticky. Below are a few ways in which FICO is integrated into the mortgage ecosystem:

• Loan Origination Systems (LOS) – LOSs such as ICE’s Encompass and Black Knight’s Empower integrate with resellers to pull a tri-merge credit report (which includes a separate FICO Score from each of the three credit bureaus). The LOS then stores the middle FICO Score as the representative score for the borrower, and uses this score in loan eligibility calculations.

• Automated Underwriting Systems (AUS) – Fannie Mae’s Desktop Underwriter (DU) and Freddie Mac’s Loan Product Advisor (LPA) require a FICO Score as an input. Without a FICO Score it is currently impossible to get GSE conforming loan approval.

• Loan pricing engines – FICO is a key input into Loan Level Pricing Adjustments (LLPAs), with secondary desks explicitly bucketing pricing by FICO bands (e.g., 720-739, or 700-719). Fannie Mae and Freddie Mac publish LLPA grids (these are essentially rate sheets that determine basis points price hits (or credits) to the loan amount based on risk factors such as the FICO score, loan-to-value ratio, occupancy type, etc.) but it is the secondary desks within lenders that operationalize these LLPAs (we will dig into this and the role of GSEs later in the report). In this sense, the LLPA grids, which are based on the FICO Score, allow lenders during the origination phase of a mortgage to efficiently establish and adjust loan pricing based on the individual risk profile of the borrower.

• Servicing and loan performance models – Mortgage servicers monitor borrower credit health post-origination by using FICO Scores. Both GSEs and mortgage servicers will rely on periodic FICO credit pulls to make updated assessments of delinquency and prepayment risk (which feeds into mortgage servicing rights (MSR) valuations and loss mitigation waterfalls).

• Secondary market and MBS structuring – when GSEs package loans into mortgage backed securities (MBS), pool-level disclosures include the weighted average credit score. For decades that score has always been the FICO Score, and this continues to presently be the case. Credit risk transfer securities issued by Fannie and Freddie (i.e., CAS and STACR, respectively) explicitly reference FICO as the risk metric.

• Regulatory capital and supervisory models – FHFA stress testing and GSE capital frameworks (e.g., ERCF) use loan-level FICO distributions as key drivers of expected default and loss models. Furthermore, banks holding MBS also bucket their capital requirements based on FICO-driven credit quality segments.

• Private Mortgage Insurance (PMI) – the FICO Score is an integral part of PMI pricing models. It is what underpins insurance eligibility decisions as well as setting insurance premiums.

• Other systems – there are a range of additional mortgage industry players that are reliant on FICO. For example, the Mortgage Industry Standards Maintenance Organization (MISMO), which is responsible for developing standards for exchanging information in the U.S. mortgage industry, has a FICO-centric data schema. Quality control and loan audit systems such as QC Ally use FICO thresholds for defect categorization. Servicing transfer data tapes (i.e., the loan-level dataset that the current servicer of a mortgage portfolio delivers to the new servicer when MSRs are sold or transferred) carry FICO fields.

Subscribe now

If you can appreciate the challenge of moving away from the QWERTY keyboard in a scenario that involves only consumers and keyboard manufacturers, imagine trying to find ways for all of these disparate parties in the U.S. mortgage industry to make the changes necessary to support a second scoring method such as VantageScore. The problem becomes even more difficult when you consider that many of these industry participants have little incentive to make the investments necessary to support a second credit score beyond FICO.

The following quote from the Founding President and former CEO of VantageScore (via Tegus) is revealing:

“My first call on Fannie was in October of 2006. They had already tested us and said, "Will you outperform FICO on predictiveness?" I thought, "Well, this is going to be a slam dunk." Here we are 19 years later. I just think there's so much inertia.”

In other words, when VantageScore was introduced in 2006, it was believed that it would be able to encroach on FICO’s stronghold in the mortgage market. Almost two decades later and VantageScore’s impact has been negligible.

The truth is the mortgage industry is just very slow to change. The following quote from a former VP at Guaranteed Rate highlights this dynamic:

“We've been waiting for 15 years for them to allow electronic signatures on mortgage closing documents. That technology has been around forever and nobody allows it in. If they're not going to allow something as simple and verifiable as an electronic signature, what makes you think they're going to allow something as potentially life-changing or industry-changing as a VantageScore to have any role in the risk management of these assets?”

The majority of financial institutions that utilize the FICO score are relatively sophisticated and use the FICO Score in conjunction with their own risk models. Introducing an alternative credit scoring model, such as VantageScore, or modifying the current FICO scoring model would require lenders to undertake a comprehensive recalibration of their internal risk models and protracted trial period. FICO management have noted that it is close to impossible to accurately and reliably map a Classic FICO Score to a VantageScore 4.0, given that the relationship is nonlinear and changes based on the economic environment. The fact that VantageScore has not been stress-tested in a real-life economic downturn is a fairly steep impediment to lender adoption.

Financial institutions are typically reluctant to change scoring methods, and we can best see this reflected in the slow uptake of the new generations of FICO Score.

Some lenders upgrade quickly to new versions of the FICO Score, while others take much longer; observe below how FICO Scores 2, 4, and 5 are still being used for the mortgage lending market, despite FICO Score 10 and FICO Score 10 T having been released. If it was easy to switch scores, then the mortgage industry would have likely shifted to using newer generations of the FICO Score that incorporate trended data. The truth is it’s a heavy lift to change scores.

Source: MyFICO

It takes years to drive adoption of these updated and more predictive generations of FICO Score, hinting at an industry that is reluctant to change.

“So going from FICO 8 to FICO 9, it took four years before 50% of the market was on FICO 9. And I think you'll see the same kind of thing. It'll be four years again before FICO 10 is half the market. So it's a very slow adoption rate.” – Will Lansing (CEO), MS TMT Conference 2025

Back in October 2022, the Federal Housing Finance Agency (FHFA) announced that a dual-score mandate would be implemented, whereby the GSEs would be required to use both a FICO Score and VantageScore on conforming mortgages. The FHFA then backed away from this proposed move. There is signal in the fact that in July 2025, after years of consultation and trying to get the industry ready for this shift, the FHFA announced that it was moving to lender’s choice, whereby lenders can choose to use either FICO or VantageScore (i.e., a somewhat watered down policy compared to the original dual-score mandate that was proposed).

A dual-score approach would simply double credit scoring costs and introduce immense complexity across the origination and securitization value chains. A Goldman Sachs research note highlighted that the FHFA had previously estimated that it would cost the mortgage industry $600 million or more to move to a dual-score mandate. These heavy implementation costs, and the fact that the absolute cost of a FICO Score is so low relative to mortgage closing costs, and that this cost is passed on to consumers, leads us to believe that there’s a low incentive for lenders to make the switch to VantageScore.

Inertia is a fairly safe bet when it comes to making changes that risk disrupting a high-value ecosystem that depends on an industry standard product like the FICO Score. There is $13 trillion worth of outstanding mortgages in the U.S.[3] – there is an extremely high bar for implementing changes that risk dysfunction in such a financially, socially, and politically important market. For this reason, we are of the view that any changes to the status quo of FICO as the credit score used in the U.S. mortgage market will be slow and marginal.

Subscribe now

Are there examples of an industry standard being displaced?

If industry standards are so defensible, there’s merit in looking at historical examples of when those standards get displaced. We view there as being two primary ways in which industry participants will pivot away from a standard: 1) technological change leads to a new industry standard that is significantly more performant; and 2) a new standard is mandated by a regulatory body.

The inexorable changes stemming from technological innovation have led to numerous industry standards being unseated. Email usurped fax. The consumption of media content moved from VHS to DVD to Blu-ray, before changing form factor entirely with the advent of streaming services like Netflix. In the case of FICO, we are not at all worried about technological change displacing the FICO Score as the industry standard.

If we look at newer generations of FICO Score, there is typically only a minor incremental performance uplift compared to the prior generation of FICO Score. In other words, the predictiveness of scores has already begun to asymptote, and we don’t foresee a new credit scoring method materializing that can predict loan defaults at a significantly higher rate of accuracy than the FICO Score. Any improvements are likely to be incremental and minor. This militates against a step change improvement that is capable of catalyzing a broader industry shift towards a new standard of assessing creditworthiness.

We would also note that standards are most easily adopted when they are free. The unique history of FICO, whereby it kept its Scores prices both fixed and at a very low level for almost three decades, created the perfect conditions for adoption of the FICO Score as the industry standard. We deem it highly unlikely that a competitor score such as VantageScore will give away its score for free for years in order to drive adoption, particularly when we consider the unique financial incentives of the VantageScore JV (which we dissect later in the report).

Finally, as it relates to the potential for regulatory change to cause a displacement of the FICO Score as the industry standard, we believe that the disruption to the all-important mortgage industry would be far too great to make these changes.

Business History

The history of FICO is fascinating, and one that is entwined with the history of lending in the U.S. It is certainly worth exploring the company’s past in some detail to understand the significance of FICO’s contribution to improving underwriting standards and expanding access to credit in the U.S. Only then can we truly appreciate the impact that the FICO Score has had on the U.S. lending landscape and the value the FICO Score provides.

The company was founded in 1956 by engineer Bill Fair and mathematician Earl Isaac. The two had met at the Stanford Research Institute while they were working as operations researchers for the U.S. military. Fair and Isaac saw an opportunity to apply their operations research methods to the private sector in a for profit enterprise. They both put in $400 as an initial investment and set out to start a consulting business that aimed to use data to improve business decisions. The duo eventually settled on consumer lending as the domain in which they would apply their analytical efforts.

Fair and Isaac sent letters out to the 50 major U.S. lenders, requesting a meeting to explain their approach of employing statistical techniques to measure credit risk. They received one reply. A lender called American Investment Company (AIC), the nation’s fourth largest consumer finance company at the time, agreed to let Fair and Isaac analyze the company’s credit files. The resultant credit scoring system, after analyzing 13,000 “good” accounts (those devoid of significant delinquencies) and 1,000 “bad” borrower profiles, purportedly enabled AIC to reduce its bad debt charge-offs by 25% with only a 3% reduction in total loan volumes[4]. The data used was rather rudimentary: how long borrowers had lived at their current residence, whether they had a telephone, and other questions such as age and marital status. Regardless, AIC served as a proof point of the systematization of credit risk analysis which led to other lenders taking interest.

During this initial phase of the Fair Isaac Corporation, the company was a consultancy that worked on bespoke credit analysis projects for lenders, a far cry from the intellectual property licensing business FICO is today. What is so remarkable about the FICO story is that the company managed to bring statistical rigor to consumer credit analysis before the digitization of credit data.

A dissertation on the history of the Fair Isaac scorecard by Martha Ann Poon made the following comments regarding the laborious process by which the company originally generated its credit scorecards[5]:

“In the late 1950s the ability to calculate involved executing a series of physically demanding activities. Credit scoring required a steady traffic of materials back and forth, between Fair Isaac, a second party firm with access to computers and engineering skill, and storage rooms of credit operations spread out across the country. To change credit screening practices on the ground, the company had to command an intricate process of moving people, paper and things around.”

In the late 1950s and 1960s, Fair Isaac Corporation pioneered a new industry by doing the incredibly labor-intensive, manual process to generate credit scorecards. FICO was essentially turning stacks of paper (i.e., consumer credit data), using physical labor, into an easy-to-use cardboard printed table that mapped out the odds of defaulting (although the credit scoring process became increasingly automated in the ensuing decades due to improvements in computing technology). Each borrower attribute (e.g., age, time on job, delinquencies, etc.) was assigned a point value which was then tallied. This numeric score correlated with the odds of repayment, and each bank or finance company could raise or lower its own cutoff score to reflect their risk appetite and willingness to grant credit. The beauty of this new system, devised by Fair Isaac Corporation, was that it provided a standardized approach to making credit decisions that was consistent and scalable, much unlike the subjective judgment of loan officers that often fell victim to biases (cough… prejudices).

Subscribe now

If we trace the history of lending in the U.S., much of it was done based on a character assessment and personal relationships. Obviously, it’s difficult to scale this approach, and the proliferation of consumer credit with the advent of the credit card demanded a more rigorous, consistent, and fair approach. FICO’s credit scorecards came at the right time and place, particularly when we consider some of the later regulatory changes to fair lending practices.

The 1970s were a period of immense social change in the U.S., and this was reflected in a number of hearings conducted by Congress during that era to create equal credit opportunity as a civil right[6]. Legislation was subsequently introduced that banned creditors from basing lending decisions on factors such as race, religion, sex, marital status, or age, promoting the idea that credit should be granted based on merit rather than an individual belonging to some group which is beyond their control. While the intent of this regulation was noble, it posed difficulties for lenders. The Equal Credit Opportunity Act (ECOA) which was passed in 1974 required lenders to provide reasons for why a borrower was turned down for a loan. Over time, the objectivity of a credit scoring method such as that offered by FICO provided the necessary cover for lenders to point to their compliance with the definition of fairness as espoused by the ECOA. As is often the case with regulation spurring change, these new fair lending rules accelerated lender adoption of the credit scoring systems that FICO offered.

Back in the 1800s, the bulk of credit activity in the U.S. was between lenders and commercial enterprises (early forms of credit scoring can be traced back to the 1830s when local credit bureaus helped firms decide whether to extend trade credit[7]). The corollary of this is that most Americans paid cash for goods; borrowing was deemed risky and irresponsible. It wasn’t until the 1900s that the advent of the motor vehicle, which was a higher sticker price item relative to average wages, necessitated some inventive forms of financing to drive sales. General Motors created the General Motors Acceptance Corporation (GMAC), making car ownership possible by popularizing the concept of the installment plan (as a fun fact, GMAC was the precursor to Ally Financial).

Retailers similarly adopted these installment plans. However, in the early 1960s, retailers began transitioning from installment plans to revolving credit plans, where customers paid a percentage of the balance and interest accrued on the rest (and notably there was no risk of repossession upon default, as there was in an installment plan arrangement). This shift was what paved the way for the modern credit card.

It wasn’t until the late 1960s that we saw unsecured consumer credit from financial institutions become a more meaningful part of the lending landscape. Up until that point, most of the credit granted to consumers was extended by retailers. This rapid expansion in consumer credit prompted a desperate need for the systematization of credit evaluation at scale. While FICO was making inroads on this front, the data used in its scorecards was siloed within the banks and retailers that it was commissioned by to provide bespoke scoring systems. What was needed was a scaled, centralized repository of consumer credit data.

Over the course of the twentieth century, there was significant consolidation of the U.S. credit bureaus – the companies that controlled the consumer credit data. This industry consolidation accelerated following the introduction of the Fair Credit Reporting Act (FCRA) in 1970, which introduced a regulatory framework that favored the larger credit bureaus that could afford to make the necessary investments in infrastructure to be compliant (many of the smaller bureaus lacked the capital and technical resources to make these upgrades).

This culminated in the emergence of the big three credit bureaus: Equifax, TransUnion, and Experian. These three credit bureaus provided exactly what FICO needed in terms of centralized credit data at scale – a development that was made possible by the digitization of data that stemmed from computing advancements. It was thus in 1984 that FICO released what was called PreScore. FICO’s PreScore used credit bureau files to screen consumers for pre-approved credit card offers (rather than blanketing an entire region with solicitations for credit, only to reject a large portion of those unscreened applicants[8]).

While the PreScore was a standardized score used for marketing credit products, it wasn’t until 1989 that FICO introduced the FICO Score, which was used as a general-purpose credit score to standardize and automate the assessment of credit risk. Rather than the hitherto manual review of consumer credit applications – which were time-intensive, costly, subject to the biases of the loan officer, and often saw inconsistent outcomes – the FICO Score underpinned an automated credit underwriting process that was fairer, more consistent, and expanded access to credit. The FICO Score also emerged at an opportune time – the recession in the late 1980s led to an increasing number of credit card delinquencies. Of the more than 200 million Visa, MasterCard and Discover card products at the time, 14% were subject to credit scoring in 1989, which jumped to more than 50% in the early years of the 1990s[9].

The pivotal moment for the FICO Score came in 1995 when the GSEs Fannie Mae and Freddie Mac directed lenders to use FICO scores for all new residential mortgage applications. Notably, this was when Fannie Mae and Freddie Mac were for-profit enterprises owned by private shareholders; these corporations voluntarily adopted the FICO Score, given that lenders were already using it to evaluate credit risk in mortgage and non-mortgage markets (i.e., FICO’s monopoly in U.S. mortgages was not granted by government decree but rather guided by market forces). This cemented the FICO Score as the industry standard for U.S. residential mortgages, but lender adoption more broadly also saw the FICO Score become dominant in other verticals such as auto loans and credit cards.

The erroneous notion that FICO is only dominant because of its government mandated use in the conforming mortgage loan market is best rebutted when we see that the majority of FICO Scores are used outside of mortgage originations. In a blog post by the President of FICO’s Scores business, Jim Wehmann revealed that 99% of FICO Scores are used outside of mortgage originations.

“In fact, the vast majority, approximately 99%, of FICO Scores used for decisioning across the consumer credit industry are used outside mortgage originations. Total mortgage originations, including GSE-purchased mortgages, constitutes only 1% of all FICO Scores used for decisioning across the consumer credit industry.”

This recount of FICO’s history helps us better understand the importance of FICO in the U.S. consumer credit ecosystem. But to properly appreciate FICO as an investment, and to assess the company’s future growth prospects, we must understand the company’s special price increases.

The remainder of this 16,000+ word deep dive is reserved for paid subscribers. We cover FICO’s special price increases, the royalties FICO receives in a typical mortgage loan origination, the economics and incentives of the CRAs (including why the CRAs are incentivized for FICO to continue raising prices), as well as views on where FICO’s scores pricing might go in future years. Consider becoming a paid subscriber to receive full access to the remainder of this FICO report.

FICO’s Special Price Increases

Given FICO’s dominance in the credit scoring industry, it might surprise readers that FICO kept its royalty rate flat for decades. The wholesale royalties that FICO gets for its mortgage FICO Score were originally set in 1989. According to a blog post by FICO CEO Will Lansing, these wholesale royalties remained at those low amounts for nearly 30 years due to “contractual and technical constraints”[10]. As a result, and despite the enormous value FICO was providing to the U.S. mortgage industry, the company’s Scores revenue growth historically had been entirely driven by growth in the volume of credit pulls. This, however, changed as FICO recut its agreements with the CRAs.

In 2012, FICO started the process of renegotiating its license agreements with each of the CRAs. This took a number of years, and, unsurprisingly, there was immense pushback by the CRAs. Eventually, FICO and the CRAs arrived at an agreement: FICO would be able to adjust its base royalty rates for FICO scores once per year, and with advance notice to the CRAs[11]. Following this, FICO made the first adjustment to its royalty for mortgage scores in 2018, pricing them at around 50 to 60 cents per score.

Then, in September 2022, FICO notified the CRAs that it was adopting a tier-based royalty structure. The royalty rate for the highest volume tier would be unchanged, but would increase for lower-volume tiers. This was only the second price increase in FICO’s roughly three decade history of providing credit scores for the mortgage space, outside of some small adjustments for inflation following the pandemic-era inflationary period. Following this change, FICO in 2023 was collecting around $0.60 to $2.75 per FICO Score. Smaller industry participants voiced concern over this tiered royalty structure, and FICO ultimately eliminated the tier-based structure. Instead, they announced a $3.50 per FICO Score wholesale royalty rate for 2024. Finally, FICO raised this per score rate to $4.95 in 2025. This means FICO increased the prices of its FICO Score product for the mortgage market by 800% in three years, a level of pricing rarely witnessed by companies of FICO’s scale.

FICO had enormous latent pricing power. After renegotiating the CRA agreements, CEO Will Lansing has aggressively exerted this pricing power. That had been a boon for shareholders: FICO’s business-to-business Scores revenues more than tripled between 2018 to 2024 (astute readers will note that this is less than the 536% mortgage scores price increase over that period, and this is due to B2B Scores including other scores categories such as auto loans and credit cards which did not have such large price increases) and FICO’s earnings multiple soared. However, as we wrote in our piece on pricing power, the pace at which prices are increased does matter, and FICO’s speedy price increases have drawn consternation from regulators (we will explore the regulatory developments later in the report).

The Real Cost of Price Increases

While a $4.95 royalty rate per FICO Score does not sound like a lot of money, it is worth understanding some of the nuance around how FICO’s price increases have contributed to credit report cost inflation. We will use an example of a two person tri-merge report for a low volume lender originating a mortgage in the U.S. In 2024, 62% of homebuyers were married couples and 6% were unmarried couples[12]. This means that two-thirds of U.S. home purchases involve two people, both of which require a credit pull (and thus doubling the royalty paid to FICO). The credit of these dual borrowers will usually be checked multiple times during the loan approval process:

• Prequalification – a pre-qualification is a quick, initial assessment of the amount someone might be able to borrow for a home purchase. It is typically used for less-serious buyers who want to window-shop for a mortgage to inform their house price budget. During the prequalification stage there will be a soft credit pull, based on unverified information provided to the lender such as income, debt, and estimated down payment. More serious buyers typically skip the prequalification stage and move to pre-approval, which is a more involved process that results in a conditional commitment from the lender. Note that a pre-approval letter will signal to a real estate agent that you’re a serious buyer, and many realtors might refuse to even show you a property absent this pre-approval.

• Pre-approval – lenders will do a hard pull credit check during loan pre-approval. A hard credit pull is when a lender accesses the borrower’s entire credit report. Notably, other lenders can see hard inquiries and it may slightly reduce the borrower’s credit score. A soft credit pull, on the other hand, will provide more limited information than a hard credit check, but it is not visible to other lenders, does not have an effect on the borrower’s credit score, and can be done without the borrower’s formal consent[13]. A pre-approval process requires a tri-merge hard credit pull, where the lender will formally access your credit reports and FICO Scores from all three major bureaus: Equifax, Experian, and TransUnion. This results in six credit pulls (i.e., three separate credit pulls for each borrower applying for the mortgage) and amounts to $29.70 worth of royalties for FICO (i.e., $4.95 multiplied by six). The tri-merge hard pull is to ensure that the borrowers qualify for a conforming mortgage (whereby a 620 FICO Score is the floor for a conforming loan). The lender will take the middle FICO Score of the three as the representative score that determines loan eligibility (for multiple borrowers the lowest of these middle scores is used). Note that within a 45-day window, multiple hard pull credit checks from a mortgage lender will only affect a borrower’s credit score as if it were a single pull[14].

• Loan processing – occasionally a soft pull or hard pull will occur during the loan processing stage to ensure that the borrowers’ financial status has not shifted, particularly if there are delays in the underwriting or home search process whereby more than 90-120 days have elapsed since the initial credit check. A mid-process discrepancy may also prompt the lender to perform a new credit pull.

• Final credit check – while this will depend on how much time passes between the initial credit report and loan closing, lenders often perform a soft pull credit check just prior to closing (typically 1-3 days before the loan closing date) to ensure that the borrower’s financial circumstances haven’t changed (e.g., acquiring any new loan facilities that may affect their creditworthiness)[15]. This generates $9.90 of royalties for FICO ($4.95 multiplied by the two borrowers).

• Possible second tri-merge pull – Fannie Mae and Freddie Mac require a FICO Score to be no older than 120 days for the mortgage to be conforming (and thus sellable to the GSEs)[16]. This means that if the mortgage loan process, from the initial credit pull through to the loan closing date, takes more than 120 days, then a new tri-merge hard pull credit check will be required. In other words, if the pre-approved prospective home buyers fail to find a property in a circa four month time frame, then a second tri-merge report will occur, resulting in another $29.70 of royalties for FICO. Given the much slower housing market – that is, the median time to sell a home was 43 days in July of 2025[17], a metric that has blown out compared to 16 days in July 2021 – we suspect that a second tri-merge is happening in an increasing number of mortgage applications.

Source: Redfin

We can thus see how in a fairly typical dual-borrower loan application, there could be six hard pulls and an additional two to four soft pulls, resulting in $39.60 to $49.50 of royalties for FICO. In the event that a second tri-merge is needed, the number of hard credit pulls increases to 12, meaning that there might be a total of 14-16 credit pull events. This would result in $69.30 to $79.20 of revenue for FICO.

FICO management has framed the FICO Score, at a $4.95 royalty rate, as being “incredible value”[18]. At $4.95, the $14.90 that FICO receives as royalties in a tri-merge bundle represents c.15% of a $100 tri-merge credit report cost, and just 25 basis points of the approximately $6,000 of mortgage closing costs. This is slightly disingenuous in light of the perhaps 10 credit pulls that occur during a mortgage approval. In this case, FICO’s royalties would amount to 83 basis points of mortgage closing costs – still small, but not quite infinitesimal.

The CRAs have contributed more to credit report cost inflation than FICO

FICO has also rebutted claims that it’s price gouging by pointing out that it is ultimately the CRA – not FICO – that determines what price resellers and end users must pay for the FICO Score[19]. The way it works is that the CRAs will each pay FICO the $4.95 royalty, but will then not only pass this cost on to resellers/end users, but will make a margin on the cost FICO Score. It’s worth parsing through an example to look at the inflation in credit report costs directly attributable to FICO’s price hikes, but then also indirectly driven by FICO via the CRAs marking up the cost of the FICO Score.

We will now continue with the dual borrower tri-merge example in a way that seeks to disaggregate the various cost buckets of a tri-merge report – namely FICO’s royalties, and credit bureau and other third party fees – in order to ascertain which parties have contributed most to overall credit report cost inflation.

Source: Bristlemoon Capital; Jefferies; Company data

Note that lenders will typically pass on the cost of the credit report to the end borrower, including these costs in the mortgage closing costs that the borrower pays. These fees are insignificant, and consumers don’t really care about them. As a former VP at Guaranteed Rate put it, “I have never had anybody complain about the credit report fee, never”[20].

There are a number of things to note about the above analysis. We know what FICO’s royalties are and can make rough estimates around the cost of a joint merged report (although notably there appears to be a wide variance in credit report costs, which is likely due to variations in reseller fees). We can then make some assumptions around the degree to which the credit bureaus mark up the FICO Score, as well as other credit file fees and reseller fees. We have heard a number of figures for the degree to which the CRAs mark up the FICO Score. Our understanding is that Equifax marks up the FICO Score to achieve a 40-50% margin for a tri-merge report, and we’ve heard that they add on a $3.50-$4.00 upcharge to the $4.95 FICO Score price (i.e., resulting in a $8.45 to $8.95 price). It’s likely that mark ups are between 70-100% of the FICO Score but we have simply used 70% for conservatism. The potential for cost variances means that the precise numbers in the analysis will vary based on what credit report cost assumption is used, but we feel that the disaggregation of the various cost buckets of a credit report is useful nonetheless.

There are a few interesting conclusions that fall out of this analysis. The first is that the increase in the price of the FICO Score is directly responsible for 33% of the dollar increase in total credit report costs between 2022 and 2025. The flipside of this is that the credit bureaus and other third-parties account for 67% of the total dollar increase in the cost of a joint merged credit report. FICO doesn’t seem like the bad guy when the credit report cost increases are framed in this manner.

However, it’s not quite this straightforward due to the fact that the CRAs have chosen to price through and make a margin on the increasing cost of the FICO Score. The incremental CRA fees associated with them marking up the FICO Score account for 24% of the dollar increase in total credit report costs between 2022 and 2025. On the one hand, we could say that FICO’s price increases were the catalyst for this portion of the credit report cost inflation – absent those price increases the cost of a credit report would be demonstrably cheaper. On the other hand, it is the CRAs who have chosen to mark up this cost and make a margin on it, rather than simply passing it through. It is both FICO and the CRAs that have contributed to this portion of credit report cost inflation, although FICO has taken almost all of the blame in the eyes of the public and regulators.

In the analysis above we looked at both FICO’s royalties as well as the total credit report cost as a percentage of mortgage closing costs. We would note that it is difficult to obtain precise data on mortgage closing costs. The Consumer Financial Protection Bureau (CFPB) has reported that median loan closing costs were $5,954 in 2022, and we’ve made some assumptions around loan closing costs for the ensuing years. An alternate study from Clever Real Estate estimates that mortgage closing costs are as high as $8,000 in the U.S. It follows that FICO’s royalties as a percentage of mortgage closing costs for a dual borrower tri-merge pull were 41 basis points in 2025, up materially from an estimated 6 basis points in 2022. Now, as we previously mentioned, there might be additional credit pulls during a mortgage application which could push FICO’s total royalties to over a percentage point of mortgage closing costs. However, this is still a relatively small amount.

Despite the enormous increase, it is still quite a stretch to argue that royalties equating to 1 percentage point of the total mortgage closing costs pose an affordability impediment to home ownership. To argue that it is, in our view, is rather preposterous. Even when looking at total credit report costs as a percentage of mortgage closing costs – which we estimate to be 1.6 percentage points in 2025 – it is still a relatively small cost bucket (but again, this rises to above 3.5 percentage points if more than one tri-merge credit report is required as well as mid-process soft credit pulls).

Nevertheless, as a sub-5% component of mortgage closing costs, we feel that credit report costs are not at a level that would be deemed egregious. Rather, we would argue that other components of total mortgage closing costs, many of which are multiples of the cost of a credit report, are much more likely to pose an impediment to home ownership: loan processing fee ($300-$900), underwriting fee ($300-$750), application fee ($200-$500+), home appraisal fee ($500-$1,000+), home inspection fee ($300-$500), title search and title insurance ($300-$2,500+), as well as a range of other junk fees which are too numerous to list in full[21].

Despite our strong view that FICO comprises only a miniscule portion of mortgage closing costs, this has not been enough to stave off regulatory scrutiny of FICO’s pricing actions.

Regulatory Woes

Bill Pulte was sworn in as Director of the Federal Housing Finance Agency (FHFA) in March 2025, and it was not long thereafter that he began attacking FICO via tweets and the media.

Source: X (Twitter)[22]

Source: X (Twitter)[23]

Source: X (Twitter)[24]

The FHFA is the government body that regulates Fannie Mae, Freddie Mac, and the Federal Home Loan Banks (the FHFA was established in 2008 during the depths of the GFC). The FHFA’s authority over the GSEs gives it a direct relationship with FICO, given that the GSEs rely on the FICO Score for the mortgages they acquire (we will explain the GSE dynamics when we discuss FICO’s role in the securitization market). However, the FHFA lacks jurisdiction to directly regulate FICO’s prices.

Pulte’s comments have directly targeted FICO in an incendiary fashion. This strikes us as bizarre. The price of FICO’s scores cannot reasonably be construed as posing a threat to home ownership in the U.S. However, the rapid pace of price increases does make FICO a convenient scapegoat for someone like Pulte, whose political grandstanding and keenness to take action possibly hint at his wider political ambitions. The risk is that Pulte seeks to push through changes to how the FICO Score is used by the GSEs that are politically advantageous for him, rather than changes that are supported by sound logic for improving access to home ownership.

The most significant recent development on the regulatory front was the announcement of lender’s choice in July 2025. Lender’s choice allows mortgage lenders to choose between two competing credit scoring models – FICO 10 T and VantageScore 4.0. The intention is to encourage competition in the credit score market. Notably, the decision to implement lender’s choice walked back an earlier policy announcement that was far more forceful in introducing competition into the credit score market: a dual-score mandate that required lenders to deliver both FICO 10 T and VantageScore 4.0 with each mortgage loan. After years of the FHFA fielding stakeholder feedback and numerous delays to implementation timelines, our view is that the FHFA was forced to abandon its dual-score mandate because requiring both credit scores on loans would have been too much change for the system to bear.

Increased Competition in the Credit Score Market?

FICO 10 T vs VantageScore 4.0

As we are moving towards a world where lenders can choose between FICO 10 T and VantageScore 4.0, it is worth exploring which score is more performant (which might help us gauge the odds of VantageScore making competitive inroads against FICO). This comes down to predictiveness regarding loan defaults as well as how wide of a population the model is capable of scoring. FICO, unsurprisingly, has pointed to an analysis conducted by the GSEs that compared FICO Score 10 T and VantageScore 4.0 and found FICO Score 10 T to significantly outperform VantageScore 4.0 in mortgage predictiveness and accuracy[25]. FICO has urged the FHFA to release this analysis to the public, but the agency has so far chosen not to release the study.

In a white paper, FICO noted that it has not had an opportunity to directly compare Classic FICO nor FICO 10 T to VantageScore 4.0, as it does not have access to VantageScore data. However, FICO did draw attention to a third-party study by the Urban Institute that analyzed the predictive strength of Classic FICO (i.e., a 20 year old FICO Score currently being used by the GSEs) and VantageScore 4.0, examining loans between 2013 to 2023 in the Fannie Mae Single-Family Loan Performance database[26]. The study looked at 90+ days past due data and then compared the resulting ROC curves from the two scores. The following quote from the Urban study is illuminating:

“Both credit scoring models effectively distinguish between high-risk and low-risk borrowers. VantageScore 4.0 is marginally more effective at identifying high-risk borrowers from among those with the lowest credit scores, though the differences are small.”

To help visualize how small the differences are between Classic FICO’s and VantageScore 4.0’s ability to delineate between high-risk and low-risk borrowers, we can observe the graph below. Note that there is barely any white space between the two curves, signaling a minimal performance uplift from VantageScore 4.0 over Classic FICO.

Source: FICO; Urban Institute study using Fannie Mae data

This is surprising, given that the 20 year old Classic FICO Score is being compared against the newest generation of VantageScore. In a separate study, FICO compared the performance of FICO 10 T to Classic FICO, and noted that FICO 10 T had significant performance uplift, with an 18.1% relative improvement in defaulters captured in the lowest decile. As we can see on the graph below, and the white space between the curves, FICO Score 10 T materially outperforms Classic FICO. The implication here when marrying up the two studies is that FICO 10 T is more predictive regarding mortgage defaulters compared to VantageScore 4.0.

Source: FICO

FICO has also noted that lenders representing over $300 billion of annual mortgage originations and $1.5 trillion in mortgage servicing portfolios have already adopted and started trading loans with FICO 10 T outside of the conforming market[27].

Former FHFA Director Sandra Thompson, who was privy to both the FHFA’s and GSE’s analyses of FICO 10 T versus VantageScore 4.0, concluded that there was minimal difference between the two scores (via Tegus):

“We came up with the conclusion that both credit score models that were submitted, there weren't many differences in accuracy. There weren't many differences in predictability or reliability. They weren't exactly equal. One was better in, let's say, inclusiveness [i.e., VantageScore]. The other was better in loss predictability [i.e., FICO], but it wasn't a material difference” [bracketed additions ours].

We don’t believe that VantageScore will steal meaningful share from FICO as the performance uplift appears to be small, and the friction of shifting to an alternate score is very high. This echoes the comments made by FICO management:

“we've been competing with VantageScore for a very long time well over a decade. And we have not experienced any kind of significant share loss to Vantage. And I would say that's because of the two things that we talked about before. One is that we have the best score. And second, that there's a lot of benefit to working with the industry standard, which is FICO.” – Will Lansing (CEO), Q3 2025 Earnings Call.

VantageScore is more inclusive?

VantageScore 4.0, the model that has been approved for use for conforming mortgages, was introduced in 2017 and was the first scoring model to use trended data (i.e., analyzing a consumer’s credit behavior over time, such as payment patterns on revolving credit, rather than just a single point in time snapshot). The key benefit of VantageScore 4.0 is its ability to score “thin-file” borrowers – that is, those with limited credit history.

The benefit of this to lenders is the ability to write more loans when using VantageScore 4.0, which helps them grow origination volumes. The flipside is if this increase in origination volumes comes at the expense of laxer underwriting standards, which could ultimately manifest in higher loan losses. In other words, there is adverse selection risk if a lender writes a loan to a VantageScore borrower that would otherwise not have qualified for the loan (or a more favorable interest rate) if they had been scored by the FICO Score, knowing that this loan will be palmed off to the GSEs. The GSEs would need to account for this adverse selection risk by remapping their LLPA grids.

VantageScore claims that VantageScore 4.0 is capable of scoring an incremental 33 million consumers. However, only 13 million of these consumers have a VantageScore above 620, and only 4.9 million of those consumers are mortgage eligible (aged between 25 and 65). Ultimately, this translates into an incremental 2.7 million mortgage loans (based on 1.83 adults per household) and $1 trillion of additional mortgage loan opportunity due to VantageScore 4.0.

Source: VantageScore

A good chunk of VantageScore’s ability to score more consumers stems from it requiring only one month of credit history, compared to the six months required by FICO (so to us, VantageScore is an inherently less conservative score than FICO). However, FICO 10 T’s use of trended data is likely to narrow the inclusiveness gap with VantageScore 4.0, given it improves mortgage approval rates by 5% relative to Classic FICO without adding incremental risk[28].

Our view is that VantageScore 4.0 is not a vastly superior score to FICO Score 10 T, which makes it less likely that lenders will push to rapidly adopt VantageScore.

FICO’s Relationship with the Credit Bureaus

FICO has a somewhat unique relationship with the three credit bureaus. On the one hand, the credit bureaus are FICO’s largest customers. They license FICO’s IP, but a large part of the value of the FICO Score derives from the credit data owned by the credit bureaus. On the other hand, the credit bureaus formed a joint-venture and launched a competing score, VantageScore, which is FICO’s most serious competitor. We would define FICO and the credit bureaus as “frenemies”, but think that they’re friendlier than what people understand when we consider the economics captured by the credit bureaus from the FICO Score (we will get to this). Let’s first understand how VantageScore came about and the details of this arrangement.

The quirks of the Vantage Score JV

VantageScore was established in 2006 as a joint venture by the three major credit bureaus: Equifax, Experian, and TransUnion. Given that the three bureaus control virtually all of the credit data used to generate credit scores, and that these three competing bureaus were entering into a contract to form a for-profit LLC, there was very careful structuring of the agreement to ensure that the VantageScore JV wouldn’t fall afoul of antitrust laws and be deemed a collusive arrangement.

This meant that VantageScore would not get compensation for credit scores its model was used to generate – that is, VantageScore would not be collecting royalties in the same fashion that FICO does. The reason for this is that under U.S. antitrust law, if VantageScore were to set a wholesale royalty rate for its score in the same way that FICO does, it would be accused of price-fixing. Rather, VantageScore receives a fixed licensing fee from the bureaus (which is divided equally amongst the three CRAs) for renting out its intellectual property. It is the bureaus who then use the VantageScore model to generate and sell the credit scores.

Each credit bureau is responsible for negotiating pricing with lenders, as there is no fixed price for VantageScore. This is a subtle but important constraint in VantageScore’s pricing strategy that introduces complexity and friction as it relates to lender’s adopting VantageScore. We do not believe this approach to pricing will change, given the abovementioned antitrust risk. The three credit bureaus compete with each other on price when selling VantageScore to lenders. We do not have a great sense for whether there is genuine competitive tension amongst the three bureaus when they are setting prices, but suspect that there is not.

Do the credit bureaus make more from a FICO Score than from VantageScore?

FICO’s pricing power depends on a few things – namely, whether VantageScore gets traction with lenders, and then how VantageScore is priced relative to the FICO Score.

In the illustrative example below, we are comparing FICO and VantageScore under different VantageScore pricing scenarios. From this, we can observe changes in fee revenue for the credit bureau, as well as changes in the credit report cost (excluding reseller and other fees). This will help us conceptualize how VantageScore might be incentivized to price its score going forward.

Source: Bristlemoon Capital; Jefferies; company data

We assume that the credit bureau is marking up the FICO Score by 70% and passing this on to the lender. We have also assumed that VantageScore is priced the same as the FICO Score – that is, at $4.95. From this, we can see that the credit bureau’s fees increase by 9% when a lender uses VantageScore rather than the FICO Score (and the credit bureau’s margin goes up, given the FICO royalty is eliminated). The credit report cost (excluding reseller fees) declines by 16%.

Just based on the math of how much the credit bureau is marking up the FICO Score, the credit bureau starts to lose fee revenues if they price VantageScore at a greater than 30% discount to FICO. This is because the credit bureaus are marking up the FICO Score by 70%, and thus revenue from VantageScore must surpass the margin they’re making on the FICO Score.

Using the same example, if instead the credit bureau were to price VantageScore at a 50% discount to the FICO Score, then the credit bureau will lose 6% of its revenues relative to if a FICO Score was used to score the loan. The credit report cost (excluding reseller fees) falls by 28% in this scenario.

Source: Bristlemoon Capital; Jefferies; company data

There is another scenario where the credit bureau chooses to price VantageScore above that of the FICO Score, perhaps on the premise that VantageScore is capable of scoring incremental borrowers, and that even at a price premium the cost to the lender will be less than using FICO.

If the credit bureau priced VantageScore at $7, a premium to the $4.95 FICO Score, then the credit bureau could manufacture a 22% increase in its fee revenues, all while the credit report cost (excluding reseller fees) declines by 7%.

Source: Bristlemoon Capital; Jefferies; company data

Due to the markup of the FICO Score by the bureaus, a credit bureau would need to price VantageScore at a more than 70% premium to FICO before the credit report cost begins increasing relative to if FICO was instead utilized to score the borrower.

This creates a unique set of incentives for the credit bureaus. Due to not having to pay FICO royalties, they can charge a premium for VantageScore in a way that juices their fee revenues, but also reduces credit report costs.

However, the problem with this scenario is that the credit bureaus would risk putting a target on their back from regulators if they started pricing at such a significant premium to FICO. We believe that the markup the credit bureaus make on the FICO Score is certainly not something the bureaus would want to draw the attention of regulators to. The bureaus have an ideal situation that we think they’re highly likely to not want to disrupt: FICO has been taking the brunt of lender and regulatory criticisms from its price increases, all while the credit bureaus get to ride on the coattails of FICO’s price increases and benefit economically from those incremental pricing-driven fee dollars.

The economics for the credit bureaus suggest that they’re incentivized to not price VantageScore at a more than 30% discount to FICO in order to maintain their fee revenues. This says to us that the risk of a race to the bottom style pricing war for scores is incredibly unlikely, which bodes well for FICO’s ability to continue raising prices (and we would again note that FICO’s royalties continue to comprise a very small portion of mortgage closing costs).

In fact, the incentives of the credit bureaus suggest that they would be in strong support of FICO continuing to aggressively increase prices, given that they directly benefit by marking up the FICO Score, and it gives them cover to raise prices of VantageScore. FICO management has in the past confirmed that the credit bureaus earn more from marking up the FICO Score than what FICO itself earns from royalties; we think there’s a very strong incentive for them to not disrupt this profitable stream of revenues.

It is also worth saying that just because FICO has been a monopoly and a second player (i.e., VantageScore) is trying to enter the market, it does not mean that a duopoly industry structure will result in irrational pricing and an erosion of industry economics. We would point to the triopoly of the ratings agencies (S&P, Moody’s, and Fitch) and how the two listed ratings agencies (S&P and Moody’s) have been very solid compounders. We don’t think any traction achieved by VantageScore would eliminate FICO’s pricing power and preclude FICO from continuing to grow its revenues at a decent clip.

It is also worth pointing out that FICO has strategic alternatives it could use to preserve its economics in the scenario that VantageScore were to try to undercut FICO on pricing. FICO could reduce its upfront scores pricing and seek to monetize later in the loan origination process (e.g., when the loan closes and/or at the time the loan is securitized in the secondary market), re-implement volume-based tiered pricing to incentivize lenders to continue scoring higher volumes of loans using FICO (or pivot to an all-you-can-eat model), bypass the CRAs as the intermediary layer and go direct-to-lender (which would eliminate the credit bureau markup in order to lower the cost to the lender), and potentially price soft and hard enquiries at different rates (e.g., soft pull prices could be reduced, or even made free, which would be offset by a large increase in hard pull prices). The company can also manufacture strong overall Scores revenue growth by pulling the pricing lever in non-mortgage verticals such as auto scores.

As it relates to VantageScore competing against FICO on pricing, the following quote from the former Vice Chairman of VantageScore (via Tegus) is instructive:

“The pricing would never be the lead attribute of selling a new scoring system because even if a competing model were cheaper, if it were an inferior model, you'd never make up in pricing what the lender would lose in additional losses. The lead proposals would always be the benefits of the model, scoring more people, which includes being more inclusive, but also being more predictive.”

In other words, it doesn’t make economic sense to take the low bid on the credit score if there’s even a slight risk of incurring higher loan losses. Any additional loan losses from adopting a score that is less effective in assessing credit risk, given the size of mortgage loans, will almost always usurp the miniscule savings derived from using the lower-priced credit score. FICO management have mentioned in recent months that they have not seen any shift in mortgage score market share following the FHFA announcing the lender’s choice mandate.

There is a segment of the mortgage market that we have not yet discussed which is wary of loan losses. It is now worth us spending some time discussing the GSEs and the securitization market, and why they act as an additional impediment to VantageScore gaining share.

The GSEs and the Securitization Market

We will take a brief detour to discuss the GSEs, as we believe this is necessary to understand why FICO is so entrenched in the mortgage market.

Freddie Mac and Fannie Mae, collectively referred to as the government-sponsored enterprises (GSEs), guarantee around 52% of the c.$13 trillion worth of U.S. residential (i.e., single-family and multifamily) mortgages[29]. The GSEs do not lend money to the public. Rather, they purchase mortgage loans from mortgage originators and then securitize those loans – that is, loans are grouped with thousands of other similar mortgages to create a “pool”, and then this pool of mortgages is packaged into a security called a mortgage backed security (MBS). MBSs are then sold to investors in the secondary market. The GSEs will guarantee principal and interest on the loans they purchase from lenders in return for a guarantee fee that is paid by the lender.

The role of Fannie and Freddie has been to ensure stability and liquidity in the U.S. mortgage market. Following the collapse of the GSEs during 2008, Fannie and Freddie were placed into conservatorship in September 2008, after which the U.S. government explicitly guaranteed all GSE mortgages (although an implicit guarantee existed before 2008, which no doubt contributed to the risky behavior of the GSEs leading into the GFC).

The GSEs remove default risk for lenders, allowing lenders to offer better terms and rates than otherwise for conforming loans (i.e., loans that meet the criteria to be sold to the GSEs, which is based on the maximum loan limit, required minimum down payment of 20% (unless private mortgage insurance is taken out), minimum credit score, and minimum debt-to-income-ratio).

Below is a visual depiction of the mortgage value chain:

Source: Milliman

This is all worth understanding as it relates to FICO for a number of reasons:

• FICO plays an important role in the securitization market as the language via which credit risk is communicated between secondary market investors. Any disruptions in secondary market investors’ ability to price the credit risk of loan pool, even if it is just the unfamiliarity and uncertainty that might come from using an alternate score like VantageScore, flows through to wider spreads, reduced liquidity (which affects GSEs’ ability to offload purchased loans to the MBS market), and ultimately higher interest rates for borrowers.

• MBS investors are likely to be skeptical of VantageScore’s ability to model prepayment risk, which is a key risk for MBS investors.

• The risk of loan putbacks by the GSEs might make lenders reluctant to adopt an alternate score such as VantageScore.

FICO Score the industry standard for mortgage securitization

FICO is the preferred score in the securitization market. It has a long and proven history as the measure of credit risk for pools of mortgages. This incumbency is powerful; it makes the FICO Score incredibly sticky as a risk-based pricing tool used by MBS investors.

MBS investors care primarily about cash flow predictability. The cash flow predictability of an MBS is driven by two things: 1) charge-offs, which are incurred when a loan defaults; and 2) prepayments, which occur when the borrower pays off the loan earlier than expected, creating reinvestment risk for the lender. Note that loan prepayments typically accelerate in a declining interest rate environment, meaning that the MBS holder is not only forgoing the future interest payments they were expecting, but will also generate less interest income on any new issue MBS due to the lower interest rates.

FICO Scores help investors assess the likelihood of both charge-offs and prepayment risk. When MBSs are being issued to investors, the prospectus will include an average weighted FICO Score. The underlying loans in the mortgage pool are grouped together based on FICO Scores, and it is the FICO Score that acts as a uniform credit standard that ultimately underpins investor confidence in these mortgage backed securities. Absent the FICO Score, there would be no practical way to gauge the risk of the thousands of underlying loans within an MBS, which would result in a less liquid secondary market, and most likely a higher mortgage rate for borrowers.

But the question now, with the FHFA pushing for greater competition in the score market by allowing the GSEs to accept loans scored with VantageScore, is the degree to which MBS investors would be comfortable investing in VantageScore-scored loan pools, or mixed loan pools. This requires us to understand the to-be-announced market and how standardization is a key feature.

The To-Be-Announced (TBA) Market

Most residential mortgages in the U.S. are securitized (rather than held on the balance sheet of the original lender), and more than 90% of agency MBS trading volume occurs in a forward market referred to as the to-be-announced (TBA) market[30]. This liquid forward market (which looks out to a horizon of a few months) allows mortgage lenders to lock in sale prices for new loans as, or even before, those mortgages are originated[31].

It is critical to understand how the liquidity that the TBA market provides for agency MBS is predicated on standardization, as this will allow us to appreciate the ramifications of attempting to introduce a second scoring method such as VantageScore into the TBA market. A white paper by the Federal Reserve Bank of New York does a good job of explaining the mechanics of the TBA market:

“In a TBA trade, the seller of MBS agrees to a sale price, but does not specify which particular securities will be delivered to the buyer on settlement day. Instead, only a few basic characteristics of the securities are agreed upon, such as the coupon rate, the issuer, and the approximate face value of the bonds to be delivered. While the agency MBS market consists of thousands of heterogeneous MBS pools backed by millions of individual mortgages, the TBA trading convention allows trading to be concentrated in only a small number of liquid forward contracts.”[32]

What this means is that the actual identity of the securities (represented by what’s called a CUSIP) to be delivered at settlement is unknown in a TBA trade. Rather, securities are segmented across six parameters: issuer, maturity, coupon, price, par amount, and settlement date[33]. In this sense, groups of MBS that share these six characteristics are considered fungible and can be delivered into a TBA trade[34]. It follows that trading is concentrated in only a few dozen TBA contracts, which further improves the liquidity of this market. It is thus clear that the TBA market is highly reliant on standardization, and the FICO Score has been a key component in this standardization, allowing MBS investors to confidently assess and price credit risk in these fungible pools.

The integration of a second, differently calculated score like VantageScore would, in our view, disrupt this established system, introducing uncertainty and potentially harming market liquidity. As the VantageScore weights credit factors differently than FICO, this would reduce the homogeneity of these loan pools, possibly to a point where a pool of mortgages underwritten with FICO scores would not be considered interchangeable with a pool underwritten using VantageScore. While it’s difficult to know precisely how the market might receive the introduction of a second score, there is the risk that the TBA market fractures into two separate trading pools, bifurcated based on the scoring model used. This is negative to the extent that it reduces MBS liquidity.

Now, given that mortgage credit risk is borne by the guarantor (i.e., the GSEs), MBS investors care primarily about prepayment risk. It is prepayment risk that is the primary source of variances in value amongst agency MBS. Investors use FICO scores to measure their pricing and prepayment models. The introduction of a new scoring model would require these models to be recalibrated and in our view would introduce a level of uncertainty that MBS investors would seek compensation for via a discount on loan pools using VantageScore (i.e., a higher yield).

However, even if investors recalibrate their internal models, there is still the adverse selection issue – that is, the possibility that loan originators select the credit score that allows them to write the most loans, even if those borrowers are riskier credits, knowing that they will ultimately sell these loans to the GSEs, which are then offloaded to investors in an MBS. This could cause investors to demand a risk premium for VantageScore-scored loans – that is, by requiring a discount for VantageScore-scored loan pools. We believe that this would ultimately be borne by borrowers in the form of higher interest rates.

We believe that investors take solace from the fact that FICO has been around for so long, providing a long history of through-the-cycle credit performance. FICO periodically updates its scoring models to incorporate new data and modeling techniques. Newer generations of FICO Score are backward compatible, meaning that, say, a 700 FICO Score 10T is compatible with a 700 FICO Score 8. This is important so that a given score continues to have meaning across a time series when assessing credit performance.

The power in the FICO Score lies in its incumbency. Given it has been used for decades, lenders can see how a FICO-scored loan has performed over time through different points in the credit cycle. The value in the FICO score is not simply in assessing risk when originating loans, but also with the ongoing monitoring of credit risk for a loan. We would also note that the trended data used by VantageScore 4.0 only goes back to 2013[35], thus lacking the same long performance history provided by the FICO score. This is relevant when we observe how meaningfully loan performance can change at different points in the cycle.

Source: Calculated Risk; MBA

The fact that the FICO Score is given away for free in the mortgage industry in virtually all instances outside of the origination of a mortgage makes it much harder for an alternate score such as VantageScore to get traction, given it takes away VantageScore’s ability to compete on price. Jim Wehmann, President of FICO’s Scores business, pointed out this dynamic:

“Notably, after its initial use in mortgage origination, the FICO Score (often after being separated from the tri-merge bundle) is generally used broadly throughout the mortgage credit ecosystem, at no incremental royalty collected by FICO beyond the per score amount collected for the initial origination use itself.”

With this said, it is highly unlikely that VantageScore will be able to make inroads in the mortgage securitization market, even if it were to give its score away for free.

Risk of loan putbacks

As a lender, there are also risks associated with selling a mortgage loan to the GSEs – namely, repurchase risk, whereby the lender is required to repurchase the loan from the GSE due to a breach in the loan’s representations and warranties. Typically, a loan default is the catalyst for a GSE to review a loan’s documentation to see whether there are grounds for a “loan putback” (i.e., when the lender is forced to buyback the loan from the GSE). Basically, when a loan is deemed to have a manufacturing defect, the GSEs can demand that the originator repurchase the loan. Freddie Mac reported that the two largest drivers of putback loans were missing key documents or loans with inaccurate income calculations[36].

The following quote from the Former Vice Chairman of VantageScore (via Tegus) illustrates how lenders are afraid of buybacks, and by extension, any changes to the underwriting process that risk a higher level of defaults:

“The reason, by the way, lenders are so concerned about prospective default, you may know this, if a loan goes into default owned by Fannie or Freddie or guaranteed by them, Fannie or Freddie will pull the loan and look at the documentation and if there's any discrepancy in the documentation, then they pull it out of the securitization and put it back to the lender. That's a big deal from a liquidity standpoint. They're big dollar transactions as opposed to a credit card. The lender is deathly afraid of buybacks. That's why they want to model, even though the loans are being purchased. They need to have a model that can accurately predict default because they don't want to be in a buyback position. That's a huge issue for the mortgage lenders.”

The originator must pay the GSE $100 par value when they buy back the loan. They then typically resell the loan for $93 to $95 (i.e., a 5-7 point discount from par)[37]. The GSEs also have up to 36 months to demand a loan repurchase (and even longer if the loan is fraudulent). This meant that there were loan putbacks occurring in 2024 at 7% mortgage rates on loans that were written when the U.S. 30-year fixed rate mortgage was sub-3%. The fact there’s a time lag between loan origination and when the loan is putback, creates a lot of financial pain for originators in a rising interest rate environment (where the prices of the loans are falling). So, put simply, originators of mortgages very much do not want to have to repurchase a loan from the GSEs (the losses from putbacks will ultimately get passed along to consumers, with the omnipresent risk of loan buybacks tightening the credit box). We believe this militates against the idea that lenders will readily transition to VantageScore, with a relatively unproven track record (and modeled performance that might have inadvertently been optimized for historical conditions).

We would finally note that the lenders pass the cost of the FICO Score on to the end borrower. If this is the case, then why is there any pushback at all from lenders? To understand this, we must look at independent mortgage banks.

Why are lenders complaining about the FICO Score if the cost is passed through to consumers?

“I was on the Mortgage Bankers Association board, I'd say to my buddies, other CEOs of the big mortgage companies, "Hey, aren't you concerned about the rising cost of credit scores?" They'd laugh at me and say, "Yeah, but that's a pass-through expense. We don't really give a shit." I said, "Okay, I get it." Therefore, we realized there wasn't going to be any big lender pushing back.” – Former Founding President and CEO of VantageScore (via Tegus)

The above quote suggests that there is unlikely to be lender pushback to FICO raising the prices of its scores (as this expense is passed through to the borrower). More specifically, the former CEO of VantageScore is referring to big lenders – an important qualification and one that hints at smaller lenders having a different sentiment towards FICO’s price increases.

To understand why some participants of the mortgage industry are unhappy with the FICO price increases, it’s worth understanding what independent mortgage banks (IMBs) are and why they are relatively more exposed to FICO’s price increases.

IMBs are monoline financial companies that are predominantly focused on home mortgage financing and other ancillary, related lines of business, such as mortgage servicing. They are predominantly small businesses with roots in their local communities[38]. These IMBs are non-depository institutions that use a combination of their own cash (c.2-5% of the loan amount) as well as warehouse lines (i.e., short term borrowings) to fund individual mortgages[39]. The warehouse lines are very short-term credit facilities, typically drawn for one to three weeks until the loans are sold. These IMBs shift the mortgages off their own balance sheets. The vast majority of these loans are either sold to “aggregators” (i.e., larger lenders), sold directly to Fannie Mae and Freddie Mac (GSEs), or issued as securities that Ginnie Mae guarantees[40]. The aggregators will either keep the loans on their own balance sheets, or will choose to also sell the loans to the GSEs.

IMBs are an important part of the mortgage lending landscape. In the U.S., they are the primary source of single-family mortgage credit (particularly for lower and middle income families)[41]. In a 2024 IMB Report, it was estimated that IMBs originated 83% of all mortgage loans and 75% of all GSE loans. In other words, IMBs are a very significant channel for loan origination volumes. The problem for IMBs is that they operate on quite thin margins.

In Q2 of 2025, IMBs and mortgage subsidiaries of chartered banks reported a pre-tax net production profit of $950 on each loan they originated[42]. As can be seen in the chart below, which looks at pre-tax net production income expressed in basis points, there were eight consecutive quarters of losses for this group of lenders between Q2 of 2022 and Q1 of 2024.

Source: MBA

Pre-tax net production income of $950 per loan is not a huge amount of money. The problem is that if the cost per loan originated increases, it can quickly erode that thin sliver of profit.

Now, typically when a loan application converts into a loan, the originator will pass on the cost of the FICO Score to the end borrower. The problem is when there are loan enquiries and FICO credit pulls, but the loan ultimately does not close. In this case, the originator must wear the cost of the (typically multiple) credit pulls. This increases their cost per originated loan (as the cost of credit pulls for loans that do not close must be amortized across the remainder of the loans that are originated).

When you’re talking about $950 of net production income per loan, if there are say $50 of FICO royalties that the originator has to wear for an unsuccessful loan application, then this represents a not insignificant profit hit (i.e., 5%). The profit hit only worsens to the extent that there are multiple unsuccessful loan applications that required multiple credit pulls. But remember that it’s not just the cost of FICO’s royalties that the IMB has to wear; it’s the total credit report cost for a tri-merge pull which is $100-$130. So, $130 for a tri-merge credit report in a loan application that doesn’t close will wipe off c.14% of the net production income of the next loan that does close.

IMBs also have a higher reliance on originations than larger banks, with those larger integrated banks having the ability to absorb losses in their mortgage division through other business units such as wealth management and investment banking. IMBs rely almost exclusively on mortgage origination and servicing for income, which makes them much more exposed to cost increases and fluctuations in the mortgage market. With that said, mortgage origination volumes have been extremely weak in recent years due to the higher interest rate environment.

Source: MBA; Jefferies

Mortgage origination volumes collapsed after the rapid interest rate increases over the course of 2022. In fact, mortgage origination volumes declined by a whopping 73% from the 16.8 million loan originations in 2021 down to a trough of 4.6 million originations in 2023[43]. Refinancing also collapsed by over 90% from the pandemic peak[44]. These lower loan volumes means that IMBs feel the cost of credit pulls from unsuccessful loan applications even more, given that these credit pull costs are spread over fewer loans. We would note that this impact is even more severe for smaller IMBs. In 2024, small IMBs with loan production volumes of less than $500 million experienced their third consecutive year of average net production losses, given the difficulty in spreading the fixed costs of originating loans over the much lower loan volumes[45]. Considering this, it becomes understandable why these IMBs have been relatively more sensitive to FICO’s rapid price increases.

We believe that many of the problems experienced by the IMBs will be cured by an improvement in mortgage origination volumes, which might quell some of their vocal criticisms of FICO’s price increases. With the Trump administration hell bent on lowering rates, this is a scenario that we view as entirely plausible. More broadly, an uptick in mortgage origination volumes would be a boon for FICO, given that it is a direct beneficiary of higher mortgage origination volumes in its Scores business.

What is the Value of the FICO Score?

In attempting to figure out the price ceiling for a FICO Score, it is instructive to try to model out the value generated by the FICO Score. Given that the VantageScore is less proven, and lenders have less familiarity with that score, we are of the belief that some lenders originating mortgages using VantageScore will require a higher interest rate (this would be via the application of a higher LLPAs if the lender perceives a higher risk associated with a VantageScore-scored loan). We have no real way to gauge what the interest rate uplift might be but have looked at different risk premium scenarios that range from a 1-10 basis point uplift in the annual mortgage rate for a mortgage scored with VantageScore, relative to FICO.

Source: Bristlemoon Capital

In the example above, for every 1 basis point of risk premium from using VantageScore instead of the FICO Score, the borrower will pay an additional $926 over the life of the loan. And this is all for an upfront saving on the credit report cost that’s in the tens of dollars.

We strongly believe that FICO is generating far more value via its FICO Score than it is receiving by way of royalties. Thought of another way, the value of the savings over the term of a mortgage are potentially 100x greater than the $50 in royalties FICO might receive in a dual borrower mortgage origination, and an even larger amount of any upfront credit report cost savings from using VantageScore over FICO. This says to us that FICO has a very long runway to continue raising mortgage score prices, predicated on this mismatch between what it’s charging and the value it is providing to the mortgage industry.

Concluding Comments and Valuation

We believe that FICO still has pricing power, and that CEO Will Lansing has the chutzpah to raise FICO scores prices even while under FHFA scrutiny. Lansing in a recent non-deal roadshow communicated that FICO’s mortgage score pricing strategy remains unchanged following the FHFA introducing the lender’s choice mandate. In fact, we foresee a scenario where Lansing knows that he has a window over the next few years to aggressively raise FICO’s scores prices before the mortgage industry can recalibrate and before VantageScore potentially becomes a more meaningful threat. Given our hypothesis that VantageScore’s pricing will be anchored to that of the FICO Score, particularly given the CRA economics from marking up the FICO Score that we explored, we think that FICO’s management would prefer a demonstrably higher starting point for scores prices before there is any meaningful prospect of competition from VantageScore (which we think would be years away, if at all).

FICO can, in a way, manufacture aggregate Scores pricing growth by pulling the pricing lever in non-mortgage verticals such as auto and credit cards. However, we believe that there are characteristics in those verticals that mean FICO’s pricing power is not nearly as great as in the mortgage vertical (e.g., smaller loan sizes, more frequent credit pulls for delinquency monitoring, etc.). Regardless, we see it as likely that FICO will adjust its Scores prices across these different verticals, including mortgage. Lansing has noted that auto dealers are making an average profit per car financed of $1,600, which allows auto dealers to absorb further FICO auto scores price increases (auto scores prices were most recently increased by a double digit percentage rate, according to the company).

As it relates to FICO’s pricing strategy more broadly, Lansing made the following comments:

“So we try to be fairly consistent with our increases. I think it's fair for them to expect increases every year. I think they do.” – Will Lansing (CEO), Barclays Americas Select Franchise Conference 2024.

During the non-deal roadshow, Lansing contrasted FICO’s price-to-value approach to that of the rating agencies, which charge 8 basis points on average to rate corporate debt. He then noted applying that same 8 basis points to the average U.S. mortgage amount of c.$375,000 results in a $300 price per FICO score (i.e., 60x greater than the current $4.95 price of a mortgage score). While we don’t think that mortgage scores will get anywhere near $300 per score, it’s an interesting analogy, and Lansing reiterated that the price increases for 2026 (which will be announced later this year) will be similar to previous years (i.e., significant!)

Below we have extended our previous example to observe what FICO’s mortgage score prices could get to. If FICO were to increase its mortgage score prices at a 35% CAGR, then the FICO Score would reach c.$22 per score. This would equate to roughly 1.4 percentage points of mortgage closing costs for a dual borrower tri-merge. While that number would be up 3.5x over the five year period, it is still a relatively small cost, particularly relative to the value FICO is providing. Assuming that the credit bureaus maintain their 70% markup of the FICO Score, total credit report costs would inflate to 3.7 percentage points of mortgage closing costs.

Source: Bristlemoon Capital; Company filings

We can make some rough estimates to see how this scores pricing converts into FCF for FICO in future years. We have focused on FCF per share, given the company’s focus on buybacks. On the Q3 2025, management made the following comment: “We have a lot of dry powder, a lot of capacity. And although we're not going to spend it all in one week, we're buyers at this level.” It’s worth pointing out that FICO has reduced its share count by 28% over the last decade, and by almost half over the past 15 years.

Source: Bristlemoon Capital; Company filings

We can also look at a situation where mortgage volumes begin to normalize, given there is a plausible scenario where interest rates are materially lower over the next few years, which would drive mortgage origination and refinancing volumes.

Source: Bristlemoon Capital; Company filings

It is difficult to know how mortgage origination volumes (and thus the volume of FICO credit pulls will progress over time). Below is a chart of the U.S. outstanding residential mortgage balances, segmented by interest rate. Over half of all U.S. mortgages have a sub-4% interest rate, so there likely needs to be a fairly significant reduction in the currently 6.55% 30-year fixed mortgage rate before homeowners are willing to move properties and give up the attractive mortgage rate they locked in.

Source: FHFA, National Mortgage Database (NMDB)

The 30-year fixed mortgage rate, however, is certainly moving in the right direction!

Source: Bloomberg

We hope you enjoyed our FICO write up. Feel free to leave a comment – we would love to hear any thoughts and pushback on our deep dive.

Disclaimer / Disclosures

The information contained in this article is not investment advice and is intended only for wholesale investors. All posts by Bristlemoon Capital are for informational purposes only. This article has been prepared without taking into account your particular circumstances, nor your investment objectives and needs. This article does not constitute personal investment advice and you should not rely on it as such. This document does not contain all of the information that may be required to evaluate an investment in any of the securities featured in the document. We recommend that you obtain independent financial advice before you make investment decisions.

Forward-looking statements are based on current information available to the author, expectations, estimates, projections and assumptions as to future matters. Forward-looking statements are subject to risks, uncertainties and other known and unknown factors and variables, which may affect the accuracy of any forward-looking statement. No guarantee is made in relation to future performance, results or other events.

We make no representation and give no warranties regarding the accuracy, reliability, completeness or suitability of the information contained in this document. To the maximum extent permitted by law, we do not have any liability for any loss or damage suffered or incurred by any person in connection with this document.

Bristlemoon Capital Pty Ltd (ABN: 22 668 652 926) is an Australian Financial Services Licensee (AFSL Number: 552045).

George Hadjia and Daniel Wu are associated with Bristlemoon Capital Pty Ltd. Bristlemoon Capital may invest in securities featured in this newsletter from time to time.

[1] https://www.safegraph.com/blog/data-standards-and-the-join-key

[2] https://www.safegraph.com/blog/data-as-a-service-bible-everything-you-wanted-to-know-about-running-daas-companies

[3] https://www.newyorkfed.org/microeconomics/hhdc

[4] https://timesmachine.nytimes.com/timesmachine/1961/07/09/118044097.html?pageNumber=125&login=smartlock&auth=login-smartlock

[5] https://escholarship.org/uc/item/7n1369x2

[6] chrome-extension://efaidnbmnnnibpcajpcglclefindmkaj/https://escholarship.org/content/qt7n1369x2/qt7n1369x2_noSplash_e82052de22b30934ab753e60742fa2e4.pdf

[7] https://www.richmondfed.org/-/media/richmondfedorg/publications/research/econ_focus/2013/q4/pdf/feature3.pdf

[8] https://www.fundinguniverse.com/company-histories/fair-isaac-and-company-history/

[9] https://www.fundinguniverse.com/company-histories/fair-isaac-and-company-history/

[10] https://www.fico.com/blogs/fico-s-adoption-and-pricing-mortgage-origination-market

[11] https://www.fico.com/blogs/fico-s-adoption-and-pricing-mortgage-origination-market

[12] https://www.nar.realtor/research-and-statistics/research-reports/highlights-from-the-profile-of-home-buyers-and-sellers

[13] https://www.redfin.com/blog/how-many-credit-checks-before-closing-on-house/#:~:text=The%20last%20credit%20check%20before%20closing%20typically%20happens%20within%201,ability%20to%20repay%20the%20loan.

[14] https://www.uhloans.com/how-many-times-will-lender-pull-credit/#:~:text=This%20is%20known%20as%20a,out%20our%20Homebuyer%20FAQ%20page.

[15] https://www.redfin.com/blog/how-many-credit-checks-before-closing-on-house/

[16] https://guide.freddiemac.com/app/guide/section/5203.1

[17] https://www.newsweek.com/houses-sitting-market-longest-period-decade-2115455

[18] https://www.fico.com/blogs/fico-s-royalty-pricing-role-and-adoption-mortgage-industry

[19] https://www.fico.com/blogs/fico-s-adoption-and-pricing-mortgage-origination-market

[20] Tegus.

[21] https://themortgagereports.com/35800/guide-to-mortgage-closing-costs-what-average-mortgage-costs-are-and-how-to-keep-yours-low

[22] https://x.com/pulte/status/1925003620560298137

[23] https://x.com/pulte/status/1927362114894585944?ref_src=twsrc%5Etfw%7Ctwcamp%5Etweetembed%7Ctwterm%5E1927362114894585944%7Ctwgr%5E8832a7f908bc5db2cf5253a73792fcd98b223a16%7Ctwcon%5Es1_&ref_url=https%3A%2F%2Fwww.mpamag.com%2Fus%2Fmortgage-industry%2Findustry-trends%2Fpulte-still-not-happy-with-fico-hints-at-action-on-credit-pull-costs%2F537114

[24] https://x.com/pulte/status/1945885972375150768

[25] https://www.fico.com/blogs/statement-fhfa-s-interim-lender-choice-policy-announced-tuesday-july-15th

[26] chrome-extension://efaidnbmnnnibpcajpcglclefindmkaj/https://www.fico.com/en/resource-access/download/54221

[27] chrome-extension://efaidnbmnnnibpcajpcglclefindmkaj/https://www.fico.com/en/resource-access/download/54221

[28] https://www.fico.com/en/ficoscore10

[29] https://libertystreeteconomics.newyorkfed.org/2025/08/a-check-in-on-the-mortgage-market/

[30] https://www.newyorkfed.org/medialibrary/media/research/epr/2013/1212vick.pdf

[31] https://www.newyorkfed.org/medialibrary/media/research/epr/2013/1212vick.pdf

[32] https://www.newyorkfed.org/medialibrary/media/research/epr/2013/1212vick.pdf

[33] https://www.newyorkfed.org/medialibrary/media/research/epr/2013/1212vick.pdf

[34] https://www.newyorkfed.org/medialibrary/media/research/epr/2013/1212vick.pdf

[35] https://www.fhfa.gov/news/news-release/fhfa-announces-release-of-historical-vantagescore-4.0-credit-scores-by-the-enterprises

[36] https://www.freddiemac.com/perspectives/sonu-mittal/optimizing-loan-quality-in-a-new-market-dynamic

[37] https://www.urban.org/sites/default/files/2023-11/GSE%20Repurchase%20Activity%20and%20its%20Chilling%20Effect%20on%20the%20Market.pdf

[38] https://www.mpamag.com/us/mortgage-industry/market-updates/independent-mortgage-banks-urge-safeguards-as-fannie-and-freddie-eye-conservatorship-exit/548706#:~:text=The%20IMBs%20urged%20regulators%20to,combined%20into%20a%20market%20monopoly.

[39] https://www.mba.org/docs/default-source/policy/white-papers/21127_mba_imb_report_final.pdf?sfvrsn=4f7d570c_1

[40] https://www.mba.org/docs/default-source/policy/white-papers/21127_mba_imb_report_final.pdf?sfvrsn=4f7d570c_1

[41] https://www.mba.org/docs/default-source/policy/white-papers/21127_mba_imb_report_wb.pdf?sfvrsn=5401c2bd_3

[42] https://www.mba.org/news-and-research/newsroom/news/2025/08/19/imbs-report-production-profits-in-second-quarter-of-2025#:~:text=The%20average%20pre%2Dtax%20production,quarter%2C%20is%2040%20basis%20points.

[43] https://ncrc.org/mortgage-market-report-series-part-1-introduction-to-mortgage-market-trends/

[44] https://ncrc.org/mortgage-market-report-series-part-1-introduction-to-mortgage-market-trends/

[45] https://www.mba.org/news-and-research/newsroom/news/2025/04/17/independent-mortgage-bankers-post-net-production-profits-in-2024

• Apple Podcasts

• Spotify

For a more detailed writeup of MELI, please see our deep dive on MercadoLibre (MELI) – Amazon in the Amazon which we published earlier this year.

We hope you enjoy the Business Breakdowns episode on MELI and as always, we welcome any well-reasoned thoughts and pushback.

Subscribe now

Disclaimer / Disclosures

The information contained in this article is not investment advice and is intended only for wholesale investors. All posts by Bristlemoon Capital are for informational purposes only. This article has been prepared without taking into account your particular circumstances, nor your investment objectives and needs. This article does not constitute personal investment advice and you should not rely on it as such. This document does not contain all of the information that may be required to evaluate an investment in any of the securities featured in the document. We recommend that you obtain independent financial advice before you make investment decisions.

Forward-looking statements are based on current information available to the author, expectations, estimates, projections and assumptions as to future matters. Forward-looking statements are subject to risks, uncertainties and other known and unknown factors and variables, which may affect the accuracy of any forward-looking statement. No guarantee is made in relation to future performance, results or other events.

We make no representation and give no warranties regarding the accuracy, reliability, completeness or suitability of the information contained in this document. To the maximum extent permitted by law, we do not have any liability for any loss or damage suffered or incurred by any person in connection with this document.

Bristlemoon Capital Pty Ltd (ABN: 22 668 652 926) is an Australian Financial Services Licensee (AFSL Number: 552045).

George Hadjia and Daniel Wu are associated with Bristlemoon Capital Pty Ltd. Bristlemoon Capital may invest in securities featured in this newsletter from time to time.

Subscribe now

A blowout quarter for Meta

Meta delivered incredibly strong Q2 results and Q3 guidance that handily beat expectations:

• Revenue growth of 22% YoY beating consensus by 6%, and a stable 2-year stack of ~44% growth for the past four quarters.

• Operating income growth of 38% YoY and 500 bps of operating margin expansion on the back of ongoing cost controls and operating leverage.

• EPS growth of 38% YoY to $7.14, smashing consensus by 21% and similar to last quarter’s beat.

• Q3 revenue guidance of $47.5 to $50.5 billion or +17% to +24% YoY, beating consensus of $46.2 billion.

While P&L performance remains very strong, we see ongoing deterioration of cash results as capex continues to step up. Operating cash flow of $25.6 billion was largely absorbed by $17 billion of capex, resulting in just $8.5 billion of free cash flow. FCF margin is down to 18% and could potentially fall to 10% in 2026 given the capex step-up compared to 40%+ operating margin.

Source: Bristlemoon Capital, company filings

The 2025 capex guide was tightened slightly to a range of $66 to $72 billion representing an over $30 billion YoY increase at the midpoint. But the real kicker is the initial outlook for 2026, which is expected to step up to ~$100 billion! Yet even with this nearly 3x capex increase over two years, the company is still compute constrained per the CFO:

“But even with the capacity that we’re bringing online, we are still not quite perfectly meeting the demand that teams have for compute resources across the company.”

The biggest driver of the 2026 capex increase will be Gen AI instead of Core AI investments. When we spoke to the company last year, Core AI was still the majority of capex (2024 capex: $37 billion), and we suspect that this year the balance may flip. By next year, we expect the majority of capex to be for Gen AI initiatives, which don’t yet have a monetization timeline. While management still maintains that the capex is fungible between Gen AI and Core AI, our back of the envelope math suggests that at a quantum of $100+ billion capex every year (and growing), it may prove difficult to earn an adequate return on investment from Core AI advancements alone.

But before we consider the AI capex vs ROIC debate, it is worth highlighting how Core AI investments have been driving improvements across the business.

Core AI investments powering growth

For those who don’t follow Meta, Core AI is essentially the investments into the company’s recommendation and advertising systems that drive engagement and revenue across the core platforms. The improvements that stem from these investments are bucketed into two categories: engagement initiatives and monetization efficiency.

On engagement initiatives, the clearest indicator of success is the reacceleration of DAU growth over the past three quarters. Despite nearly 3.5 billion people already using Meta’s apps daily, DAU is somehow still growing at 6% YoY. This suggests to us that AI – whether it be through better recommendations or Meta AI itself – is driving higher daily engagement.

Source: Bristlemoon Capital, company filings

Other engagement improvements that were highlighted in Q2 include over 20% YoY growth in video time spent on both Facebook and Instagram, driven by enhancements to the recommendation systems. This comes on top of a 7% increase in time spent on Facebook and 6% increase on Instagram reported in the six months through Q1, which came on top of a 8%/6% increase in time spent on Facebook/Instagram in the nine months through Q3 2024. So there do appear to be strong incremental gains in time spent across the apps from better AI recommendations. The company is also making good progress on longer-term content ranking innovations such as cross-surface foundation models and extending LLM architectures into existing recommendation systems.

Subscribe now

We generally subscribe to Ben Thompson’s (of Stratechery) view that ad networks are at their healthiest when revenue growth is being driven by ad impressions rather than price per ad. Strong growth of ad impressions supports robust ad revenue growth in future quarters, while an over-reliance on eCPM growth tends to degrade advertiser returns over time. For the past few quarters, ad impression growth had slowed materially while price per ad grew double digits, but in Q2 there was a meaningful inflection in ad impression growth driven by rising engagement on the core platforms and some further ad load optimization on Facebook. Meta has also started inserting ads into Threads and the WhatsApp Updates page, though these won’t be material contributors to ad impression growth this year (or next, for WhatsApp).

Source: Bristlemoon Capital, company filings

On monetization efficiency, CFO Susan Li noted that the 9% growth in price per ad benefited from improved ad performance, suggesting that underlying ROAS continues to increase. Susan spoke extensively about improvements to ad retrieval and ranking systems, each of which drove 3% to 5% increases in conversion on Facebook and/or Instagram. During the quarter, Advantage+ was also rolled out as the default option for sales and app campaigns, and adoption has been higher than expected.

"The Andromeda model architecture we began introducing in the second half of 2024, powers the ads retrieval stage of our ad system, where we select the few thousand most relevant ads from tens of millions of potential candidates. In Q2, we made enhancements to Andromeda that enabled it to select more relevant and more personalized ads candidates, while also expanding coverage to Facebook Reels. These improvements have driven nearly 4% higher conversions on Facebook Mobile Feed and Reels."

"In Q2, we improved the performance of GEM by further scaling our training capacity and adding organic and ads engagement data on Instagram. We also incorporated new advanced sequence modeling techniques that helped us double the length of event sequences we use, enabling our systems to consider a longer history of the content or ads that a person has engaged with in order to provide better ad selections. The combination of these improvements increased ad conversions by approximately 5% on Instagram and 3% on Facebook Feed and Reels in Q2."

We think it is easy for investors to dismiss or underappreciate small improvements to engagement and ad efficiency, especially when viewed against massive capex investments. But an 8% increase in time spent could, ceteris paribus, translate into a $15 billion increase in revenue, and a stack of 3% improvements to ad efficiency could support a 10% increase in eCPM with no degradation to ROAS.

The example below shows how the cumulative benefits from Core AI investments can generate an attractive incremental pre-tax return on capital at current capex levels.

Source: Bristlemoon Capital

Subscribe now

Finding religion in superintelligence

Moving on from Core AI, it is clear that Mark Zuckerberg has found his new messianic calling: to usher in the Metaverse superintelligence. From our perspective, it is interesting watching his descent into an AI maximalist. Throughout 2023, Zuckerberg appeared content to do his own thing with AI (open source, product-focused, school project-ish etc.) and let the frontier labs compete for the state-of-the-art models. In early 2024, Zuckerberg changed his position and believed Meta needed to pursue AGI. Fast forward to 2025, he has been on a multi-billion-dollar binge to acqui-hire the best AI talent in a bid to build superintelligence. This seems to be spurred by his observation of the exponential rate of AI progress (so far).

"I think it's basically just what we're continually observing how this works, what the trajectory or the pace of AI progress has been. I think it continues to be on the faster end. And that I think informs a lot of the decisions, from everything from the importance and value of having the absolute best and most elite talent-dense team at the company, to making sure that we have a leading compute fleet so that the people here can do, so the researchers here have more compute per person to be able to lead their research and then roll it out to billions of people across our products, making sure that we build and drive these products through all of the different things that we do, which I think is one of the things that our company is the best in the world at, is basically when we take a technology, we're good at driving that through all of our apps and our ad systems and all that stuff.

So, yeah, I mean, we're just going to push very aggressively on all of that, but, at some level, yeah, this is—there's sort of a bet in the trajectory that we're seeing, and those are the signals that we're seeing, but we're just trying to read it."

As Meta shareholders, it is not clear to us why Meta needs to achieve “superintelligence” (whatever that means) at all costs. We understand that Meta has enormous consumer distribution and cash flow, but we don’t understand i) why a personal AI assistant for the median human needs to be a “superintelligence”; ii) how a superintelligent personal assistant can realize its full potential without access to email, calendar, encrypted messages etc; and iii) how a highly engaging personal assistant does not cannibalize time spent on Meta’s monetized surfaces.

Even as we give Zuck the benefit of the doubt when it comes to (eventual) monetization, we have to wonder what it might look like. With OpenAI hiring Fidji Simo to build out an advertising business, it is evident that subscriptions alone are insufficient to cover the cost of a freemium service where the substantial majority of users are non-paying. Yet Meta’s engagement-driven ad model is not well suited to conversational use cases – WhatsApp ads sit in the separate Updates tab, and Susan revealed on the follow-up call that the company abandoned plans to introduce Messenger inbox ads after initial testing.

We believe Meta will ultimately figure out a sustainable monetization strategy for “superintelligence”, but the $100 billion a year and growing capex burden adds a certain level of urgency. Investors are currently cheering Meta’s strong P&L performance and letting the enormous capex outlay slide. But should revenue growth disappoint, we imagine the market reaction could be reminiscent of 2022, especially if Zuck is unrelenting in his pursuit of superintelligence. As we mentioned earlier, even if the Gen AI investments are fungible with Core AI, we struggle to see how forcing $100 billion a year of capex into the core platform can deliver adequate returns.

Source: Bristlemoon Capital, Bloomberg consensus

Two observations here: 1) a low-teens post-tax ROIIC is not yet value destructive in the sense that it is likely still above Meta’s cost of capital, but presumably shareholders would prefer buybacks; and 2) obviously if the superintelligence investment was not meeting internal expectations, Meta should cut capex. Our biggest concern is that instead of rationally reducing capex, an irrational decision is made to throw even more money at the problem (in line with current LLM scaling laws).

Subscribe now

Disclaimer / Disclosures

The information contained in this article is not investment advice and is intended only for wholesale investors. All posts by Bristlemoon Capital are for informational purposes only. This article has been prepared without taking into account your particular circumstances, nor your investment objectives and needs. This article does not constitute personal investment advice and you should not rely on it as such. This document does not contain all of the information that may be required to evaluate an investment in any of the securities featured in the document. We recommend that you obtain independent financial advice before you make investment decisions.

Forward-looking statements are based on current information available to the author, expectations, estimates, projections and assumptions as to future matters. Forward-looking statements are subject to risks, uncertainties and other known and unknown factors and variables, which may affect the accuracy of any forward-looking statement. No guarantee is made in relation to future performance, results or other events.

We make no representation and give no warranties regarding the accuracy, reliability, completeness or suitability of the information contained in this document. To the maximum extent permitted by law, we do not have any liability for any loss or damage suffered or incurred by any person in connection with this document.

Bristlemoon Capital Pty Ltd (ABN: 22 668 652 926) is an Australian Financial Services Licensee (AFSL Number: 552045).

George Hadjia and Daniel Wu are associated with Bristlemoon Capital Pty Ltd. Bristlemoon Capital may invest in securities featured in this newsletter from time to time.