If you missed it, here is part one:

And part two:

On Amazon’s Q2 2026 earnings call, Andy Jassy gave the following project-level IRR breakdown:

If the demand isn’t there, we won’t spend the capital. For servers and networking equipment, on average, it takes a little less than 3 years to break even on that investment. The servers currently have a useful life of at least 5 to 6 years. And most of our AI capacity these days is being contracted for at least 5-year terms. That means that we’re driving significant free cash flow on the servers and networking equipment in the 2 to 3 years after we break even.

Mostly Borrowed Ideas has a great write-up on AMZN’s earnings that I highly recommend.

However, many people are not convinced, claiming these project IRRs are not that attractive if you do the math. Here’s one example:

The basic math is very simple and since I haven’t seen anyone do it, much less provide the spreadsheets, I did it here. If you spend $100 and have a payback of roughly 3 years with a 6-year useful life, you get like 25% IRR, given a reasonable discount rate for Amazon of 10% to 15%. It’s very positive NPV.

However, if you read the transcript you’ll see that datacenter spend is generally about two years before the company can start making cash flows on the project, while the majority of spend is for chips and networking equipment that are only purchased a few months before their project begins to generate cash flows.

If you assume that it’s roughly 2/3 of spend is for chips with a 6-month lead time and 1/3 is for data center with a two-year lead time, you get a weighted average of about one year. The math looks like this.

While an 18% IRR is pretty intriguing these returns are pre-tax and so this may lead to only low teens after tax returns depending on the rate. Accelerated depreciation from the OBBB makes these projects more favorable.

Amzn Capex Analysis 8

32.5KB ∙ XLSX file

Download

Download

I’ve provided a spreadsheet with some more scenarios if you think CapEx takes longer, if you want to be more realistic and use the mid-year convention, or if you think that the useful life on this networking equipment can be extended to the 7-8 years Andy Jassy mentions as a possibility.

It’s important to note that these projections are conservative, as they assume the entire project is funded with equity instead of a significant portion being funded with debt, along with the associated lower cost of capital. Additionally, the one-third of data center CapEx with a two-year lead time should probably be a weighted average one-year lead time, as you spend throughout the two years before the data center becomes a cash-generating asset.

If you believe Andy Jassy’s project economics, assume a 2-year lead time, and no mid-year convention the lowest return is you get is a pre-tax 14% IRR with cash flows coming over the next eight years. Andy Jassy also claims that the next batch of chips and networking equipment in the data center will have even higher returns, as they don’t have to cover the initial data center construction outlay.

Of course, the risks are:

• Tenants default on these five-year contracts

• Significant improvements in chip technology obsolete the current investments

• Future demand for data centers falls precipitously, either due to model efficiency improvements or less demand than expected, plateauing of models, more open-source models, etc. The hyperscalers are unable to earn good returns past years 5 or 6 because of a massive data center overbuild.

Overall, I think demand for AI is going to be very strong for the next few years, and the incremental returns so far have been very attractive for hyperscalers and look likely to continue while these conditions are in place. I’m sure many people will join me in monitoring these details, but for now I’m very bullish on the incremental economics.

Amzn Capex Analysis 8

32.5KB ∙ XLSX file

Download

Download

Disclaimer: This post is not financial advice, and is not a solicitation to purchase or sell any securities. For more info, please review the disclaimer.

After Q2 earnings, Meta dropped 8.5% the next day. While many bulls insisted that the market was overreacting and that results were actually good, I have to admit that the results were weaker than I expected, and many market participants summed up the negative sentiment surrounding Meta.

With Meta currently my biggest position, I want to take some time to articulate the Meta bear case to help clarify it for myself and to make sure that I understand the other side of the trade as well as possible.

I’ve separated the concerns into three categories: the specific quarter, the general AI / Capex trade, and then legal / other.

Quarter:

1. META revenues were just in line with consensus. We aren’t seeing accelerating growth, despite the accelerating spend in capital expenditures so far.

2. META expenses increased, 55% y/y . This led to reduced operating income y/y and was especially driven by a 67% increase in research and development costs. This expense is largely for unproven products and the company’s AI push rather than the core business.

3. Concerningly META COGS was also up more than revenue suggesting that incremental margins may be worse as META and the Meta ads system becomes more compute intensive.

AI / Capex:

1. Meta is shifting from the business model you’re familiar with: a social media platform with strong network effects, to an attempt to compete with leading AI labs and hyperscalers, committing hundreds of billions of dollars.

   1. Meta’s existing business seems to be in the pile of software that is easy to recreate with the help of new AI tools. While the social network has some core network effects, LLMs may revolutionize how we interact with technology and where we go to for entertainment. If AI creates personalized entertainment will people spend as much time on Facebook and Instagram?

   2. Part of the significant CapEx with uncertain future returns funds META entering other business lines with very established, well-funded, and fast-growing incumbents like Anthropic and OpenAI, as well as Open Source models like Kimi on the AI model development side, and then hyperscalers like Amazon, Google, and Microsoft. Despite Meta’s claims of potential optionality, they are yet to show revenue from any other uses for their compute, and enterprise sales is not a core competency.

2. In addition to all these factors above, Reality Labs continue to lose $10-20B /year. Meta spends aggressively on capex without providing much clarity to shareholders and without caring what they think. A bet on Meta is ultimately a bet on Zuckerberg at the helm and spending as he sees fit.

3. Capex actually understates the amount that Meta is committing to these projects. If you look at the growth in Meta’s off-balance-sheet purchase commitments and lease obligations, you can see a massive ramp-up over the past two years for future capex and performance obligations.

Meta Contractual Commitments And Lease Obligations 8

12.3KB ∙ XLSX file

Download

Download

Legal / Other:

1. Future litigation is pending, and while the specific damages are uncertain, the cost is very likely to be material according to the company’s own admission. This leads to significant headline risk in the short term and uncertainty, as well as the potential for truly massive judgments against Meta. If you extrapolate the recent Meta and YouTube lawsuit about addictive product design to 10% of the U.S. user base, $4.2M/person * 18.2M affected users for a $76T judgment which dwarfs any possible enterprise value. judgments just in the U.S. To compound this, Meta reported litigation and severance expenses that caused the company to miss earnings estimates this quarter as well.

2. If you read Meta’s legal proceedings it’s almost 10 pages long!

Overall, I think META’s core network effects remain strong, and you want a technology-obsessed founder at the helm of a cash geyser with strong potential adjacent businesses tied to their world-class distribution. The price you pay vs the current normalized operating income and growth, even adjusted for reasonable litigation settlements is very attractive.

For balance I’ll conclude with a couple of bullish tweets:

Concentrix has one of the highest normalized trailing free cash flows that I’ve ever seen: 35%+. However, given their lack of moat and substantial risk of AI disruption, along with their heavily levered capital structure and limited flexibility on their financial covenants, I do not have a position at this time.

What do they do?

Concentrix primarily earns revenue as an outsources / call center staffer. However, they also provide basic website design, project help, payments, integrations, etc. If these sound like things that AI is getting more and more effective at — you’re absolutely right! Concentrix stock is down about 60% in the past twelve months as the market worries that this business is dying. I share the market’s skepticism, and at the current price I agree with Ian Bezek’s piece: I am not buying the dip following week Q2 earnings.

Revenue Split:

• 25% retail, travel, and ecommerce.

• 25% tech and consumer electronics.

• 15% communications,

• 15% banking, financial services, and insurance.

• ~10% from healthcare with surprisingly the largest year-over-year revenue decline

What’s their moat?

Concentrix does not really have a moat. CNXC boasts that the average contract duration with their top 25 clients is 16 years and that they have 160+ of the Fortune Global 500 as customers, but this is a double-edged sword. Concentrix clients often value their large labor force, allowing them to scale up and down projects and global coverage. However, these giant companies are also the ones most aggressively spending on AI and seeking to generate ROI from it. Call centers, website setup, documentation, and integration are all low-hanging fruit in today’s world.

Who are their competitors?

Teleperformance (TEP FP) is the closest comp. TEP also has a global footprint, and a similar number of employees and revenue though TEP is headquartered in France. I don’t think either firm has truly distinguishing features. Per BBG, both firms trade ~4x EV/2026E EBITDA but 50% of Teleperformance’s EV is equity, whereas only 25% of Concentrix’s is, leading to a higher free cash flow to equity yield for Concentrix.

TTEC Holdings (TTEC) – Much smaller company with 1/5th the revenue and around 1/5th the EV but even more leverage and a smaller market cap. Their revenue is already declining.

TaskUs (TASK) –Another smaller outsourcing company with about $1.2 billion in revenue and a $1 billion EV, so less heavily levered than CNXC.

Other higher quality outsourcing firms that are more consulting rather than call center oriented include Accenture (ACN), Cognizant (CTSH), and Infosys (INFY).

What are the unit economics?

At its core, Concentrix is mostly a labor outsourcer, so this is a spread business on billable hours. Given the variety of contracts, customers, terms, etc., the best way to get a sense of you in economics is probably to look at the firm-level results. In 2025, the revenue per employee was around $21,600, with gross profit of $7,500, implying a cost of goods sold per employee in the $14,000 range. It’s probably reasonable to assume about $10,000 of that is wages to the employee, and average hours are around 40 a week, which tells you that Concentrix is hiring employees for about $5 an hour on average.

Adjusted EBITDA is about $3.8k, and free cash flow is about $1.2k per employee. Multiply by 450,000 employees, and you get roughly $560 million in free cash flow generated by Concentrix in 2025.

What does the future look like?

At this point, and for the foreseeable future, there’s still a need for employees that have passed certain regulatory certifications, for instance dealing with healthcare and billings claims, but healthcare is only about 10% of Concentrix revenue. Concentrix also claims to be helping companies adopt AI or review prompts for quality control, but the future trend of more automation and AI displacing these workers is likely to continue.

Organic growth has averaged low to mid single digits the last five to ten years, with the other half of growth driven by acquisitions. So far, organic growth hasn’t really turned negative, and the company is still guiding for flat to 1% organic growth this year, which is negative in real terms, after inflation, but still important as it means contracts haven’t been negotiated lower yet.

I expect a mid-single-digit revenue decline over the next five years, which, given the operating leverage on this model, suggests high single-digit to low double-digit declines in earnings. However, I don’t have insight into the company-level break-even and when this company potentially becomes persistently unprofitable.

What’s the valuation?

CNXC trades at a 35+% FCF to Equity yield and ~4x 2026E EBITDA. Price / Sales is 0.16x, the lowest multiple in its history. EV/Sales is 0.7x which is also the lowest multiple in its history. Clearly the valuation is low on historical numbers and the market is pricing in extreme concern.

What are the main risks?

Obviously, AI lowering the number of representatives a company needs and thus the number of employees Concentrix can earn a spread on is the number one risk. Revenue growth is very uncertain.

One potentially overlooked risk is that Concentrix has a 3.75x time gross leverage restriction. With about 4.6 billion of gross debt, that means Concentrix Evita must be at least $1.22B. While EBITDA was around $1.50B last year, it’s on pace to be about $1.42B this year. This means a 14% decline in EBITDA would trip this covenant, likely restricting any dividends or buybacks and requiring other concessions to the debt holders.

I’m surprised that these leverage restrictions are so tight. If this was a situation where I could buy Concentrix equity and expect to get 30-40% in free cash flow returned to equity holders while the debt is just perpetually rolled, I would likely be willing to make that investment and bet that the cash flows decline more slowly over time. Given the likelihood that cash flow is going to have to be exclusively prioritized towards paying down debt, I am much less optimistic about substantial returns to equity holders over time. Additionally, Concentrix’s overall EV is not much of an outlier compared to its comp set and may be reasonable.

Conclusion

Overall, I’m always tempted by something with a 30%+ free cash flow yield if it seems like those cash flows can persist. However, while the headwinds have not influenced business results much yet, the ominous clouds looming on the horizon plus a balance sheet overloaded with debt keep me as a bystander rather than a passenger on this flight.

Appendix:

Cnxc 2021 2025 Fcf Calculation

16.1KB ∙ XLSX file

Download

Download

Concentrix VIC write-up (2025)

Concentrix VIC write-up (2023)

Reddit Concentrix write-up (2023)

I’m in the midst of getting my roof replaced. During this process, I learned about the industry-standard Xactimate software that adjusters use to estimate damages and the insurance payout. Here’s a nice intro to using the software. I’m always interested in learning about profitable business niches, so today I’m spending one hour researching and an additional hour writing up Verisk Analytics (VRSK), the owner of the Xactimate software.

What do they do?

Verisk has multiple different software products and business lines, and it’s the result of many acquisitions over the past twenty five plus years, but it’s all focused on Insurance. The company started as a regulatory privileged way to standardize insurance forms, allowing regulators to review insurance products and make sure they were conforming. In 2009, it went public in a 100% secondary offering where the existing insurance company shareholders sold portions of their stake.

Verisk boasts an impressive 100% customer coverage in the top 100 property and casualty insurers in the US and has products for virtually every type of insurance.

They are a classic small part of the insurance company’s budget with significant pricing power and strong market share.

What’s their moat?

Verisk has strong network effects and data advantages because they collect data from the vast majority of the insurance industry. Their Xactimate software has localized price lists that are updated periodically with the market price for roof repairs, new windows, etc. This data is gathered locally and has economies of scale. It would be expensive and difficult for a competitor to acquire all this information locally.

For the same reason that newspapers tend towards duopolies or monopolies it makes sense for VRSK to spend $1,000 to have unique, up-to-date data in a market and charge $2,000 earning a 100% ROI. However, for a competitor to come in, they would have to spend $1,000 meaning a competitor would have to capture 50% market share just to breakeven. Given the dominant starting position of Verisk, the duration of their contracts, and their integrated product offerings, new competitors are generally not economically feasible.

The other strong evidence of a moat is their pricing power. As you can see in their investor presentation, they expect to raise prices by roughly 3-4% a year, virtually in perpetuity.

Who are their competitors?

Verisk claims a lot of competition in their segments, including insurance companies with in-house software. Notably, in their 10K, they do not list any company as an explicit competitor.

What are the unit economics?

With so many product lines, economics are very difficult to disentangle, but gross profits are consistent in the 65+ percent range, and operating margin has been 40+ since 2021. Management has mentioned that acquisitions and their higher-growing products tend to be slightly lower margin, but also margins are increasing with growth and scale.

What does the future look like?

I think management’s target of 6-8% growth looks very reasonable. I have free cash flow growing at around a 10% CAGR over the last 10 years, and I think 6-10% going forward seems very reasonable.

I’ll discuss AI briefly in the risks, but I don’t see AI agents significantly encroaching on industry-standard software, contracts, or data, as most of these have already been generated. AI may help VRSK be slightly more efficient in tweaking forms from one year to the next. However, given the trust, small percentage of a big product, and regulatory conservativism, even if I came in and gave away insurance forms or software for free that were of similar quality to VRSK’s, I think my market share would be approximately zero.

What’s the valuation?

I have a trailing free cash flow of $1,102 for a yield of approximately 4%. If you underwrite 6 to 8% growth in the future, that gets you to a medium-term shareholder yield in the 10 to 12% range. I’m really impressed by the quality of this company, so that may be reasonable, but given the slightly low organic growth, I would love for this to be a 5% to 5.5% free cash flow yield.

What are the main risks?

• Poor Acquisitions – VRSK had kind of a lost decade due to some questionable acquisitions outside of insurance and likely mismanagement. Verisk seems to have gotten religion and be pretty concentrated in their specific niche, but I would keep an eye on future acquisitions.

• AI Risk – VRSK has traded as an AI loser especially the last 12 months. Given their market dominance, strong network effects, and winner-take-all dynamics, I don’t see it. I would love to hear better arguments espousing significant AI risk in the future, but I think this is quite low probability.

Conclusion

Verisk is a sprawling collection of insurance-related products, most with seemingly dominant market share and persistent pricing power. Management has articulated a target of 75% of free cash flow returned to shareholders, and I think this is a strong candidate for a quality compounder stock at a 4% free cash flow yield. That’s a very reasonable price for a high-quality company and if we see a continued sell-off due to AI fears, this can be a very compelling investment.

Appendix:

Verisk - VIC writeup

ISO history (precursor to VRSK)

VRSK Wikipedia Overview

Vrsk Investor Day Presentation March 2026

9.21MB ∙ PDF file

Download

Download

Vrsk Investor Presentation June 2026

1.31MB ∙ PDF file

Download

Download

Vrsk 2016 2025 Fcf Calculation

18.4KB ∙ XLSX file

Download

Download

Vrsk Chatgpt Research Report 7

92.1KB ∙ PDF file

Download

Download

TL;DR PGY is an attractively valued but opaque and speculative fintech with moderate network effects and the potential for mid-teens growth. Pagaya also has 20%+ short interest, a significant warning sign, so size accordingly.

Correction: When I calculated normalized adjusted FCF I failed to subtract SBC (Stock-Based Compensation). After subtracting SBC I get ~$60M in normalized FCF instead of $110M and estimate about $80M for 2026. That gets PGY owners a 5+% FCF yield that I think grows 10-15%, but is significantly less than the 8% I initially calculated. The valuation section below has been updated.

Pagaya has been one of the most challenging businesses for me to analyze. It’s fundamentally difficult to assess a financial firm’s underwriting, especially over a limited period of time or in the midst of credit cycle and model changes.

There are a lot of items that I would like greater disclosures on including historical credit quality and charge-offs by vintage, fee splits, their managed investment funds, VIEs, and go forward risk retention requirements to name a few. If you have thoughts I’d love to chat!

What do they do?

Pagaya started by partnering with banks and other financial institutions (e.g. US Bank, Ally, and SoFi) to offer loans to consumers that would otherwise be denied by the bank. Instead of the bank rejecting the loan, they give Pagaya the option of making the loan through the bank’s infrastructure, and the customer never knows the difference. Pagaya lends primarily in three segments: Personal, Point-of-Sale (POS), and Auto.

• Personal loans – are simple unsecured loans to consumers with an average income of $115k.1

• Point of Sale loans – are Buy Now Pay Later (BNPL) loans when consumers are checking out.

• Auto – offers loans at the time of purchase to used and/or new car buyers.

Pagaya seeks to be an intermediary offering attractive financing options. They partner with the financial institution that has a relationship with the customer allowing institutions to capture a fee without taking any credit risk on the loan. Pagaya, in turn, repackages the loan into an asset-backed security (ABS) or other financing vehicle and sells that to investors like hedge funds, credit funds, insurance companies, and pensions.

In theory, Pagaya should have limited credit risk as they seek to retain little if any of the loan on their balance sheet. In practice, Pagaya is often mandated to have some risk retention tranche and has historically lost money on the investment securities they’ve been forced to hold for risk purposes. Recently, these losses have moderated, and Pagaya is tightening their underwriting to bring these closer to profitability going forward. Pagaya has also improved their unit economics, and in theory, even if the entire loan goes bad, Pagaya’s losses should be limited to the revenue they made.

What’s their moat?

Pagaya claims to have extended credit to more than 1 million Americans and evaluated applications for more than $3.7 Trillion in loans since inception. Pagaya argues this gives them a data advantage where they use their machine learning models to more accurately segment and offer credit to borrowers. For example, PGY might be willing to extend credit to borrowers with high income but a fledgling credit score, such as recent college graduates, or borrowers with prior bankruptcies that are now being responsible and paying their bills on time. Pagaya has also expanded beyond their “decline monetization” model, lending to consumers the bank would otherwise reject, which now only represents about 50% of their revenue.

Pagaya benefits from network effects. As they partner with more financial institutions, they can enable more loans and use more data to assess consumer credit conditions and borrower behavior. PGY is also growing their investor network, and are the largest US personal loan ABS (Asset-Backed Security) issuer allowing them to get more attractive financing terms. In Q2 2025 they received their first AAA-rating on an auto ABS, and continue to reduce concentration risk.

Pagaya hopes for the rare quadruple win:

1. Win by helping financial institutions increase the lifetime value of their customers by earning fee income even on loans that don’t make the balance sheet.

2. Win by helping consumers get access to credit at better terms or when they would have been rejected entirely.

3. Win by helping investors access attractive ABS securities.

4. Win by earning a fee for facilitating all the above.

Who are their competitors?

Anyone offering credit instead of using Pagaya to make the loan is a potential competitor. Pagaya has mostly partnered with financial institutions and sought to be incremental through their “Decline Monetization” program where rejected loan applications are sent to Pagaya or by offering credit instead of the financial institution. Pagaya has branched out to offering loans when the bank doesn’t want to deploy their balance sheet (First Look) as well as helping banks market and make more loans (Affiliate Optimizer Engine and Direct Marketing Engine).

Pagaya’s partnerships with other financial institutions must be win-win otherwise the partner would terminate the transaction. However, as Pagaya’s network grows stronger, they have more bargaining power in the relationship. As they have more partners competing against each other, rival banks have to worry about customers getting a Pagaya enabled loan on a different platform and Pagaya is able to negotiate a more favorable split of fees.

Recently, Pagaya’s relationship with Klarna has taken a turn for the worse. Formerly, Pagaya listed Klarna as their anchor POS partner but in Q1 of this year Klarna cut Pagaya out, deciding to vertically integrate. Pagaya, in turn, is suing Klarna, claiming that Klarna misappropriated trade secrets and other confidential information. We’ll discuss this more in the section on litigation below.

What are the unit economics?

Pagaya claims to be very selective, approving less than 1% of loan applications. However, their slides also indicate only 1% of offers are accepted by borrowers. I have reached out to investor relations to clarify this, and the risk for adverse selection is much higher if only 1% of their loan offers are accepted versus them intelligently subsetting the banks rejections and finding the incremental 1% of borrowers worth lending to.

When Pagaya makes a loan, they often charge the borrow an upfront fee, the origination fee, in the 7-10% range. Pagaya pays 60-80% of this fee to the financial institution keeping the remainder for itself. The rest of the fees on the lending product side (light blue below) are charged as servicing and maintenance fees over time. Pagaya generally seems to keep a greater percentage here although they don’t disclose the weighted average split. Overall, these loans are short term and high interest rate with an average 2-3 year duration and ~20% Weighted Average Coupon.

Pagaya’s net revenue, the money they keep after paying financial institution partners, is called FRLPC or (Fee Revenue Less Production Costs). This is the key revenue metric the company highlights.

After Pagaya makes these loans they then sell them to investors. As long as Pagaya makes good loans they should recognize a small gain on sale when they sell them in the market. Pagaya also earns some ongoing maintenance and service fees for managing the ABS securities. However, thanks to the Dodd-Frank Act Pagaya is required to maintain a portion of the ABS securities, often the riskiest tranche. At the end of Q1 they held $489M of required risk retention securities down from $504M at year end. This is an ongoing cost of business, and I estimate about $155M in losses a year from these securities, which is about 15% of the total balance or 30% of the required risk retention securities.

While 15-30% charges might sound aggressive, and $155M is more than the $107M of losses on investments in loan and securities they reported in 2025, this is often the junior tranche most exposed to charge-offs, and management provided key unobservable inputs that include a loss rate around 16%.

Over time Pagaya has been successful in negotiating the fee split with their lending partners to reach a 4 to 5% revenue (FRLPC) take rate on their network volume, the loans they facilitate.

What is the litigation situation?

Pagaya proudly pointed to longstanding customer relationships to show that their network was win-win rather than cannibalizing lending fees from partnering financial institutions. Then on March 25th of this year “Klarna, Inc. notified Pagaya that it would cease to operate under the License Agreement and the Commitment Agreement, purporting to terminate them on 180-days’ notice, and stated it would not enter into any new financing agreements with Pagaya.” - PGY redacted complaint.

Pagaya was taken off guard with only an inkling this was happening starting in February 2026. Pagaya responded with a complaint linked below, accusing Klarna of misappropriation of trade secrets, unjust enrichment and other contractual violations. Klarna is scheduled to file a response by July 15th, and then Pagaya may file an additional response, and a trial may commence from there. Pagaya has reiterated their guidance since this March 25th cancellation of the license agreement by Klarna, and I think that Klarna has declined in significance as a partner for Pagaya in the last year or two as auto volume has grown more rapidly than point-of-sale volume.

Financial institutions canceling their partnership outright is always a risk. However, as long as Pagaya is incremental, either in enabling loans that the banks would otherwise make or providing a valuable fee stream to financial institutions that doesn’t tie up their balance sheet, Pagaya should still be able to maintain win-win partnerships. It’s unlikely that a bank will be willing to make every loan that Pagaya is willing to make and always have balance sheet capacity to spare.

What does the future look like?

Management targets 2-4 new partnerships a year, adding 4 in 2025, in addition to revenue expansion with existing relationships. Right now, management is excited about several US regional banks they expect to become new partners by the end of the year.

Pagaya’s FRLPC, net revenue on the transaction, is likely to remain in the 4-5% range in line with management guidance and what they have achieved historically. Network Volume has grown at a 21% CAGR since 2021 and I estimate this will grow at 10-15% over the next 5 years. Pagaya has shown impressive operating leverage with SG&A expenses declining in 2025 versus 2024 despite revenue growing 20%. I wouldn’t count on further operating margin expansion here, but SG&A can remain fixed as a percent of revenue. This is conducive to cash flow growing at double-digits or better.

What’s the valuation?

Pagaya has a complicated capital structure. The simplest answer is an EV of $1.3B if you give them full credit for the Investments in Loans and Securities at Fair Value. Given the poor historical performance I’d haircut that number by an additional 25% for an EV of $1.5B. I am also not giving PGY credit for their $150M of fees receivable or other assets which should reduce the EV by another $100M. I’m attaching the FDSO workbook in case you want to make your own assumptions.

Pgy Fdso 7

17.1KB ∙ XLSX file

Download

Download

I estimate normalized Free Cash Flow around $110M in 2025 and I expect $120M+ for 2026. I estimate normalized Free Cash Flow around $60M in 2025 and I expect ~$80M+ for 2026. That gets Pagaya owners a 5+% FCF yield that I think is growing 10-15% a year.

What are the main risks?

• Continued losses from risk retention securities. Bulls argue that the prior losses were mostly a result of the sharp rise in interest rates and that Pagaya has tightened their underwriting, something they’ve continued in 2026. However, if their losses return to prior levels, then the company will likely need to raise equity and this could be existential.

• Economic downturn leads to poor credit performance and reduced go-forward investor ABS demand. Even if Pagaya’s underwriting is improved, the cyclical nature of credit markets could saddle them with significant losses when the economic cycle turns. Pagaya hasn’t lived through a full credit cycle yet, and investors may not be willing to underwrite these loans in the future.

• Funds managed by Pagaya suffer losses and outflows, leading to significantly reduced revenue. Related party revenue was $652M or 52% of revenue. This is flat in dollar terms but down from 81% of revenue in 2023. If there are significant redemptions and Pagaya is not able to replace these funds, then Pagaya may become unprofitable again. While I wish the company offered more disclosure here, this is not a growth area and seems to be declining in importance over time.

• Like Klarna, customers vertically integrate, cutting out Pagaya. This is the risk I’m least worried about, as I think Pagaya is incremental for most financial institutions, even banks that have their own credit underwriting. However, if we see more customer defections, then that would be a big warning sign.

Conclusion

Overall, I’m impressed by the network effects and potential quadruple win. Shareholders could earn 20%+ 15%+ annualized on the current FCF yield plus growth prospects. It’s not just management puffery, the historical financials show massive improvements: taming investment losses as well as slashing SG&A overhead. Their dramatic inflection to gap profitability is a powerful signal. However, I have concerns about future credit performance and lending is almost always a black box. Pagaya’s scale, financial transformation, and value proposition give it a right to win. While their are many risks, I think the future shareholder earnings are some of the most attractive on the market today.

Appendix:

M. V. Cunha's Substack

Pagaya Technologies (PGY): Investment Thesis Explained

Just before the 4th of July weekend, I asked you for stock ideas to dig into during the break…

Read more

a year ago · 98 likes · 7 comments · M. V. Cunha

Convequity

Pagaya: From First-Loss Pain to Forward-Flow Discipline

SUMMARY…

Read more

9 months ago · 6 likes · 9 comments · Convequity

Iceberg Research Short Report part 1 (2/11/25)

Iceberg Research Short Report part 2 (2/18/25)

Pagaya Short Report on VIC (11/7/24)

Pagaya skeptical analysis (2/21/25)

Pgy 2023 2025 Fcf Calculation

15.6KB ∙ XLSX file

Download

Download

Pgy Historical Financials 5y Chatgpt 7

45.1KB ∙ XLSX file

Download

Download

Pagaya Kpis Chat Gpt

44.7KB ∙ XLSX file

Download

Download

Pgy Network Volume And Frlpc By Vertical 2021 2025 Chatgpt 7

49.3KB ∙ XLSX file

Download

Download

Pgy Klarna Lawsuit Redacted Complaint 5

1.23MB ∙ PDF file

Download

Download

If you missed it, here’s Part I in this series analyzing META :

As you’ve no doubt seen big tech is spending hundreds of billions of dollars a year on AI. This is the fundamental driver for NVIDIA’s dominance and ascension to the most valuable company on the planet. While these capital expenditures are widely talked about, there are many concerns that the transition from relatively capital-light to capital-intensive business models will cause a huge bubble and shareholders will lose lots of money.

Alphabet (I’ll call them Google from here to match the stock ticker) is doubling their capital expenditures this year and forecasting a significant increase next year as well. They’re raising cash from equity and debt sales to invest even more aggressively in AI. $300B of Capex in one year seems possible. While no one knows the return on those future outlays, Google’s financials give us a clue as to how these investments are likely to perform. I think Google is likely to continue to earn strong returns, but given the amount of capital being deployed, particularly in Google Cloud, the hyperscalers like Google are more exposed to a pullback in AI demand than other companies like META.

Google has increased their invested capital by ~60% in the past 12 months. Google highlights that their 2026 expected capital expenditures are six times greater than 2022 Capex. 6x in 4 years. wow. Estimates are for Google Cloud to grow 60+% to $96B in Revenue for 2026 and then more than 50% to $145B in 2027. I estimate Google Cloud growth is responsible for around half of Google’s 2025 Capex and more than 1/3rd of the massive $180B+ in Capex they’re forecasting for this year.

Before we speculate about the future, let’s take a step back and look at Google’s historical financials in the table below. We see a spike in operating income particularly in year two of Covid (2021) without a corresponding increase in invested capital, leading to great ROIC for Google. Afterwards, there was a digestion period before Operating Income and Invested Capital began to ramp in 2024 with AI, Gemini was officially released in December 2023. In recent years, the ROIC has fallen but still remains excellent and above 2019 levels.

In 2025, incremental ROIC was a relatively anemic 22.8%. To be clear, this is still a pretty phenomenal result. If you can do better than 22.8% before taxes on your corporate project, you should probably do as many of those as possible! However, this is a dramatic decline from 75% incremental ROIC in 2024.

Part of the reason for this decline in ROIIC (Return on Incremental Invested Capital) is that it’s taking longer for Google to build new data centers and other construction projects. From 2011-2019 Google’s construction in progress was 70-75% of that year’s Capex suggesting it took about 9 months to get capital projects operational. In the last 3 years, the average construction in progress as a percent of capital expenditures was around 100%. This shows delays or longer capital project timelines as well as the continued ramp up in capital expenditures.

To adjust for this we can attribute this year’s increase in operating income to last year’s investments as they mature and begin to generate returns. Google’s lagged incremental ROIC increases to 45%. This is still below the historical average for Google. The true incremental ROIC likely lies somewhere in the middle between 22% and 45%. In Q1 2026, incremental (non-lagged) ROIC accelerated back to almost 40%, suggesting that Google is still seeing excellent returns on their capital projects.

As mentioned earlier, the big news in the past couple of weeks is that Google made a huge departure from their historical capital allocation and issued $80 Billion. I recommend reading this post for some more context and analysis:

Yet Another Value Blog

What is $GOOG seeing in AI? Part 1: history's biggest, weirdest equity raise

I was shocked by Google’s big capital raise (and the rumors of META’s big potential raise that ended the week…

Read more

2 months ago · 48 likes · 2 comments · Andrew Walker

With Google’s capital expenditures ramping up, no one knows what the incremental ROIC will be, as we are 12-24 months out from those expenditures beginning to generate revenue. Google has been very clear that they’re striving to maintain ROIC discipline, and keeping that in mind along with the potential for them to use all these capital investments for their own internal products like YouTube search as well as DTC and B2B AI offerings if third-party demand falls through.

Unfortunately, Google doesn’t really break out which segments they’re investing substantial incremental Capex in and what those returns are. YouTube remains very capital light, and the improved recommendation engine is probably extremely high (100%+) ROIC. In contrast, Google Cloud ROIC may be in the teens or lower if we see a slowdown in third-party demand for cloud compute.

If you estimate a 30% ROIC for Google Cloud and a one-year lag between expenditures and generating revenue then roughly $50B of 2025’s 91B in Capex was for Google Cloud and another ~$65B of this year’s 180B is for Google Cloud.

Overall, Google has one of the best businesses in the world, and they have pivoted aggressively to include AI overviews and summaries in their searches, allowing them to claim “Google Search is delivering AI to more people than any other product in the world.” I’m optimistic that Google will see strong returns on their incremental capital expenditures here, and I think they were smart to proactively raise capital and get the public blessing of Berkshire Hathaway.

However, I’m less confident in the future returns and ability to repurpose this hardware if third-party demand fails, then I am from META. At the same time Google trades at a much richer valuation than META. I still like Google, but in a pullback the hyperscaler business model is at much greater risk of commoditization. Consequently, I like Google but have a heavier weighting in META at this time.

Appendix:

Google call explaining the equity raise and AI Capex opportunity:

GOOGL June Investor Presentation Transcript

GOOGL June Investor Presentation

For at least a decade, I’ve heard claims that the market is broken, indexing is ruining investing, and value investing no longer works. People complain that the market is dominated by passive flows, or investors aren’t paying attention. “Stocks that are cheap just get cheaper, as if the market has no regard for proper intrinsic value.” In this environment, how can a value investor make money buying underpriced (hated) stocks?

Despite the naysayers, value investing lives on. The financial value of a company to shareholders is the present value of all future cash flows that the shareholders will receive. If management acts faithfully on behalf of shareholders (a big if) then we don’t need to worry about the cash flows the shareholders receive, and can instead focus on the future cash flows of the company. The financial value of a company is the present value of all future cash flows. As Warren Buffett says “Growth is part of the value equation.”

Imagine you find a company, ValueCo, trading at ten times earnings, but NormalCo and all the comparable companies trade at fifteen times earnings. You believe this company is 50% undervalued and should trade up to 15 times earnings. But it never does. Assuming the market will instantly re-rate is a rookie mistake. What do you do? To simplify we’ll assume earnings remain perfectly flat each year for this company and the comparable companies.

In year 1, with no change in multiple, ValueCo earned a profit of $10/share, and with no growth opportunities paid the $10 to shareholders. Shareholders earned a 10% return on their investment (10x $10/sh = $100 initial share price, $10/100 = 10%).

NormalCo also earned $10/share, and with no growth opportunities paid out $10/share to shareholders. NormalCo shareholders earned $10/150 or 6.66%. If the multiples remain constant, rather than earning 50% on the re-rating of ValueCo to be in line with peers, ValueCo shareholders will just earn 3.33% more annually in perpetuity.

Quick aside, I'll generally use the term "earnings" for simplicity here but what we're really talking about is “owner earnings” or the free cash flow the company generates each year.

More generally, if earnings are flat in perpetuity then the shareholder return is the earnings yield of the company.

When valuation multiple is constant and growth equals 0:

Now imagine a more complicated scenario. We will keep the same earnings per year for both companies of $10 per share and the same multiples: 10x for ValueCo and 15x for the NormalCo, giving us a price of $100 per share and $150 per share respectively.

Let's assume that each company is able to grow their earnings at 10% a year. However we will also assume a return on invested capital of 10%, meaning that all earnings are required to be reinvested back in the business. In this case, as long as the multiple remains constant, the returns to shareholders will be identical whether they buy cheaper ValueCo trading at 10x or the more expensive NormalCo trading at 15x.

The general rule is that if all earnings are being reinvested back into the business and the multiple remains constant, then the shareholder return is just the growth in earnings each year.

When valuation multiple is constant and all earnings are reinvested in the company:

Putting these two together we get the return equation. If multiples and growth remain constant, shareholder return is earnings yield plus growth. This is the most foundational equation in investing.

So how does this apply to hated value companies? Ultimately, every stock is fully owned at every moment. For the multiple to go up, the current owners or new owners must think the company is worth a higher multiple than it currently trades at. For the multiple to decrease, the current owners must reduce the multiple they think the company is worth paying for, and any incremental buyers must agree.

If you buy a hated, low multiple, company with good cash flows and growth, then one of three things can happen:

1. The hated company can become less hated, and you benefit from multiple expansion. This is the ideal case, and most value investors assume this should always happen. The majority of current complaints about value investing are when this doesn’t happen.

2. The multiple remains constant, and your returns merely collapse to the returns of the underlying business. Another way of thinking about this case is that shareholders earn the returns they would get from buying the entire company at the current price and never selling.

3. The company becomes even more hated and the multiple declines, which is often caused by other investors having a more pessimistic view of the future for this company. If the company’s underlying fundamentals persist, the future returns are even better.

Case three is the “bad” case and often called a “value trap.” What’s to stop a company from perpetually being more hated even while they have good earnings and growth prospects? As long as the earnings persist, case three is only temporarily bad and leads to higher future returns. Eventually the business shines through and investors are richly rewarded.

A quick example: ValueCo trades down from 10 times earnings to eight times earnings. Assuming those earnings remain constant (0% growth) the earnings yield increases to 1/8 or 12.5% from 10% increasing the return that shareholders earn going forward.

Ultimately, the backstop is case number two: if the earnings yield plus future growth is attractive enough, there will be significant pressure to purchase the company, whether that’s by private equity, a management buyout, other companies that see the attractive financial returns, or even the company itself repurchasing its own stock in the market. The catalyst for value realization is the distribution of capital back to shareholders or a change in control where earnings are distributed back to the new owners.

Value investors stereotypically focus on the current valuation multiple, which in the return equation is the earnings yield; however, it’s important to also consider the future growth of those earnings. If earnings are fully reinvested, then returns are driven entirely by growth. True value investors consider the full equation and forecast the returns from owning the business. If the earnings persist and management is responsive to owners, then value investing works and there is no such thing as a value trap.

Appendix:

Value Investing Examples

15KB ∙ XLSX file

Download

Download

Big tech is spending hundreds of billions of dollars a year on AI. This is the fundamental driver for NVIDIA’s dominance and ascension to the most valuable company on the planet. While these capital expenditures are widely talked about, there are many concerns that the transition from relatively capital-light to capital-intensive business models will cause a huge bubble and shareholders will lose lots of money.

META is one of my favorite positions at the moment and is currently being punished for a huge ramp up in capital expenditures (Capex) that management thinks is necessary to compete and grow in this AI boom. While the stock price has come down, Meta’s growth continues to be very strong. In this most recent quarter, Meta grew revenue 33% y/y. However, this doesn’t tell us whether the huge increase in capital expenditures is a good investment so I did some analysis.

The key metric is Return On Invested Capital – ROIC for short. ROIC is typically defined as:

EBIT or operating earnings is the money a business earns from its ongoing operations, after all variable costs, overhead, etc. but before any financing or tax considerations. Invested Capital is all the money invested into the business by people who have claims on cash flows from the business.

If I spend $100 on a lemonade stand and make $20 before interest and taxes each year, then I have a 20% (20/100) return on invested capital. If I’m considering opening another lemonade stand with similar returns, I can compare this to the return I would get from other projects like investing in the stock market or opening a hot dog stand.

People’s standards for a good ROIC can vary, but financially a bigger number is always better, and generally big tech companies have been characterized by impressively high (at least 20%+) returns on invested capital in their business. The downside, if you can call it that, is that these companies have been so profitable that they can’t find good investment opportunities to reinvest all their cash each year!

If we look at this table, we can see META’s invested capital has doubled in two years, from 2023 to 2025. Also, while the ROIC is still excellent it is trending downward and has fallen significantly from 2019’s 92% level.

With META planning to spend $135 billion this year, growing their existing invested capital by more than 50%, investors are nervous about the prospective returns. $135B in 2026 is almost double the $70B META spent on Capex in 2025. Visions of Reality Labs’ $80B+ of losses and counting are swimming in front of investor’s eyes.

A more careful analysis looks at the incremental invested capital. Perhaps META was a good business in the past, but is now wasting lots of money on bad ideas, and so the company’s overall return on invested capital is going down.

When looking at incremental invested capital, it’s important to remember META must spend money first, and then later on the equipment is put into production and actually earns a return. Meta is buying chips from NVIDIA, buying land, constructing buildings, purchasing servers and finally turning on the actual data centers. The project has to be powered up and operational before META can start earning money from these assets. Generally the lag time from expenditures to making money is about one year.

One way we can corroborate this one year time lag is by looking at META’s Construction in Progress, which is money they’ve spent (Capex) but the project is not finished and thus not producing any return yet. $61B is ~90% of the $70B META spent on Capex in 2025, or roughly 10-11 months of Capex.

We analyze incremental return on invested capital by looking at how much META increased their invested capital in a time and then see how much META earned a year later. On $25B of Incremental Invested Capital that META invested in 2024 they earned 14B in 2025 for a 56% ROIC. While lower than the past 5 years, this is still a great return. At these sort of rates, I hope META spends all their money on Capex!

Turning to the most recent quarter, we can see that in the 12 months ended 3/31/25 Meta invested $32B and then earned $5.3B in incremental operating income in Q1 26 for an annualized return of 67%. If you just look at Q1 25 Capex, they invested an incremental $10B that quarter, and due to the aforementioned $5.3B in incremental operating income for Q1 26, earned 55% on just Q1 25 Capex.

Overall, no one knows what ROIC META will earn on their current Capex. Most of the past year’s Capex is still Construction in progress and not earnings revenue / profits. However, META has a long track record of excellent returns on invested capital and the recent trends in incremental ROIC are both 50%+ and above the overall ROIC of the company! Finally, if Capex proves to be disastrous (generating very low ROIC), I am not overly concerned. This is discretionary growth Capex that META could quickly halt if they aren’t seeing returns. However, you need to believe that META won’t just incinerate hundreds of billions regardless of the economics, and given their Reality Labs history that can be difficult for investors.

Appendix:

Here’s a BBG enabled spreadsheet with some of my ROIC analysis if you’d like to dig in further.

Meta Incremental Roic Analysis

104KB ∙ XLSX file

Download

Download

This morning POET Technologies (POET) released a press release announcing an order for up to $500 million of their semiconductor components. The initial headlines look promising and talk about “a strategic joint development and commercial technology partnership to advance a new class of wafer-level photonic integration for frontier AI infrastructure.”

However, the details show that Poet is paying Lumilens in warrants, the right to buy shares, to induce Lumilens to buy components from POET. POET is offering 22.9M warrants, 15% of the entire company, to Lumilens if they’ll buy up to $500M of optical components in the future. How much are these warrants worth and how much is Lumilens on the hook for net?

Yesterday POET shares closed at $14.37. The warrants are for 9 years and have an exercise price of $8.25. Plugging into a basic Black-Scholes Calculator and you get a value of $13.45 per warrant.

At POET’s current stock price, assumed to be $19 as it’s been volatile today, these warrants are worth $17.95.

10% of the warrants, 2,292,140, are currently exercisable with a current market value of $40M. For Lumilens this is a classic Heads I Win, Tails You Lose transaction. They get paid $40M upfront with no obligation to buy anything. All purchases are “subject to the … ultimate qualification of the modules.” If Lumilens does end up buying anything, they get approximately 80% off due to the warrants they received.

For further details I’ve attached a spreadsheet showing how the warrants are valued and an even more conservative warrant valuation that is still more than 50% of the purchase price for these components.

In conclusion, this morning POET paid a little over $25M in warrants to issue a press release with Lumilens. These warrants are now worth $40M out of the $50M initial purchase amount. This press release helps make up for their recent disastrous cancellation with Marvell Semiconductors and is a significant step up from the $5M order they announced in October 2025 for delivery in 2H 2026. However, POET is paying at least half, 250M, of the $500M headline number to get Lumilens to agree to this order / partnership. At current prices, POET is paying over $400M (80%) of the cost of components. But a $100M purchase order wouldn’t pump the stock the same way.

Buyer beware.

Poet Purchase Analysis 5

674KB ∙ XLSX file

Download

Download

This weekend I had the pleasure of attending Berkshire Hathaway’s annual meeting. Many people rightfully look up to Warren Buffett as one of the greatest investors of all time. It’s common to hear the refrain that if you’re young, smart, and interested in investing, you should do what Buffett did in the 1940s and 1950s and look for small-cap companies. After all, Warren claimed to be able to compound at 50% a year if he were only managing $1 million (2024).

Many emerging managers have a small or micro-cap focus and claim that these companies are underfollowed and present a better pond from which to fish – resulting in better returns.

In contrast, the Russell 2000 small cap index has underperformed consistently since 2016, and cumulatively since the Russell 2000’s inception in the 1970s.

Most academic literature reports small cap outperformance over the past 100 years, and perhaps they will again, but I think this underperformance is structural. I predict the R2K will underperform the S&P 500 over the next 10 and 20 years. Here are some of the reasons:

1. Companies don’t stay small on purpose. If a company is small and has been around for a while, there’s some limiting factor to growth. It could be: return on invested capital is not very good, the business is very cyclical, competitors are constraining growth, the end market is declining, or poor management. Whatever the reason, quality companies in growing markets get bigger over time. If the company has been public for 10+ years and is still a small-cap there’s likely a problem with the company.

2. Public company costs are greater now. Thanks to SOX, other regulation, and the increased prevalence of shareholder lawsuits (everything is securities fraud), public company costs have gone up. If you’re a very frugal public company, you might be able to minimize these to the order of half a million dollars a year or so. Most of the time, though, I hear $1M to $3M as the typical small cap public company cost (e.g. ChatGPT link).

3. Adverse selection has increased. In the fifties and sixties, when Buffett got started, the stock market was relatively sleepy. Investors had just weathered the Roaring Twenties, then the Great Depression and 20 years of turmoil and generally poor returns. Now, markets are popular! There are millions of retail investors excited about the market, and tens of thousands of finance firms. There are TV channels dedicated to markets. The S&P 500 has compounded at 15% over the past 10 years.

   Most importantly, since the junk bond revolution in the 80s, private equity has exploded in number of firms and AUM. Virtually any small cap you’re looking at has been evaluated by private equity, and for one reason or another they have passed. Management is also much more able to leverage outside capital, and so a deep value company that is truly cheap is likely to be the subject of an MBO.

4. The internet has democratized investment research. Even little-known small caps often have write-ups on Value Investors Club, Substack, etc. And with the global nature of markets, millions of people from all over the world can be looking through the same roughly 10,000 U.S.-listed companies.

In addition to these structural factors mentioned above, it’s worth remembering that even in public market transactions, every trade has a counterparty. If you’re buying an ‘undiscovered’ stock, someone already owns it and is selling it to you! In 2026, we have enough history of people successfully boring, stable companies at 6-8x times normalized earnings for management, directors, customers, suppliers, and many others to be aware of this strategy. If the company is cheap and management is not buying more with either their own or company funds that already speaks to poor management, which is often a headwind to future returns.

You are also much less likely to have the situations where owners have forgotten about the shares and don’t see simple publicly accessible information implying that their shares have a lot of value.

Small caps generally have a lot more variance, and simpler business models, but the average small cap has significant challenges, otherwise they would have grown by now! Any benefits from fishing in this pond need to be weighed against the adverse selection and significantly negative base rate of companies in this category to make sure you’re buying something that’s really cheap and earnings that are actually sustainable. Happy Hunting.

Recently, there have been some tweets highlighting how cheap WIX is on a trailing free cash flow basis, or how it’s been left for dead by investors in the ‘SaaSpocalypse.’

I generally like these tweeters, but this analysis dramatically misses the mark. Let’s dig in.

In 2025, WIX had net income of $50 million and operating cash flow of $580 million.

They did a complicated Dutch Auction in conjunction with a private placement (more on that in a minute) and now have an enterprise value of about $3.4 billion. See the fully diluted shares outstanding spreadsheet attached.

Wix Fdso

16.5KB ∙ XLSX file

Download

Download

$3,500 divided by 500 equals 7, and that’s where some of these hyperbolic trailing free cash flow multiple claims come from.

Here’s the statement of cash flows:

First, the intuition: if we bought the entire company how much would we expect to earn in a normal year of ownership? We start with the normalized historical number and then we can forecast how much the business will grow or shrink in the future.

In order to calculate free cash flow to the owners, we have to subtract stock-based compensation. Stock-based compensation is a real expense, and the stock-based compensation listed in the statement of cash flows is management’s best point estimate for how much it would cost if WIX paid the employees in cash instead of shares.

Second, there are a lot of one-time changes in net working capital. If WIX decides not to pay suppliers (or employees) for a while this keeps cash in WIX’s pocket and looks like a source of cash. However, not paying suppliers or employees is generally not sustainable so you shouldn’t count on this in normalized Free Cash Flow.

If you do the work you will find that WIX’s normalized Free Cash Flow for 2025 was $83M, a far cry from the $500M of free cash flow the company reported. That means WIX is trading at 41x trailing free cash flow! This still might be a good price given their continued growth and AI-enabled base44 offering, but it’s not a slam dunk or signs of a company being “left for dead.”

I’m including the spreadsheet and you can customize column G to make your own adjustments and calculate normalized free cash flow.

Wix 2025 Fcf Calculation

11.6KB ∙ XLSX file

Download

Download

Finally, I have a significant problem with the private placement the company did right before their Dutch auction. On the one hand, we’re supposed to believe their shares are undervalued, and thus they’re trying to buy them back in the open market. On the other hand, WIX sold units, which are even better than shares since they have one-quarter warrant coverage, at a lower price right before the Dutch auction. This just looks like a sweetheart deal to Durable Capital Partners and makes me highly skeptical of WIX management.

Reader beware.

Appendix:

If you’re trying to decide how to normalize Free Cash Flow line items, consider asking an LLM. I found the individual comments to be useful, although it didn’t do a very good job when asked to do the entire normalization adjustment.

Wix Normalized Free Cash Flow Discussion 4

201KB ∙ PDF file

Download

Download

This is a little different from my usual writing. Over the years, friends and family have often asked me: “What is shorting?” So here’s the simple version first, followed by the mechanics: how shorting actually works, what it costs, and why it can get dangerous fast.

Conceptually, shorting is betting against a stock. You make money if it goes down and lose money if it goes up.

Imagine Dwight wants to short AAPL. First, Dwight must find someone who already owns Apple stock – we’ll call him Jim. Dwight makes an agreement to borrow Jim’s shares and promises to return them later, either when Jim asks for them back, or when Dwight wants to give them back and close the trade.

After borrowing the stock, Dwight sells it in the market at the current price, $100. At some time in the future, Dwight will have to buy Apple stock and give it back to Jim. If Apple is trading at $75 when Dwight buys it back, then he’s made $25 a share in profit ($100 - $75). However, if Apple is trading at $200 a share, Dwight has locked in a loss of $100 ($100 - $200).

In practice, it’s difficult to find shareholders of every company you might want to short. Instead, you use your prime broker (e.g. Fidelity) to borrow the shares from millions of shareholders across the country. Fidelity doesn’t work for free. To short Apple shares, you have to pay a fee, often called a short borrow rate. This fee is quoted as a percentage per year.

That’s the simple version. Now let’s make it more realistic.

Dwight is convinced the moon landing was fake and that the Artemis II mission is going to fail. Dwight borrows Jim’s Boeing shares and sells them at $100 a share on day zero. Dwight owes a 7% annual interest rate paid daily (calculated on a 360-day year) for the privilege of borrowing Boeing shares. However, after Dwight borrows the shares, he sells them and gets cash. Jim doesn’t want Dwight to run off with the cash, so instead Dwight must keep the cash in an escrow account with his prime broker. Fortunately, Dwight earns interest on the money in the escrow account, which is invested in Treasury Bills.

Initially, the stock sells off slightly as people are worried about the upcoming launch. But after a successful launch, the stock jumps higher. Upon the Artemis II crew’s successful return, Dwight “covers his short,” buying back the Boeing shares at $115.

The excel sheet below shows Dwight’s loss of $151.21 after borrow costs.

Short Borrow Illustrative Example

23.2KB ∙ XLSX file

Download

Download

I’m including a couple of additional cases. If Jim and other shareholders stop lending out their shares, this can cause a “short squeeze,” forcing short sellers like Dwight to cover. The key thing here is that in a short squeeze situation, the “borrow rate” – the annual interest rate Dwight has to pay for borrowing the shares – often increases at the same time the stock price is going up. So the position is moving against Dwight, and it’s becoming more expensive to stay short. Additionally, short sellers pay borrow fees on the market value of the position while they earn interest on the initial short-sale proceeds.

In the best case, you short a stock and it falls. As the stock declines, you continue to earn income on the initial short-sale proceeds while both the borrow cost, and the liability – the cost to buy back the shares – decrease . However, if a stock you’re short goes up, your short proceeds remain constant while borrow fees increase. You pay more in borrow costs, and the liability gets bigger because buying back the stock now costs more. Meanwhile, your interest income remains fixed because it’s based on the original short-sale proceeds*.

One final point: short selling is much more path dependent than long investing. As a long-term investor, if a stock falls or stays low for a while, that doesn’t impact the eventual return as long as you hold on. On the short side, in a steady decline situation, the borrow cost can be minimal, and it’s easy to hold the short for an extended period of time until the stock reaches fair value. However, if the stock rips higher, the cost of staying short can explode. You also have the problem of a bigger position in a stock that’s going against you. If a short doubles and then doubles again, you have not lost 200%; you have lost 300%. And now you have a liability that’s 400% of the initial size.

Bottom line: be careful short selling. Stocks are famous for taking the stairs up and the elevator down, so short-term shorts can be extremely lucrative. Just look at CAR over the past two days. But timing and sizing can be incredibly tricky. Oftentimes shorting is more an exercise in investor psychology, knowing short-term catalysts, and position sizing rather than understanding the fundamentals of the business. Just look at CAR over the past 2 months!

Appendix:

Here is a link to a good borrow cost estimator for different securities.

*The actual details are more complicated because your prime broker will adjust your margin requirements depending on the market value of your shorts.

What do they do?

Primarily, CareCloud (CCLD) handles billing on behalf of physicians and medical offices, submitting claims to insurers for reimbursement. This business segment, called Revenue Cycle Management (RCM), accounts for about two-thirds of the company’s revenue (p.49). CCLD also claims to offer AI-enabled SaaS solutions for electronic health records, practice management, customer service, and remote patient monitoring. Finally, CCLD has a long-term contract to manage physician practices through its Medical Practice Management segment.

In practice, CCLD is a low-margin billing outsourcer, processing claims via its 3,300 Pakistani employees, who make up 90% of the employee base. While they claim to be pivoting towards AI, the most concrete use case appears to be paying the Executive Chairman’s son $15k/mo for AI consulting services (p.F-32). Pakistani employee count has remained roughly flat over the past five years suggesting the core processing work remains untouched by AI.

How does CCLD make money from RCM? CareCloud charges a percentage of the amount billed to insurance. While they don’t disclose the exact split, the estimated range is 3-7% of the amount billed.

What’s their moat?

CCLD boasts a practice renewal rate (not to be confused with net revenue retention) in the 90+% range since 2021. However, aside from their low-cost Pakistani workforce, the main moat here is switching costs, as it’s painful for practices to switch systems. This is a partial moat, but since they compete as one of the lowest cost options, their customer base is the most likely to switch if faced with a superior AI offering.

CCLD’s number of practices served increased to 2,700 in 2020, following their acquisition of CareCloud, whose name they also took. Practices declined slightly to 2,600 in 2024 before increasing to 2,900 in 2025. Healthcare IT Revenue has increased only 15% since 2020, suggesting CCLD does not have any pricing power.

Concerningly, CareCloud claims that “Over the past several years, organic growth has been a Company-wide focus.” While organic growth has been improving, it remains negative: -2% in 2025, -5% in 2024 and -9% in 2023.

What are the unit economics?

The Medical Practice Management segment posted roughly 10% operating margins before overhead costs over the past few years, but it was unprofitable in 2025 and appears to be unprofitable after overhead costs going forward.

CareCloud doesn’t break out unit economics for its RCM offering though we mentioned the 3-7% take rate. RCM is the primary component of the Healthcare IT segment, and overall Healthcare IT Earnings Before Tax margins were roughly 22% after adding back amortization of acquired intangibles.

Margins have improved over the past couple of years as CareCloud seems to have achieved some economies of scale due to their past acquisitions. Historically, CareCloud has incurred significant asset impairments from unsuccessful acquisitions. However, CCLD took a break from acquiring companies in 2023 and 2024 which reduced asset impairments in 2024 and 2025 and boosted margins.

What does the future look like?

Given CareCloud’s history as a hodgepodge of acquisitions and negative organic growth, it is hard for me to view this as anything other than a melting ice cube. I don’t have confidence in their AI-enabled product solution, and in reality CCLD seems to be continuing the same business plan of outsourcing to their predominantly Pakistani workforce.

What’s the valuation?

For both 2024 and 2025, I estimate that CareCloud’s free cash flow was about $24 million. However, CareCloud has two series of preferred shares that have to be included to get to the correct enterprise value. The Series A converts into an additional 8 million CCLD shares, and the Series B liability is about $42 million. When you add those together, you get an enterprise value of about $220 million.

Dividing $24M in FCF by the $220M enterprise value gives you an 11% free cash flow yield. This high free cash flow yield keeps me from wanting to short the company, but given the negative organic growth, poor capital allocation, and red flags, I’m not sure that 11% is a high enough yield to compensate owners for the risk of owning this company.

What are the red flags?

• Organic Growth: Management flags organic growth’s importance but fails to provide historical organic growth and seems to be trying to hide that organic growth has been negative for years.

• Related Party Transactions: The CEO’s son is on the payroll for $38k/year, the Executive Chairman’s son gets $15k/month for AI consulting services, and during 2020 a telehealth VIE was formed with the Executive Chairman’s wife but as of 12/31/25 it “had not yet commenced operations.” Over the past 2 years, CCLD spent $2.6M upgrading facilities leased from a related party. Those improvements cost 4.5x the 2 years of rent paid on the same facilities.

• Medical Practice Management: CareCloud is stuck in a contract to manage these medical practices for another 13 years at what look to be unprofitable prices.

• Preferred Shares Outstanding: There has been significant historical dilution from the preferred shares, and a significant preferred share overhang remains.

What are the main risks?

• Organic growth continues to disappoint.

• Management continues their pattern of value-destroying acquisitions resulting in significant asset impairments.

• AI native solutions provide a better experience and take market share.

Conclusion

Overall, CareCloud bills itself as an AI-enabled solution for medical practices. In reality, CareCloud appears to be a melting ice cube outsourcing insurance billing to Pakistan. They have significant recent free cash flow hidden by their amortization expense. However, given their declining organic revenue, history of troubled acquisitions, and the red flags articulated above, I’m avoiding this stock for now.

Ccld Scratchpad

13.5KB ∙ XLSX file

Download

Download

I’m not seeing much intelligent analysis of the OAI 17.5% guaranteed minimum return.

There are three things embedded in that eye-popping 17.5% rate, but people are only focusing on #1.

1. Open AI wants financing to build out more data centers.

2. Open AI is trying to acquire more enterprise customers for their custom models, which they think will be sticky implementations of their enterprise AI products.

3. Anthropic’s Claude is the default in most enterprise AI use cases at this point.

Because of Claude’s dominant market share in enterprise AI (#3), OAI has to pay higher Customer Acquisition Costs in an attempt to get lock-in for enterprises that may become huge customers for OpenAI tokens in the future.

“There’s a big race to lock in as much enterprise, as many desks as possible,” said Matt Kropp at Boston Consulting Group’s AI unit, adding that once a company has a customized ​AI model integrated into its systems, it becomes much harder to switch to a competitor.

“I can see that there’s a huge amount of scalability there.”

There are only a few big firms with the ability to influence enterprise rollout across a host of smaller firms, and OpenAI really wants to remain competitive in the SMB space. They’re pursuing these few big private equity firms that can influence dozens or even hundreds of SMBs to adopt a specific enterprise AI platform.

Finally, the headline 17.5% rate sounds inflated because OpenAI can effectively PIK (Pay-In-Kind) in the form of tokens or minimum spend agreements with the PE JV partners, turning a lot of the 17.5% interest cost back into further revenue that they expect to be high margin.

Disclaimer: This post is not financial advice, and is not a solicitation to purchase or sell any securities. For more info, please review the disclaimer.

No one knows how AI will play out, but it’s possible that continued development and adoption of this technology drives dramatic declines in white-collar jobs as we know them.

People have a bad habit of only looking at things from their own perspective. When people consider their jobs, they think of their salary instead of how they are providing value. The salary you get paid comes from providing value to people. If you give people something they want, they’re happy to pay a portion of the value created for them. This will not change, no matter how advanced AI gets.

AI is a powerful force multiplier. If you know of a problem—something you want to do, or a pain point others have—AI can enable you to solve that problem more cheaply and effectively. This creates massive consumer surplus and wealth for everyone involved.

I recently took my son on a ski trip. We had a great time, but there were multiple solvable frictions: signage, queues, lift tickets, and chaos as beginners struggled with the lifts. AI won’t kill skiing, but it may make the operations cheaper. That would allow operators to lower ticket prices, letting more people ski for less, or earn superior profits while maintaining current prices. If AI makes this easier, more entrepreneurs will consider developing mountains that currently aren’t used for skiing. Available mountains are the new constraint. In the limit, robots will be able to move giant piles of dirt around and make snow. I can’t wait to ski in Texas.

An investing thought experiment: in a scenario where AI displaces many white-collar roles and there is a spike in unemployment among white-collar professionals, will time spent on social media – like Facebook, Instagram, or WhatsApp – go up or down? The market is currently pricing Meta as an AI loser, trading at roughly 16 times Family of Apps operating income.

What Citrini and many others miss in these write-ups is the massive increase in efficiency leading to massive decreases in prices, giving us much more flexibility to spend time how we want—whether that’s browsing social media, going on ski trips, or playing video games. The creation of digital assets has never been easier: AI-assisted illustration, video-game asset development, and new ways to create and share meaningful experiences with friends. AI is coming for digital experiences first and will enable them to be cheaper, better, and faster—and you’ll have more time to enjoy them. Coupled with robotics, AI is eventually coming for physical experiences as well.

Instead of being concerned about potential wage destruction from AI, recognize that creating better experiences for yourself and those around you is limited only by your imagination. The more effectively you do this, the better your life will be as you generate wealth by solving other people’s problems, improving their lives as well.

Citrini’s piece was demoralizing psychologically, first, because it focused on the demand side – where we can see the potential destruction – and second, because it preyed on white-collar workers’ insecurities that their job is predicated on intermediation. “Swaths of companies built on monetizing friction for humans disintegrated.”

“Their moats were made of friction. And friction was going to zero.”

This was the best part of the article and highlights that friction is not a real moat. If someone else can make a better user experience and that takes away your business, then your business doesn’t have a competitive advantage.

Citrini mentions real estate commissions going away, with many transactions having no buyer’s agent. I would love this future. I had a startup seeking to make this a reality without any AI required. It could save you thousands or even tens of thousands of dollars when you buy a home. However, people are skeptical of new ideas and change slowly, especially for the most expensive transaction of their lives. They often want hand-holding and, most importantly, someone to trust.

AI is nowhere near this level for now, but eventually we may get huge savings and increased productivity in the real estate market, allowing people to transact more freely or take those savings and spend them on other things that they care about. This is good for humanity.

“For every new role AI created, though, it rendered dozens obsolete. The new roles paid a fraction of what the old ones did.”

The biggest fallacy: there is no fixed number of jobs. If one million jobs become obsolete, human desires continue to increase and creative humans who understand what they, or others, want are able to meet those needs in new and interesting ways. New jobs can always be created. Maslow’s Pyramid is never filled.

If we find that coordination problems are reduced and “the middle layers” that keep “the economy lubricated” are not necessary, we should eliminate them! We can instead solve bigger and bigger coordination problems or specialize in solving ever more personalized and bespoke problems. Each of us has dozens of minute problems that aren’t currently addressed. I want an AI app that can diagnose the squeak from my washing machine. I won’t pay someone to come out unless it gets worse. I don’t want to take the time to troubleshoot it (this essay won’t write itself!), but I’d love this problem to be solved.

AI got better and cheaper. Companies laid off workers, then used the savings to buy more AI capability, which let them lay off more workers. Displaced workers spent less. Companies that sell things to consumers sold fewer of them, weakened, and invested more in AI to protect margins. AI got better and cheaper.

A feedback loop with no natural brake.

This is the core of the argument and patently false. AI “wants” more chips, more compute, more energy, better algorithms, in order to be more useful and solve our problems. We want cheaper food, more recreation, more options, cheaper computer components, more video games, more experiences, more time scrolling Instagram. Greater productivity means being able to meet more of these desires.

The core idea that income can be permanently impaired by AI displacement in aggregate is logically impossible. If productivity increases, then real income increases. All productivity is owned by someone. Some people’s incomes may diminish as they search for more productive ways to use their time or as AI eliminates their old role. But if AI raises productivity, then aggregate production, which must equal aggregate consumption, is increasing and therefore real incomes increase. Anything produced is owned by someone and so if humanity is able to produce more things that are valuable, that only makes people collectively better off.

Conclusion

• Frictions reduce wealth, they don’t create it.

• Creating value for others is the best way to maximize compensation.

• Human desires are infinite. AI boosting productivity will enable more to be satisfied.

• Say’s Law – Supply creates its own demand. There is no such thing as being too productive as people want the next scarce thing: unique experiences, recreation, excitement, skiing.

I love the imagined headlines, and I agree that frictions are not a moat. I’m also not refuting the idea that the market could crash 38%. Future productivity is not guaranteed. But if AI enables greater productivity, we will be wealthier not poorer.

Appendix – some other takes from around the internet:

Comparison to Marxism and Prices as a counter-weight

Macro agreement

General rebuttals

Micro Disagreement

DASH opinions (agreeing with Citrini)

More DASH opinions (agreeing with Citrini)

Alternative future - bull case

Citadel’s take (disagreeing with Citrini)

In 2026 Carvana has been the subject of short reports from both Gotham City Research and Abelian Analysis.

Gotham City Research’s report is imposing, running to more than 70 pages as well as supplementary filings and annual reports for two related parties. I will summarize the recent short reports as well as the company’s response and Gotham City’s follow-up.

Why me?

I have followed CVNA for years and lost money shorting their 2023 resurgence. I also bought a car from Carvana and, while I wasn’t blown away, I understand how it could be the future of used car buying.

Gotham Report

Gotham City’s short report primarily boils down to the following key claims:

1. DriveTime is owned by Eric Garcia II, the father of Carvana founder Eric Garcia III. DriveTime is inflating Carvana’s profits by over a billion dollars between 2023 and 2024 in an attempt to benefit from the higher multiple given to Carvana versus the privately held DriveTime.

   1. Evidence of DriveTime attempting to inflate Carvana’s profits includes $1B in net debt issuance for DriveTime over the 2022-2024 time period.

   2. DriveTime had negative adjusted EBITDA (after portfolio debt expense) in 2023 and 2024 compared to high single-digit adjusted EBITDA from 2009 to 2013.

   3. DriveTime has scaled significantly in the past decade with revenue growing 3.5x and 65% greater revenue per employee, yet operating margins are worse now than they were in 2010-2013. Gotham claims this is evidence of artificially low margin due to DriveTime subsidizing Carvana.

2. Gotham City also claims that Bridgecrest, another related party owned by Eric Garcia II, is inflating the gain on loan sales for the company.

3. Gotham City also claims that Carvana’s 10-K will be delayed and the 2023 and 2024 10-Ks restated, along with Grant Thornton, their current auditor, resigning.

One counterpoint: Gotham claims Eric Garcia II extracted roughly $300 million from DriveTime, yet also suggests he’s trying to maximize family wealth by boosting Carvana’s reported economics. Since Carvana trades at a much higher multiple it would make sense for Garcia to subsidize Carvana’s earnings with the $300M instead of withdrawing it in cash for himself. The market, in turn, would have given Carvana an even higher multiple, and the Garcia’s would have profited even more as they’ve sold billions of dollars of stock in the last two years.

Gotham Report Analysis

Now let me provide my analysis of the following three points:

1. There is clearly evidence of related party transactions as Carvana discloses in their 10-K. Other short sellers like Hindenburg have also pointed to these transactions, which may be inflating profits or decreasing warranty costs as well as other financing and servicing costs. These arrangements are suspicious enough to me that although I don’t see a smoking gun, I would likely avoid investing in Carvana or significantly decrease my estimates of their run-rate profitability to take into account potentially temporary favorable terms here.

2. Gotham City also claims inflated gain on loan sales. This argument is much less compelling to me, as we can see the gain on loan sales decreasing as a percentage of adjusted EBITDA. Furthermore, it obviously makes sense that Carvana may be originating loans that are immediately profitable to Carvana. Dealer financing is a core profit driver for used car dealers. After dealers make a loan, they are then able to sell these loans and immediately recognize a profit on the loans they’ve made to customers wanting to buy their cars. The rest of this claim reduces back to suspicious related-party transactions per #1 above.

3. Gotham City claimed Carvana’s 10-K would be delayed and Grant Thornton would resign. This has clearly proven false as Carvana filed their 10-K on time. It’s too soon to see if the 2023 and 2024 10-Ks will be restated because that could happen at any point in the future, but so far it has not occurred. I would judge this claim to be extremely optimistic, overly promotional, and ultimately false.

Abelian Report

An Abelian Analysis short report boils down to a claim that credit quality is deteriorating for Carvana asset-backed loans as they have started to offer a small subset of loans to very suspect car buyers with poor credit ratings. Abelian’s main claim is a forecasted increase in defaults due to poor consumer credit and car repossessions leading to a significant increase in Carvana’s cost to package and sell future asset-backed securities hampering their ability to sell cars.

Abelian Report Analysis

Much of Abelian Analysis’s post is about the future, and I don’t have a crystal ball. To some extent, we’ll have to wait and see if this year the ABS is underperforming, leading to more expensive financing for these asset-backed securities. However, I’ve heard many pitches and claims like this in the past, and often times credit quality is much stickier and better than people forecast. If there are signs of significant losses, Carvana is likely to be able to raise the rates they charge lower-quality consumers and, in turn, pass those higher rates on to investors. So, it is hard for me to imagine this as an existential threat to Carvana’s business model.

Conclusion

Overall, there are persistent suspicions around the related-party transactions between DriveTime and Carvana. However, Gotham City’s report lacks a smoking gun, and the poor operating results they point to can also be explained by weak auto loan performance in 2022-2024. Abelian’s report is more forward-looking and claims future auto loan performance will be bad. However, I don’t believe this is an existential risk to Carvana’s business model as they can adjust pricing and shouldn’t be disadvantaged compared to other auto dealers.

I have no position in Carvana and remain skeptical of some of the related party transactions. I don’t see Carvana having a significant moat here, but perhaps I underestimate the logistical difficulty of the automotive network they have compared to eBay autos, CarMax, or Amazon getting into the space. At current prices, CVNA is too expensive for me, and I will remain on the sidelines.

Resources:

Cvna Short Report Gotham City Research 1

4.69MB ∙ PDF file

Download

Download

Swanson Investment Research

Carvana: Dissecting the Gotham Short Report

Gotham City Research recently published a short report on CVNA: “Carvana: Bridgecrest and the Undisclosed Transactions and Debts…

Read more

6 months ago · 10 likes · 3 comments · Austin Swanson

Thread responding to Gotham Short Report

JPM executive summary response to Gotham Short Report

Gofi 2024 Annual Report

184KB ∙ PDF file

Download

Download

Drivetime 2024 Annual Report

762KB ∙ PDF file

Download

Download

Cvna Supplemental Materials 1

2.82MB ∙ PDF file

Download

Download

Abelian Short Report

Twitter thread - Abelian data correction

Hindenburg CVNA Short Report (Jan 2025)

Kerrisdale CVNA Short Report (Feb 2024)

Note: GRVY was a PFIC and I believe is likely to remain one unless they significantly reduce their cash balances. This may have tax implications for US investors. This is not tax advice and please consult your own tax advisors.

Gravity Co. Ltd (GRVY) is a Korean video game developer. They primarily develop mobile games which was 81% of revenue in 2024. Over 80% of their revenue comes from the Ragnarok game series. They have a license agreement with the Ragnarok creator until 2063. The company is controlled by GungHo Online Entertainment Inc. who owns 59% of the company and also distributes the games in Japan.

Gravity doesn’t have much of a moat beyond customer affinity for the Ragnarok IP. As the company mentions: “Ragnarok Online has been on the market for 23 years and has reached maturity in most of our principal markets.” The company’s revenue has been reasonably stable, and Gravity has been profitable since 2016 thanks to their transition to mobile gaming.

The main attraction is the cash. With over $400M of Cash & Short-Term Investments, GRVY has an Enterprise Value of $70M USD and is trading at 1.3x the operating profit of approximately $53M for 2025 they announced today. Revenue is recovering this year and profitability was pressured as they announced the lowest operating profit since 2019. However, the future still looks promising and app revenue trends are moderate to strong.

The primary problem is that capital allocation has been terrible at GRVY with the cash balance quadrupling over the past 6 years. GRVY has always had a conservative balance sheet but this continued obstinance has cost shareholders hundreds of millions of dollars and led to a new activist campaign this year at GungHo, the primary shareholder. If this activist campaign is successful in putting pressure on GungHo, then it would likely cause Gravity to distribute cash, either to their parent GungHo and other shareholders via special dividends, or through share buybacks.

Risks:

• The cash is fake – This is unlikely as the cash buildup matches the cash flow from the financial statements and cash should be very easy to audit. However, I’m less familiar with Korean audit practices.

• Poor capital allocation continues – Management continues to hoard cash and the stock is “dead money.” This is probably the most likely outcome, in which case I expect the shares to generally appreciate by about the amount of positive cash flow each year. Downside is pretty limited and at a greater than 10% earnings yield this isn’t a disaster.

• Ragnarok IP loses relevance – With Ragnarok sales impaired the company might become unprofitable and begin to burn through their huge pile of cash. I think this is unlikely given the established nature of the IP and I’m encouraged by their recent app trends and increase shots on goal via other apps.

• GungHo cheats minority shareholders – GungHo might be able to find a way to have GRVY buy back GungHo’s shares at high prices while not offering the same option to minority shareholders or make a disastrous acquisition of another video game developer.

Overall, if management begins to intelligently allocate capital, particularly through special dividends or share buybacks, it will address 3 of the 4 main risks and likely lead to a doubling in market cap or better for GRVY. Given the recent catalysts along with the cash balance and history of profitability I think there’s sufficient margin of safety to make GRVY an attractive investment. Disclosure: Beeli Capital is currently long GRVY.

Grvy Analysis 2

19.2KB ∙ XLSX file

Download

Download

Appendix:

Q4 25 Revenue Update

Q4 25 Earnings Presentation

"From $100K to $1M" & More.

Gravity Co., Ltd. ($GRVY)

This write up was done in collaboration with AppInvestor…

Read more

8 months ago · 35 likes · 12 comments · Yegor and AppInvestor

Gravity writeup (2023) Clark Square Capital

Altay Capital - Mostly Value Investing

Gravity - Cheap but an Obvious Value Trap

Gravity Co., LTD. is a video game developer based in South Korea and trades in the U.S. (GRVY) and is extremely cheap. It’s cash rich and trades at TTM P/E of 8.3. Net income TTM is $35.5m USD vs $295m market cap…

Read more

4 years ago · 6 likes · 2 comments · AltayCap

What do they do?

"It's a me, Mario!" Home to the world’s most famous plumber, Nintendo designs, manufactures, and sells entertainment consoles (currently the Switch and Switch 2) and video games for those consoles, along with taking a cut of third-party games and licensing royalties for their intellectual property.

Notable intellectual property includes: Super Mario, Legend of Zelda, Pokémon, Donkey Kong, Metroid, Kirby, Animal Crossing, Pikmin, Splatoon, Super Smash Bros., and Fire Emblem.

Hardware is generally sold at low margins while sales of both their first party games – particularly Zelda, Mario, Mario Kart, Super Smash Bros, Splatoon – and their exclusive games (Pokémon) drive operating profit.

What’s their moat?

The primary moat is based on the IP of their recognizable characters and the advantages they have from selling software (video games) that are deeply integrated with their hardware. Nintendo is a master of entertainment, storytelling, and videogame design with a host of iconic characters.

Who are their competitors?

XBOX (MSFT) – 3rd place console system for the last few cycles. Xbox is pursuing an all-you-can-eat subscription aggregation strategy with Xbox Game Pass due to a lack of hardware differentiation or compelling exclusive content.

PlayStation (SONY) – the console winner so far this generation. The Nintendo switch outsold the PlayStation 4 by units, but likely generated less in hardware sales revenue due to the lower price point.

iPhone (AAPL) – For the past decade console sales growth has been anemic, while mobile gaming has seen huge growth. While smartphones are not traditionally seen as a direct competitor for Nintendo, better iPhone games reduce the need for a dedicated video game console. Nintendo is particularly susceptible to competition here as portability is a key selling point of the Switch and Switch 2. A smartphone already goes with you everywhere and is more convenient, although the lack of joysticks or buttons mean game input is often worse.

Other Entertainment (Social Media, NFLX and YouTube, etc.) – In the same way that Netflix likes to talk about how their main competition is sleep, Nintendo’s real competition is any other form of entertainment you might select instead of buying and playing their video games. Primary competition includes social media like Facebook, Instagram, TikTok, etc., as well as Netflix and YouTube.

What are the unit economics?

Nintendo doesn’t break out their unit economics but we can estimate them. Over the last two years, gross margins have been in the 60% range. However, in the last nine months, due to the introduction of the Nintendo Switch 2, gross margin has declined to 37%. I estimate the blended hardware gross margin for 2026 is 19%, leading to roughly break-even operating profit per unit. However, the software gross margin is close to 95%. SG&A has been flat the last 3 years providing the potential for significant operating leverage.

What does the future look like?

Video game console release is cyclical. While Nintendo is seeking to normalize revenue opportunities through their annual subscription, a dedicated user base in the 100 million-plus range, and stretching out software sales, the core driver of more software sales is success on the hardware front. I’m comfortable estimating 100 million+ units sold in the lifetime of the Nintendo Switch 2, given the strong sell-through and backwards compatibility with Nintendo Switch games. However, the ability to reach or exceed the Nintendo Switch number of units remains to be seen.

Video game sales peaked during covid and then declined 34% from the highs. Going forward I’d expect 200M+ units sold as the normalized base. Note, Nintendo’s fiscal year ends 3/31 and for the first 9 months of FY 2026 they had a weak slate of first party releases while FY 2027 is expected to be stronger.

What’s the valuation?

The current market cap is $71.8B (~¥11.3T). I’m giving them credit for $18.2B (~¥2.9T) in Cash, Securities and Investments based off their latest balance sheet. With no meaningful debt this gives an enterprise value of $53.5B (~¥8.4T). Because we’re looking at Enterprise Value I will be excluding Interest & Investment Income as well as FX G/L. NOPAT is a reasonable proxy for Cash flow and over the past 5 years Nintendo’s has averaged around $2.5B (~¥397B) for a trailing multiple of 21.2x. Last year’s operating income was particularly depressed, and this year’s operating income is also depressed due to the launch of the Switch 2. Going forward, analysts estimate NOPAT around $2.6B (~¥418B) for FY 2027 and then growth to $3.3B (~¥529B) in FY 2028. If you believe that Nintendo can grow their normalized earnings by 26% in FY 2028, then it trades at 16x 2028E NOPAT.

Given the inherent cyclicality, I’m unwilling to underwrite LT secular growth above 5% so I’d want at least a 6-7% FCF yield leading to a 14-17x NOPAT multiple. Basically, I think Nintendo trades fairly to a slightly optimistic 2028E NOPAT. Additionally, because of the boom and bust nature of console development, I think Nintendo is likely to maintain a fortress balance sheet enabling them to ride out adverse conditions. This significant cash balance reduces upside because if Nintendo’s EV/2028E NOPAT increased from 16 to 20 (25% multiple expansion) the market cap only increases by 19%.

What are the main risks?

Cyclical video game hardware sales – Reception for the Nintendo Switch 2 seems to be very strong, and reviewers are impressed by this latest iteration. I plan to get one in the next 12 months. However, if their next console is a dud like the Wii U, then normalized revenue is likely to shrink significantly. This uncertainty is embedded in Nintendo’s business model. In the more distant future, augmented reality or virtual reality goggles may enable experiences like those in Ready Player One. Dramatically changing how we think about video game consoles and entertainment.

Failure to innovate and develop intellectual property – There's a new Super Mario World movie coming in 2026, development of Nintendo World at Universal theme parks, and many of their games recycle the same IP. While it’s hard for me to imagine at this time, poor storytelling and execution could lead to brand fatigue and weigh on console and video game sales in the future.

Conclusion

Overall, Nintendo is an interesting company with a fortress balance sheet and beloved intellectual property. However, given the cyclical nature of hardware sales and the resulting video game software unit sales – I think this company is deserving of a mid-teens multiple or ~6% free cash flow yield given the boom and bust nature of the overall business and difficulty estimating normalized future free cash flow.

They have some interesting ancillary growth opportunities, and Switch 2 sales could easily continue to be strong with revenue growth surprising to the upside over time. Margins should improve as well, as software sales become a higher percentage of overall revenue. However, given the cyclicality of revenue, I’m unwilling to underwrite high single-digits revenue or earnings growth for the next 5-10 years.

Despite the 45% drawdown from recent highs, Nintendo is still a pass for me.

Appendix:

Emerging Moats Research

Why We Own Nintendo Stock (Ticker: NTDOY)

Reminder: these are show notes that should be read in conjunction with the podcast. Do not expect these notes to be a polished research report…

Read more

3 years ago · 9 likes · Brett Schafer

Emerging Moats Research

Why I Own Nintendo Stock: Round Two (NTDOY)

YouTube…

Read more

2 years ago · 8 likes · 4 comments · Brett Schafer

Best Anchor Stocks

🎙️ Nintendo: The Switch 2 Edition w/Ryan O'Connor

Hi there…

Read more

a year ago · 4 likes · Best Anchor Stocks

Margin of Safety Investing

Nintendo $NTDOY

Note: I use Stratoshphere to help with my research, to follow my holdings, screen for ideas and get insights to company specific KPI’s that drive the business results. I also use it for charting in m…

Read more

5 years ago · 3 likes · Simon Handrahan

Best Anchor Stocks

Reading between the lines

Nintendo reported 9-month earnings on Tuesday and the stock suffered (in Japan) one of its worst reactions in a long time (it was down 12% at one point). The following days weren’t great either, and …

Read more

6 months ago · 18 likes · 2 comments · Best Anchor Stocks

By now, you’ve probably used AI to summarize a filing, polish a write-up, or draft a bull/bear case. I’ve found these to be helpful use cases, but not revolutionary.

Here are some more creative ways I’m using or considering using LLMs in finance.

1. Short Report Scrivener – Bearish reports on publicly traded companies, especially learning from certain authors’ style. I’m able to quickly screen for red flags and pull key metrics for many suspect or potentially overvalued companies. LLMs subject the company to scrutiny and highlight where management claims are not holding up.

2. Subject Matter Expert – Particularly for technical claims e.g. in semiconductors, telecom / spectrum, advanced materials, etc. There is wide variability in the usefulness of approaches here and I’ve experimented with various prompts. If you’ve found a good one consider mentioning it in the comments! I find LLMs are very useful for summarizing the key technical claims and know a lot of the vocabulary, but refuse to have an opinion on viability or desirability of most management claims.

3. Litigation Expert – First pass expert for Mergers and Acquisitions, various corporate legal matters including a company’s potential legal liability and details around business combinations, or acquisition agreements. LLMs can spot mistakes in the drafting of legal documents. This can be hugely helpful on the credit side, but even though I focus on equities this is still helpful. If you feed an LLM court filings they can quickly summarize and lay out the chronology of the case. However, Gemini and ChatGPT often don’t seem to have access to the court documents.

4. Patent Attorney – Evaluating a company’s moat, defensibility and coverage of patents and intellectual property. Particularly useful for short reports (#1)

5. Management Whisperer –

a. Proxy Analysis, LLMs can pull the last three to ten years of historical proxy statements to evaluate management compensation targets, subsequent company performance, and determine if the company changed the targets to still give generous executive compensation despite poor corporate performance.

b. Earnings Call Analysis, LLMs as an aid in evaluating management credibility. They can look at past deliverables and see if management has been consistent or has a tendency to overpromise and underdeliver. This can be as simple as seeing if the FX headwinds that caused difficult comps this year were properly flagged as tailwinds last year. It can also include more complex tasks like evaluating whether management met 3–5 year goals laid out at prior investor days, and grading whether this was due to sector performance, management execution, or circumstances beyond their control (e.g. Covid).

6. Historical Investment Pitch Evaluator – I’m working on a project using LLMs to assist in grading prior investment pitches simplifying to a few key predictions. Examples of key pitch predictions include: multiple expansion, top-line acceleration, or margin improvement. Then I’ll assess whether business and stock performance matched the key criteria articulated, and determine if the stock pitch was just lucky, or if analysis nailed the key determinants of future business and/or stock performance.

So far, I’ve seen LLMs do well at summarizing information, using correct terminology, and providing helpful primers when I’m getting up to speed. LLMs also excel at improving readability and simplifying more obscure or complicated syntax. However, in my experience, ChatGPT, Gemini, and other AI models tend to avoid taking a strong opinion or view on anything, instead trying to hedge and return an outcome that’s in the middle of the distribution.

I’ve also started using Wispr for dictation, which has been very user-friendly and accurate. I intend to use that in the future, especially to allow a more active lifestyle while still generating and consuming content. As always, the primary content will be my own, but I continue to seek to find ways to utilize these tools to speed up production, consume more information, and polish the output.

Today, QXO Inc. announced an agreement with Apollo and other lenders to sell Series C Convertible Preferred Stock worth $1.15B to fund any acquisition that would be agreed to within the next 6 months. The Preferred has a dividend rate of 4.75% and a conversion price of $23.25. Investors reacted favorably with the stock up ~18% today.

However, the “cheap financing” narrative can be misleading. I’ve analyzed the terms and believe that QXO did a good job getting close to fair value on the preferred stock. This deal both shows Brad Jacobs’s financing savvy and leads analysts and shareholders to expect an acquisition announcement in the next 6 months fueling the rally.

I estimated a 300bps spread (potential triangulation here) to the Fed Funds rate which gives a FV for the dividend payments of about 71 cents on the dollar. The option is more difficult to value as the Series C Preferred is perpetual. Black-Scholes doesn’t work for infinite lived options. However, the details indicate that QXO can force conversion after 2 years if QXO is greater than $40.69 or in 3 years if QXO is >$34.88. There’s also an option for QXO to retire these convertibles at 107% after year 7, 104% after year 8 and at par after year 9.

As a simplifying assumption, I used a 1,000 day option life and a 60% vol to price the option. The 1,000 day life sits squarely in the forced conversion window, and 60% vol is a discount to recent volatility while reflecting the much lower Beacon Roofing volatility. These assumptions result in a value of $7.05/sh and overall option value for the preferred of 30 cents on the dollar.

Alternate link where you can change assumptions.

Ultimately, the assumptions lead to an almost $10,200 value per $10,000 par indicating that Apollo paid a very modest discount to FV. However, both sides likely knew that there would be positive reflexivity upon announcement making the economics even better for Apollo.

Qxo Convertible Preferred Analysis 1

143KB ∙ XLSX file

Download

Download

While this financing was good news, today’s price change seems like an overreaction. I’ve trimmed the position but am still bullish on the LT outlook for QXO and Brad Jacobs’s deal-making prowess.

Finally, if anyone has more theoretically rigorous ways to calculate the value of this complicated perpetual convertible preferred, in contrast to my simplifying assumptions, I would love to chat!