Third Point Q3 2026 Investor Letter
YurolaitsAlbert/iStock via Getty Images
Dear Investor:
During the Third Quarter, Third Point returned -6.8% net in the flagship Offshore Fund.
The top five winners for the quarter were Atom Computing Inc. (a legacy private position), Warner Bros Discovery Inc (WBD)., Meta Platforms Inc (META)., Indra Sistemas SA (ISMAF), and Bausch Health Cos Inc (BHC). The top five losers for the quarter, excluding hedges, were SK Hynix Inc (SKHY)., Flex Ltd (FLEX)., MasTec Inc (MTZ)., CRH PLC (CRH), and TTM Technologies Inc (TTMI).³
Portfolio Updates
Benign index performance masked significant turmoil in markets during the Third Quarter. The Morgan Stanley TMT Momentum⁴ factor recorded its largest drawdown since inception, falling more than 50% from its late-June peak to mid-August, while the broad momentum factor registered its highest volatility in 45 years. With two months left in the year, the Goldman Sachs High-Beta Momentum⁵ basket has already recorded 39 distinct one-day moves of more than 5%. Outside of 2020, which had 41, that is more than any year since 1999.
The market’s drawdown had many causes, among them renewed focus on AI capex and how it is being funded, the rise of open-source models, and concerns around growth slowing at frontier labs. Crowding and leverage fueled an explosive reaction during the depths of July and August, from the unwind at Situational Awareness to retail liquidation of leveraged memory-stock ETFs in Korea. Many of the names that drove our Q2 returns—including SK Hynix, Flex, and MasTec—were caught up in the rout.
Oil and rates also contributed. The June ceasefire with Iran collapsed in the first week of July, the Strait of Hormuz closed, and by mid-September the US blockade of Iranian ports and attacks on tankers had taken Brent from roughly $72 on July 1 to a peak of $109. Oil-fed inflation prompted higher rates: headline CPI re-accelerated to 3.7%, and new Fed Chairman Kevin Warsh raised the funds rate to 3.75–4.00%, its first hike since 2023.
Part of the move in rates is inflation, and part of it is inflation expectations: after five years of “temporary” supply shocks stacked end to end (COVID, Ukraine, tariffs, DRAM shortages, Iran), investors appear to be seeing high inflation as permanent. The less-discussed part is supply. Nation-sized companies are now competing with sovereigns for the same bond buyers to fund their AI capex plans, which is one of the reasons the 10-Year is above 5% although the economy is still growing. In Q3, anything sensitive to fuel, rates, or housing was sold with little discrimination, exchange hedging volumes hit records, and the divergence between indices sitting near their highs and the potential violence underneath—in factors, in breadth, in rates—underscored just how tricky this market has been to navigate.
We continue to see a tug of war play out. On one side: a strong earnings season (S&P 500 earnings tracking close to 30% growth for Q3, the third straight quarter above 25%), aggregate consumption of AI usage that is holding up, and what we still believe are the early innings of AI adoption—with the Muse launch and growth metrics disclosed in Anthropic’s S-1 being the latest evidence that usage is compounding faster than the market expected. On the other side: the 10-Year hovering north of 5%, an unresolved situation in the Middle East, oil prices that remain high, and investors who are being a lot more circumspect about AI as a monolithic investment theme.
The AI debate has moved from whether demand is real to whether the capital can be funded and earn a return. The bear math is hard to dismiss; consensus has hyperscalers and frontier labs spending $5–7 trillion between 2025 and 2030. At a low teens hurdle and reasonable assumptions around asset depreciation, spenders need to generate $2-3 trillion of incremental annual AI revenue, a considerable sum in the context of $6 trillion in global IT spend today. The strain is now showing up in credit, not just equities: several data-center financings in the high yield and leveraged loan markets carry double-digit yields, top-rated data-center ABS that priced below 5% in January yields closer to 7% today, and CDS on Oracle (ORCL), Meta, and Nvidia (NVDA) hit record levels in late October. Some banks and insurers are approaching their exposure limits to the sector, and the financing the neoclouds still need increasingly depends on backstops from the same hyperscalers and chipmakers that already carry the largest capital needs of all, so credit investors are starting to price AI risk back onto balance sheets the market had assumed were insulated.
While the quantum of capital being spent seems hard to justify, there are several reasons to believe that total projected spending is overstated and buildout costs will be significant but more rational and stretched out over more years. First, there are tangible signs that AI is addressing markets far larger than just global IT spend, potentially as large as the $60 trillion knowledge worker TAM. Second, the capacity being built today is quickly being consumed: the revenues at the largest public cloud platforms are accelerating to rates not seen in many years and their backlogs are growing at triple-digit rates. The two leading frontier labs are, in aggregate, generating well over $100 billion in run-rate revenue and still growing at rates unprecedented at this scale. Lastly, economics on infrastructure also appear sound: according to the hyperscalers who are financing the majority of the spend, AI servers and racks are bought months ahead of contracted demand and pay back in under three years. We still think the world is structurally short the building blocks of compute and power, but with open-source models now within six months of the frontier and capital markets signaling that the next several trillion dollars of buildout will not be financed on the terms the first several hundred billion were, our view is that who captures the economics matters a lot more than it did in June. Investors are no longer underwriting AI revenue; they are underwriting who pays for it.
Our base case on the Middle East is a muddle-through stalemate in which oil stays elevated and the economy slows without breaking. The tail risk is real damage to Gulf infrastructure, which appears to be something neither side has an incentive to cause, while the underappreciated path is a deal after November 3. We believe Iran’s leverage is eroding, the administration wants the war over, and China wants cheaper oil. More broadly, the midterms look like a clearing event for political uncertainty. In the meantime, anxious investors are selling the small, the cyclical, and the rate-sensitive, while strategic buyers keep paying takeover premiums for the same assets.
In light of these concerns, in a market that depends on the successful IPOs and continued growth of two primary frontier model companies, we can see a wide range of scenarios playing out from here from optimistic to apocalyptic. While we don’t believe that the current data center build out shares the economic deficiencies of the infrastructure buildout during the dotcom bubble, we see signs of excess all around us, particularly the frenzy in private markets like so-called “seed” rounds for companies that are pre-revenue and pre-product at $1 billion plus “unicorn” valuations. Similarly, frenzies for allocation to anointed so-called “growth-stage” companies and a proliferation of amiable but financially unsophisticated individuals who we’ve seen selling access to private deals via SPVs that charge outlandish fees.
This does not take away from the fact that some extraordinary companies are being built and that AI technology will forever change the way we conduct business, even our own. We believe that this is a good time to pause and take account of who we expect the winners and losers will be. We are also of the view that such uncertainty will inevitably lead to lower multiples given the fragility of markets, as evidenced last week where companies from Service Now (NOW), Intel (INTC) and Reddit (RDDT) could lose a third of their market capitalization in a single day.
All of this calls for significantly lower gross and net exposures and a more selective approach, and that is where we have taken the portfolio. We have cut our semiconductor and data-center-adjacent positions, and as well as the hyperscalers, which we had earlier added to based on our assessment of their strong business models backed by recurring revenue, and reasonable valuations. We were broadening the portfolio already mid-year, seeking more balance, and that continued through the quarter with positions in financials, live events, industrials, commercial aerospace, and consumer-facing companies that we feel are equipped to navigate an air pocket or are actively using it to gain market share.
On the short side, our gross equity short exposure (~21% of NAV) is entirely single-name, and we think dispersion will continue to present opportunities here. Current themes include GLP-1 losers, cable, P&C insurance, and housing, to name a few. We reduced short exposure to AI disruptees—software, information services, IT services—because those shorts were amplifying our AI longs and creating basis risk rather than hedging it. We had some success shorting “AI overearners, ” but it is difficult to press those given the dramatic recent de-rating.
With interest rates where they are and a roughly 20x forward multiple underpinned by 30% earnings growth, the index is not where we want our risk. We expect that the opportunity lies underneath it: in the dispersion between the companies that will actually earn a return on the AI buildout and those that are merely levered to it, and in the parts of the market that were sold for reasons that had nothing to do with their fundamentals. Having expressed such caution before just prior to market rallies, we are being equally cautious on the short side as we are on the longs. We are willing to miss the early stages of a market rally or a rotation back into tech in order to preserve capital, exercise patience, and wait for investments that better fit our event-driven framework. Our suspicion is that we could be getting close to an environment like we had in March 2020, or towards the end of the GFC, where the next wave of opportunities will come from credit markets.
Atom Computing
Atom Computing was our top winner during the Third Quarter. We first invested in late 2021, when we led the Series B financing and joined Atom’s Board of Directors. Based on Atom’s technical progress and commercial traction, we subsequently led the Series C which closed in early 2026. Atom Computing is led by its CEO Benjamin Bloom who founded the company in 2018 to pursue the Neutral Atom Modality for Quantum Computing. Ben holds a bachelor’s degree from MIT and PhD in physics from the University of Colorado-Boulder where he helped develop one of the most accurate atomic clocks ever constructed. Of the seven primary approaches being pursued to build practical, scalable, fault tolerant Quantum Computers, Atom employs the Neutral Atom modality. Our research led us to Atom Computing as we determined that the Neutral Atom approach’s combination of practical scalable design, high quality qubits, noise immunity, and qubit longevity made it one of the most likely to achieve the goals desired by governments, enterprises, and research facilities. Since our initial investment in 2021, we have partnered very closely with the company, been supportive of their organizational development, and have been pleased with their achievements of milestones (from 100’s of qubits in 2021, to 1200 qubits, to their expected announcement of the 10,000 qubit system shortly).
Atom Computing raised a Series D-1 financing (at a stepped-up valuation) in July 2026 based on their technical and commercial progress as well as the announcement (May 2026) of Atom being selected to receive a $100 million investment by the US Department of Commerce under the CHIPS Act. Last month, Atom Computing was selected as one of four companies announced for DARPA’s QBI (Quantum Business Initiative) Stage C awards. The four were Atom (private), Diraq (private), IBM (IBM) (public), and IonQ (IONQ) (public). Note that while Atom was one of four to receive DARPA-C, it is one of the two that are private companies, and the only company based on the Neutral Atom Modality thus far awarded. DARPA-C awarded Atom Computing potential non-dilutive funding up to $300 million (contingent on hitting specific development milestones over the next 36 months). The QBI program is designed to rigorously verify and validate whether any quantum computing approach can achieve utility-scale operation by the year 2033 (with near-term deliverables sooner). The recent markup was primarily driven by news of a major sector investor for its new Series D-2 financing, which will close in the coming weeks at a meaningfully higher valuation than the Series D-1 round (and over 28x the valuation where we originally entered the investment in 2021).
APi Group (APG)
APi Group provides fire protection, electronic security, elevator, and specialty infrastructure services. The business is built around inspecting and maintaining systems that customers need to keep buildings safe and operating. More than half of revenue comes from recurring inspection, service, and monitoring, much of it driven by regulatory requirements. We believe APi has the ingredients of an attractive long-term compounder: resilient demand, opportunities to improve margins, and a fragmented market in which to reinvest its cash flow.
Most of the value sits within Safety Services, which accounts for roughly three-quarters of segment profits. Fire protection is the largest business, and APi holds the leading position in North America. The work is local, relatively small-ticket, and recurring. APi can inspect and service equipment from any manufacturer, allowing it to manage a customer’s needs across different systems and locations. Its scale supports purchasing and service capabilities, while local branches retain the customer relationships.
APi’s inspection-first strategy is what makes the economics interesting. Every dollar of inspection revenue generates an additional $3–4 of annual service work and gives APi a repeatable way to deepen customer relationships and grow without relying solely on new construction. North American inspection revenue has grown at a high single digit rate or better for 25 consecutive quarters.
Elevator maintenance is a recently added business platform with similar characteristics: required inspections, recurring service, a fragmented provider base, and aging equipment that needs ongoing attention. It also creates opportunities to sell additional services to existing building customers. Specialty Services adds exposure to infrastructure and industrial investment, including data centers through fiber, HVAC, and fabrication work. These projects provide additional growth and can establish relationships that lead to future service revenue.
Acquisitions give APi another way to expand this model. Management estimates an addressable market of approximately $160 billion, much of it served by smaller operators. APi has completed nearly 150 bolt-on acquisitions over two decades, historically paying approximately 6-8x EBITDA for these businesses. We think today’s opportunity is to improve purchasing, expand service offerings, and bring greater operating discipline to an established local franchise. Chubb Fire & Security demonstrates the potential on a larger scale. Its EBITDA margins increased from below 10% at acquisition in 2022 to approximately 15% today, bringing the effective acquisition multiple down to high single digits.
APi’s margins and recurring revenue mix have steadily marched upward. Management’s 2028 objectives include an adjusted EBITDA margin of at least 16% and more than 60% of revenue from inspection, service, and monitoring. The company’s cash generation provides capacity to keep investing: pro forma net leverage stood at 2.4x as of 3Q, below management’s 2.5-3x target range despite substantial first-half acquisition spending. We see a multi-year runway for double-digit earnings per share growth, and the stock trades ~18x our FY28 EPS.
Corporate Credit Update
The corporate credit book gave back its strong second-quarter performance in the Third Quarter. A handful of positions contributed positively, with shorts in telecom, building products, and healthcare driving gains. On the long side, Bausch Health performed strongly on continued solid operating results and a favorable patent ruling. Our largest individual detractor was the GSE preferreds. At current levels, the market appears to be pricing in only a 15% probability of an IPO, which strikes us as too low.
AI infrastructure positions were the largest detractors from performance. We continue to see rapid AI adoption and seemingly insatiable demand for compute, but sentiment toward the sector has deteriorated sharply for several reasons.
First, concerns about eventual overbuilding are mounting. Meta’s announcement late in the Second Quarter that it might sell excess compute capacity—if it ultimately had any—was interpreted by the market as a signal that a large, well-capitalized competitor could resell significant compute supply. In July, SpaceX guided year-end 2027 compute capacity “closer to 10 GW than 5, ” which was initially met with skepticism. However, Elon continues to reiterate this target and recently provided more details on the build-out during the 3Q earnings call.
Second, opposition to data centers has intensified. Demand for compute is strong, but local communities increasingly resist hosting the infrastructure required to support it. This issue has found traction on both the left and right, and, as elections approach, political rhetoric has grown more hostile. The market is increasingly concerned that new state and local leadership will seek to impede data center development. Ironically, restrictions on new construction would increase the value of the completed powered shells, a distinction the market has so far largely overlooked.
Third, an unparalleled wave of investment-grade and high-yield issuance to finance AI infrastructure is creating significant technical pressure on credit markets, particularly against a backdrop of rising interest rates. Goldman Sachs expects hyperscaler debt issuance will increase ~60% YoY in 2027 and eclipse $400 billion. More impactful will be the ~$300 billion of expected global project finance issuance, up 2-2.5x YoY, to fund powered shell construction and chip purchases. Many of these deals will require guarantees or credit support from IG counterparties.
AI infrastructure spending as a percentage of GDP is now approaching levels last seen during the railroad boom. While some of this capital is funding model training, a substantial portion is going toward the construction of compute infrastructure. The resulting investment boom has driven an extraordinary increase in debt issuance, weighing on credit valuations. We are already seeing the issuance tsunami pressure spreads with hyperscaler 10-year spreads +50bps YTD and HY Project finance bonds ~250bps wider since the tights in June.
Basic economics suggests that this demand for capital will ultimately be constrained by higher financing costs or declining returns on investment as capacity catches up with demand. Paradoxically, the exceptionally attractive returns currently available on some of these investments—including leases with investment-grade counterparties that generate ROIC in the teens—may be accompanied by further declines in the prices of the securities financing them. As long as the economics of new investment remain compelling, we consider it likely that issuers will continue to access the capital markets. The marginal buyer, not the average economics of the underlying investment, ultimately sets the price of the securities being issued.
Over time, in a frictionless commodity market, the cost of capital would converge with the return on invested capital. We believe, however, that meaningful differences in execution and jurisdiction will ultimately drive greater dispersion in performance across these issuers. For now, the market largely treats them as interchangeable. This is creating mispricing in individual securities, but we believe the near-term path of least resistance remains lower. Investors are effectively treating some of these issuers as though their securities were “death spiral preferreds, ” subject to perpetual dilution as additional capital is raised.
The irony is that the underlying assets may ultimately prove enormously valuable. When capital markets eventually close to these issuers or the cost of financing becomes prohibitive, we expect compelling opportunities to emerge for investors willing to distinguish between the winners and losers.
Amara’s Law states that we tend to overestimate the impact of a technology in the short run and underestimate it in the long run. We believe the AI trade is transitioning from the former to the latter: investors are beginning to underestimate AI’s long-term potential after initially overestimating its near-term impact.
Numerous anecdotal examples demonstrate meaningful productivity gains from AI, yet the evidence remains less apparent in aggregate economic data, lending some credence to the bubble thesis. We remain confident that demand for compute will continue to grow, although the ultimate distribution of that demand between open-source models and proprietary large language models remains uncertain. Regardless of which models prevail, the need for compute infrastructure should persist, and we believe data centers will benefit.
We believe hyperscalers (and certain related project finance structures) currently offer the most attractive way to maintain exposure to the theme. We do not expect issuance fears to materially change their cost of capital: the incremental financing needs are relatively modest compared with their balance sheets and cash flow generation. Hyperscaler access to relatively inexpensive capital provides a significant advantage as the market works through the current wave of infrastructure investment and financing.
We had hedged our credit positions in infrastructure with short equity positions and have since unwound significantly all of this exposure and continue to find attractive relative value shorts in BB-rated data center bonds. The technical as well as range of fundamental outcomes is creating more dispersion in AI infrastructure credit.
During the quarter we made a convertible loan investment in Nscale, a neocloud. Although volatility and uncertainty remain elevated in the sector as investors weigh the aforementioned concerns about overall AI demand and financing needs, it has also created what we believe are attractive investment opportunities. The Nscale loan has the downside protection of debt (at an attractive yield to maturity relative to public comps) and provides equity upside in our base case in what we see as an improving idiosyncratic outlook.
Execution, access to power, and cost of capital will separate winners from losers in this sector in our opinion. Nscale recently hired an experienced data center planning and construction team, and we expect this new group to support delivery of its pipeline as forecast. The projects largely consist of vertically integrated data centers where Nscale plans to own the powered shell and the GPUs, which should drive higher margins and extend the duration of cash flows. With respect to access to capital, in our assessment, the balance sheet is stronger than peers given limited parent-level unsecured debt and Nvidia (a significant equity owner) has committed to provide guarantees that should help Nscale finance compute projects at a reasonable rate.
We believe our instrument is structured with attractive asymmetry and the conversion into equity will be struck at a discount to the level at which Nscale goes public. As their cost of capital goes higher due to the sector massively de-rating, our potential return increases since we are buying in at a lower valuation.
Structured Credit Update
Across credit markets in Q3 2026, the rate selloff and start of credit spread widening introduced volatility after a relatively benign first half of the year. Our structured credit performance was essentially flat over the quarter, with trading gains in shorter duration consumer ABS and the monetization of segments of our residential mortgage exposure counteracting the move in rates. While the rate selloff extends mortgage duration in our portfolio, our current securitizations have a cost of financing close to current Treasury rates enabling a steady current cash yield to our exposure.
As we see volatility increase across credit, we are constructive on the following sub-asset classes through year-end:
1) Residential mortgage securities: When rates first sold off with a 60bps move in September alone, 30-year fixed mortgage rates moved north of 7%. Non-bank mortgage originators had existing inventory on their balance sheets that were immediately subject to pricing pressure as the originated coupons were too low and their ability to generate mortgage volume going forward is challenged as the demand for mortgages at these rates is much lower. While we expected these originators to sell loans at a discount, many of the portfolios did not trade and we have seen the new issue supply of residential mortgage securitizations grow, pushing spreads wider on already higher base rates. We purchased BBB exposure at an 8.5% yield that was at a high 6% yield a few weeks earlier. Over the next six weeks, we believe more opportunities may emerge to add new issue and secondary exposure north of 10%.
2) Leveraged loans and CLOs: While the FOMC hike meant SOFR would be at an elevated level for longer, the cost of debt servicing for leveraged loans became more expensive to sustain for longer. Within CLO portfolios, leveraged loans first rallied despite growing concerns around LME maturity walls in 2027 and 2028. We are now seeing widening in leveraged loan spreads and opportunities for wider trading CLO tranches and potential repricings
3) Consumer ABS: Within this segment, the subprime cohort and certain vintages in specific consumer asset classes are under pressure. We have seen impairment in the student loan market where 2023-2025 vintage securitizations are likely to take actual credit losses while subprime auto loan delinquencies continue to rise. We expect the current market opportunity set to provide liquidity in the next few weeks given broader volatility, but we anticipate a more pronounced credit cycle in the coming year as higher rates and headwinds for the consumer create corrections in the credit markets.
Business Updates
Wally Cook joined Third Point in 2026. Prior to joining Third Point, Mr. Cook was an Associate at Veritas Capital and began his career in the Financial Institutions investment banking group at Goldman Sachs. Mr. Cook graduated Phi Beta Kappa and summa cum laude from Dartmouth College with an A.B. in Government.
Sincerely,
Daniel S. Loeb
CEO
Editor’s Note: The summary bullets for this article were chosen by Seeking Alpha editors.