filo/iStock via Getty Images

Dear Partner:

The Greenlight Capital funds (the “Partnerships”) returned -3.6%1 in the third quarter of 2025, net of fees and expenses, compared to 8.1% for the S&P 500 index, bringing our year-to-date return to 0.4%, compared to 14.8% for the index.

We ordinarily end our letters with a quote. This time, we’ll start with one:

“It was my understanding that there would be no math during the debates.”

— Chevy Chase (playing President Gerald Ford on Saturday Night Live)

In that context, let’s discuss Artificial Intelligence (AI), weaving in another pop culture reference from Michael Lewis’ *Liar’s Poker *– the BSDs.2

Some BSDs are throwing around some very big numbers. In August, Tim Cook promised President Trump that Apple (AAPL) would spend $600 billion in the U.S. over the next four years. For context, last year Apple spent $10 billion on Capex and $33 billion on R&D worldwide. Apple’s entire U.S. payroll is estimated at $15 billion.3 Analyst forecast models for Apple – at least the ones we’ve seen – show no evidence of the increased spending implied by Mr. Cook, and the company has not released a detailed plan regarding this spending. We don’t know if Mr. Cook was serious, or whether he was simply glad-handing President Trump to achieve favorable treatment on tariffs.

In September, Mark Zuckerberg promised President Trump that Meta would spend *at least *$600 billion in the U.S. through 2028. Moments later, Mr. Zuckerberg was caught on a hot microphone telling the President, “Sorry, I wasn’t ready… I wasn’t sure what number you wanted to go with.” Perhaps Mr. Zuckerberg’s “at least” comment was designed to merely outshine Mr. Cook.

Not to be outdone, Sam Altman of OpenAI (OPENAI) claims he wants to invest trillions in data centers and related infrastructure. To date, OpenAI has raised about $64 billion, while it is expected to have $13 billion of revenue in 2025 and lose many billions for years to come.

McKinsey & Company estimates cumulative total spending of $6.7 trillion on data centers globally through 2030, including $5.2 trillion in capital expenditures on those equipped to handle AI processing. But where will the nearly $7 trillion come from? In 2024, before the major ramp-up in capital spending, the Magnificent 7 generated about $500 billion in combined cash flow. Roughly half went to buybacks and dividends, about a third to capital spending, and the small remainder was split among M&A, debt repayment, and cash accumulation. Even if they cut buybacks and dividends, the Magnificent 7 can’t do the whole AI spend without taking on a lot of leverage.

The total book equity of the Magnificent 7 is about $1 trillion. Unspent dry powder in the private equity and venture capital industries stands at roughly $2.5 trillion and $300 billion, respectively. Wall Street’s biggest year for IPOs and secondary offerings raised about $300 billion. Even if all this capital were dedicated to AI investment, it would still leave a multi- trillion-dollar gap – one that would need to be filled through debt and leases. For context, the record year for U.S. nonfinancial corporate debt issuance was $1.2 trillion.

And what about the revenue that might create a return on such a massive investment? Today, total global annual advertising spend is about $1 trillion, while all paid subscriptions in media, software, cloud, and related services accumulate to another $1 trillion.

According to Bain & Company, AI revenues would need to reach $2 trillion by 2030 to generate an adequate return on the AI investment. In other words, that would be about 100% of today’s global advertising and subscriptions combined.

Something has to give.

Let’s back up for a moment. Reflecting upon the internet bubble with the benefit of hindsight, whoever was the most bullish about the internet in the year 2000 was still not bullish enough. At the turn of the century, we had no smart phones, no tablets, no Wi-Fi, no Google, no Facebook, no TikTok, no cloud, and no video streaming. The internet has become much more important than even the biggest optimists believed.

There is a good chance that 25 years from now, AI will turn out to be even more important than we currently imagine. Whether that’s a good thing or not is an open question.4 But, the path from here to there is likely to be very bumpy for investors.

Current AI investment is focused on large language models (LLMs), which, as we are all experiencing, are very impressive. However, these models have significant limitations because they don’t truly think. They first get trained and then they get used. The industry calls this usage “inference,” but that term is misleading, as the models don’t infer or reason in a human sense. Instead, they rely on sophisticated statistical correlation techniques and trained pattern recognition to create responses to queries. The responses are so impressive that they give the illusion of having been created through inference or reasoning, when, in reality, they are doing neither.

While some have criticized the models for hallucinations (false answers expressed with confidence) and other types of errors, it seems very likely that LLMs will improve over time. We recall, with a bit of nostalgia, Apple’s first release of its Maps application, when it told us we could drive from our office in Midtown Manhattan to New Jersey by going straight through the Hudson River. Even so, while response quality, hallucinations and other errors are likely to improve over time, LLMs, as they currently exist, are not designed to think or reason.

The BSDs are nonetheless pushing the world to spend trillions with expectations of so much more. Their worldview is that computers will surpass human intelligence and achieve Artificial General Intelligence (AGI) within a few years. From there, AGI machines will assume the process of developing future AGI in recursive self-improvement, such that within a few additional years, computers will become smarter than all humans combined (Artificial Super Intelligence or ASI).

Once this happens, it is predicted that there will be enormous advances in robotics and other autonomous machines (including trains, planes and automobiles) that will create massive revenue streams to justify all the spending. *The Jetsons *– and then some – could be here soon.

Will all of this happen in the time frame suggested or even at all? We don’t know. The BSDs leading us down this path don’t know either. The optimists might be right, or it may turn out to be much more challenging than they think. We’ve seen analogous over- optimism with technologies like self-driving cars. For the past ten years, full autonomy has always been said to be just ‘one year away,’ yet it still hasn’t materialized. The truth is that nobody knows when (or even if) human-level AI will be achieved.

Significant scientific breakthroughs in model architecture, learning paradigms, computing efficiency, or something fundamentally new will be required to achieve AGI. The experts in the field simply *believe they will come. *Perhaps they will be proven right, or perhaps the challenge of getting computers to genuinely reason like humans will prove far more difficult than expected. What we do know for certain is that the BSDs that want the world to spend trillions have huge financial incentives to be believers. In case you haven’t noticed, Wall Street is also being paid a lot to promote the story.

And promote it they do. Recently, a leading bank published a report that anticipates $1.1 trillion of AI revenues in 2028 at 70% margins. Regardless of whether this prediction proves accurate, we must question what is truly AI revenue, and who stands to profit from it.

Here is how we see it (these numbers are illustrative):

• The consumer or business user spends $1 on a ChatGPT subscription (OpenAI revenue).
• OpenAI provides the service by spending $2 on Microsoft AI infrastructure (Microsoft (MSFT) revenue).
• Microsoft spends $0.60 on leasing GPUs from CoreWeave to handle the compute load (CoreWeave (CRWV) revenue).
• CoreWeave spends $2.40 on chips from NVIDIA (NVIDIA (NVDA) revenue) and $2.40 on other Capex (revenue to a variety of companies).5

So altogether, $1 of loss-making customer revenue cascades into more than $8 of aggregate AI revenue across the supply chain. Microsoft and NVIDIA show excellent margins that are subsidized by the losses of OpenAI and CoreWeave. In fairness to CoreWeave, its capital spend can be reused until it depreciates, and we are stating these amounts because CoreWeave’s Capex goes directly into the reported revenue of NVIDIA and its other suppliers.6

Rather than external enterprises or consumers spending new money, much of the AI revenue comes simply from AI companies buying products and services from each other. The largest announced third-party deal is KPMG’s annualized $400 million spend with Microsoft, which includes cloud services that might not be entirely AI related. In the context of this discussion, $400 million is but a pittance.

As we all know, investing in the internet in the late 1990s and holding through 2002 proved to be a very poor decision. It is often difficult for investors to separate the importance of innovations from the merits of the related investments. The capital spending numbers being thrown around today are so extreme that it’s really, really hard to understand them. There is a reasonable chance that a tremendous amount of capital destruction is going to come through this cycle, even if AI ultimately turns out to be everything it’s cracked up to be… and more.

Chevy Chase’s SNL line about no math was a good joke. But when it comes to AI, doing math is essential. The figures simply must add up, and right now the math is… challenging.

We write all of this to explain why we are refusing to participate in the excitement. It has been a good year for the S&P 500 and a great year for the few dozen companies central to the AI story. It has been harder to make money on long investments outside of this ecosystem, because most of the rest of the economy has been floundering. While others are doing better than us for the time being, many are taking risks that we find hard to get comfortable with.

In our experience, when the tide turns, it does so quickly and without warning. Even 25 years later, it’s still not clear why the internet bubble popped when it did. Our view is that it was due to the last buyer buying and the last short seller covering – a phenomenon that is very difficult to time. This remains the most expensive market we have experienced, and we don’t see a better option than continuing to be cautious.

In the third quarter, macro made 2.2% net, our longs made 0.8% net, and our shorts detracted 6.6% net. Similar to last quarter, we had negative alpha from our longs and minimal alpha from our shorts.

Macro continued to make notable positive contributions to our performance, led by an almost 17% increase in the price of gold. Green Brick Partners (GRBK) also advanced 17% and would have been a material winner, but most of the gain was offset by a housing hedge we have in place. In the near term, we continue to be cautious on the housing market.

We had a material short loser that cost us 1.8% net in a profitless financial services company that transitioned from a near-term bankruptcy candidate to a meme stock. In the long book, Kyndryl Holdings (KD) fell 28%, which reversed gains achieved earlier in the year. There was nothing terrible about the quarterly result, but it was much less impressive than previous reports. Lanxess (Germany: LXS)(OTCPK:LNXSF)(OTCPK:LNXSY) fell 16% in the quarter, as any kind of a cyclical recovery appears more elusive than we expected earlier in the year.

We established a medium-sized position in Pacific Gas & Electric (PCG) at an average price of $15.15 per share. PCG is a California-based regulated utility that transmits and distributes electricity and natural gas. In theory, regulated utilities operate as local monopolies that earn a guaranteed return on capital invested. This includes passing through costs from catastrophes such as the wildfires in California. In practice, these cost recoveries are sometimes complicated by politics, which can also lead to unexpected outcomes. As a result, California-based regulated utilities have a much higher cost of capital than non- California-based regulated utilities. This higher cost of capital directly passes through in higher prices to California residents. Affordability has become a hot-button political issue in California, and state officials are focused on solving this through reform.

In January, Los Angeles suffered a disastrous wildfire. While PCG was not involved, Edison International (EIX) was, and related claims are expected to consume most (or all) of PCG & EIX’s shared California Wildfire Fund, an important defense against wildfire- related damage claims. On the heels of this, PCG’s earnings multiple collapsed to a low of about 8x. We invested with the view that the legislature was likely to put in place funding support for the California Wildfire Fund and make further wildfire risk reform a priority. Since we invested, we’ve seen positive progress on each of these initiatives and believe that as tail risk recedes, PCG shares should close the discount to the nearly 18x average multiple for the sector. PCG ended the quarter at $15.08.

We want to update you on our small investment in Coya Therapeutics (COYA). COYA is a clinical-stage biotechnology company whose lead drug, aimed at treating ALS (Lou Gehrig’s disease), has just entered clinical trials. While results won’t be known for about a year, the compound appears very promising and, if the trial succeeds, it has an excellent chance to receive “fast track” approval from the FDA. COYA is a small position for us because its market capitalization is only about $100 million, and we are the largest shareholder.

COYA’s scientific approach was recently profiled in *Nature.*7 Additionally, last week Dr. Shimon Sakaguchi – a member of COYA’s scientific advisory committee – was awarded the Nobel Prize in Medicine for his work on immunology in an area consistent with COYA’s approach. His research forms the foundation of COYA’s focus on ALS, Alzheimer’s and other related conditions. The Nobel Prize validates this science and underscores COYA’s leadership in advancing a potentially transformative new class of treatments.

Of course, there is no way to know how the trial will turn out. But, at a time when AI- related business plans that are not much more than a PowerPoint presentation are being funded at multi-billion-dollar valuations, we’d prefer to speculate on COYA at a $100 million valuation.

After just over five years, we exited our investment in Teck Resources (TECK) with a large gain. Some of you will remember that we pitched this at the Sohn Investment Conference in 2021, and since then the thesis has mostly played out. The company divested its metallurgical (MET) coal business, completed its capital spending program and opened the huge new Quebrada Blanca mine in Chile. Copper prices have strengthened and the company transitioned to returning capital to shareholders.

Nonetheless, there were a series of setbacks along the way. Quebrada Blanca construction took longer than expected, was over budget and has subsequently produced less copper than anticipated. While TECK initially proposed a creative and value enhancing plan to split off the met coal business, that plan was rejected by shareholders, who did not seem to appreciate the opportunity. Ultimately, the met coal business was sold for a mediocre price. Finally, the company recently announced a “merger of equals” with Anglo American plc, which was a bit less than a merger of equals due to Anglo American shareholders being awarded a pre-closing dividend. In any case, we don’t believe TECK did a great job of maximizing value upon exit. As we assess the transaction, we are less enthusiastic about owning the combined company than we were about owning TECK as a standalone company. After the announcement, TECK traded up due to market excitement related to the anticipated deal synergies. While we had trimmed our position opportunistically over the years, we decided to now make a final exit. Over the five-year holding period, partly because the thesis worked and partly because we did a good job trading around the position, we achieved a 52% net IRR on this investment.

Save the date for our 30th Annual Partner Dinner, which will be held on January 20, 2026 in New York.

At quarter-end, the largest disclosed long positions in the Partnerships were Brighthouse Financial, Fluor, Green Brick Partners, Lanxess and Solvay. The Partnerships had an average exposure of 90% long and 60% short.

“It’s always a low when life begins to imitate an old Chevy Chase movie.”

— David Levithan

Best Regards,

Greenlight Capital

Footnotes

1 Past performance is not indicative of future results. Please refer to information contained in the disclosures at the end of the letter.

2 Big Swinging D*cks (footnoted and censored because this is a family-friendly letter).

3 We have seen some commentary that U.S. spending includes supplies purchased abroad from U.S.- headquartered companies. We do not believe that counting this would be consistent with the spirit of Mr. Cook’s promise to President Trump. Apple does very little product manufacturing or final assembly in the U.S.

4 For a very good discussion of this, we recommend Sam Harris’ *Making Sense *podcast episode #435 - “The Last Invention.”

5 We are ignoring expenses to other vendors like electricity providers. Next year CoreWeave is projected to only spend $1.50 of Capex per dollar of revenue.

6 We estimate the $8 of revenue is amplified into $100-$200 of stock market wealth.

7 Autoimmune response to C9orf72 protein in amyotrophic lateral sclerosis, Nature, October 1, 2025, https://www.nature.com/articles/s41586-025-09588-6.

The information contained herein reflects the opinions, estimates and projections of DME Capital Management, LP d/b/a Greenlight Capital (collectively with its affiliates and predecessor entity, “Greenlight”) as of the date of publication, which are subject to change without notice at any time subsequent to the date of issue. Greenlight does not represent that any opinion, estimate or projection will be realized. All information provided is for informational purposes only and should not be deemed as investment advice or a recommendation to purchase or sell any specific security. Greenlight has an economic interest in the price movement of the securities discussed in this letter, but Greenlight’s economic interest is subject to change without notice. While the information presented herein is believed to be reliable, no representation or warranty is made concerning the accuracy of any data presented.

GREENLIGHT® and GREENLIGHT CAPITAL, INC. with the star logo are registered trademarks of Greenlight in the United States, European Union and other countries worldwide. All other trade names, trademarks and service marks herein are the property of their respective owners who retain all proprietary rights over their use. These materials are (a) confidential and may not be disclosed, distributed or reproduced without the prior written permission of Greenlight, and (b) intended solely for the use of the person to whom the materials have been delivered by Greenlight.

Unless otherwise noted, performance returns reflect the weighted average total returns, net of fees and expenses, for “New Issue Eligible” investors invested since inception in the dollar interests of Greenlight Capital LP and Greenlight Capital Offshore Ltd. (collectively, the “Partnerships”). Performance returns generally include all positions held at the master and feeder funds, but exclude the gold interests. The performance of the dollar interests and the gold interests have been different. Upon request, Greenlight will provide the specific performance of each type of interest to help investors understand these differences over time. Combined performance of components of the Partnerships is presented for illustrative purposes only and does not reflect the performance or risk profile of any individual fund. Please contact Greenlight for additional information regarding the performance of the portfolios from which this information was derived or extracted. Returns are net of the modified high water mark performance allocation of 10%. This is the performance allocation applicable to an investor whose capital account is below its modified high water mark. Investors whose capital accounts are not below their modified high water marks would experience different performance. Returns are also net of the weighted average of the actual management fees paid by investors. An investor’s actual returns may differ from the returns presented due to several factors, including the timing of each investor’s capital activity, the applicable management fee rate, which may be greater or less than the weighted average of actual management fees depending on the date and amount of such investor’s capital contribution, and the applicable performance allocation rate, which may be 10% or 20% depending on whether such investor is below such investor’s modified high water mark.

All figures are unaudited. Greenlight does not undertake to update any information contained herein as a result of audit adjustments or other corrections. Past performance is not indicative of future results. Each investor will receive individual statements from the funds’ administrator showing actual returns. Reference to an index does not imply that the Partnerships will achieve returns, volatility or other results similar to the index. The S&P 500 is a long-only index of primarily large-cap stocks used to represent the performance of the U.S. stock market. Greenlight funds take long and short positions and may invest in non-U.S. and non-equity securities, and therefore differ (often materially) from the composition of the index. The total returns for the index do not reflect the deduction of any fees or expenses which would reduce returns.

All exposure information is calculated on a delta-adjusted basis and excludes “macro” positions, which may include, but are not limited to, government debt, currencies, commodities, credit default swaps, interest rate swaps, volatility indexes, credit indexes and derivatives on any of these instruments. However, equity indexes and derivatives on such instruments are included in long/short exposure. The largest disclosed long positions represent individual issuers to which the Partnerships have the highest exposure. Greenlight, in its discretion and in the interest of investor protection, may exclude from this list any position that has not been disclosed but would otherwise be included, and instead include the Partnerships’ next largest position.

All weighting, exposure, attribution and performance contribution information is inclusive of positions held both directly and indirectly through the master fund, reflects estimates of the weighted average of such figures for investments by Greenlight Capital LP and Greenlight Capital Offshore Ltd. (excluding gold interests), and is the result of classifications and assumptions made in the sole judgment of Greenlight. All exposure calculations include the impact of month-end redemptions and subscriptions as of the first day of the following month.

Net performance contributions by attribution category apply an allocation of operating expenses, management fee and performance allocation to the applicable gross contribution. These pro forma net contributions are shown for illustrative purposes only to meet regulatory requirements. Different assumptions than those made by Greenlight can result in different net performance contributions at the category level. A description of the calculation methodology is available upon request.

Changes in quoted prices and market returns of publicly traded investments do not reflect the deduction of fees or expenses from the Partnerships. These fees and expenses are reflected in the net performance of the Partnerships.

The fund terms, performance returns, and portfolio characteristics reflected in this document are not indicative of future returns or portfolio characteristics and do not modify the terms of the funds as detailed in each fund’s Confidential Offering Memorandum. No guarantee or representation is made that Greenlight’s investment program, including, without limitation, its investment objective, diversification strategies or risk monitoring goals, will be successful. Investment results may vary substantially over time. No assurance can be made that profits will be achieved or that substantial or complete losses will not be incurred.

With respect to the investment strategy utilized by Greenlight, there is always a significant degree of market risk, including the risk of a complete loss of capital. Greenlight may employ speculative investment strategies, such as the use of leverage, short selling, derivatives trading and commodity interests trading, which involve the potential for significant risk of loss. Prospective investors should be aware that they will be required to bear the financial risks of this investment for an indefinite period of time. Prospective investors should read the more detailed Risk Factors contained in the Confidential Offering Memorandum.

The specific investments identified and described are not representative of all the positions held, purchased or sold, and in the aggregate, the information may represent a small percentage of activity. It should not be assumed that any position identified has been or will be profitable. There can be no guarantee that similar investment opportunities will be available in the future or that Greenlight will be able to exploit similar investment opportunities should they arise. The information presented is intended to provide insight into the noteworthy events, in the sole opinion of Greenlight, affecting the Partnerships. The opinions expressed represent the current, good faith views of Greenlight at the time of publication and are provided for limited purposes, are not definitive investment advice, and should not be relied on as such.

THESE MATERIALS SHALL NOT CONSTITUTE AN OFFER TO SELL OR THE SOLICITATION OF AN OFFER TO BUY ANY INTERESTS IN ANY FUND MANAGED BY GREENLIGHT OR ANY OF ITS AFFILIATES. SUCH AN OFFER TO SELL OR SOLICITATION OF AN OFFER TO BUY INTERESTS MAY ONLY BE MADE PURSUANT TO DEFINITIVE SUBSCRIPTION DOCUMENTS BETWEEN A FUND AND AN INVESTOR.

Original Post

**Editor’s Note: **The summary bullets for this article were chosen by Seeking Alpha editors.