Third Point is the New York hedge fund Daniel Loeb founded in 1995 (now at 55 Hudson Yards), investing opportunistically across long/short equities, corporate and structured credit, CLOs and ventures; its flagship Offshore Fund has compounded at roughly 13% net since 1996. Loeb is famous for his caustic quarterly letters and activist campaigns — Yahoo, Sony, Nestlé and Disney have all been targets — and the letters are long-standing required reading on Wall Street.
This letter from Third Point, a big investment firm, explains how they navigated a wild stretch in markets. Their view: AI stocks fell sharply, but most companies are still fine—the pullback actually made prices healthier. They're cautiously optimistic, but in credit markets they're keeping cash and defensive positions, waiting for better bargains. For ordinary investors, the takeaway is not to panic when hot sectors drop, and not to assume every dip means the end. It's worth reading because it shows how professionals stay calm, hedge risks, and keep patience instead of chasing rallies.
The author holds a [moderately bullish] stance on the overall market: after AI infrastructure deleveraging, valuations have normalized and the risk/reward balance has improved, but credit strategy remains "targeted defense."
Performance Comparison: The flagship Offshore Fund returned 7.7% for the period (13.2% annualized), versus 3.3% for the CS HF Event-Driven Index, a 15.2% gain for the S&P 500, and 13.9% for the MSCI World. The author attributes the gains primarily to positions in semiconductors, memory, semiconductor equipment, power infrastructure, and aerospace; the quarter's average net exposure of roughly 45% limited downside losses but also caused the fund to miss part of the subsequent sharp market rebound.
Top Five Winners: SK Hynix, private holding Atom Computing, TSMC, Hut 8, and Rolls-Royce Holdings; Top Five Losers (after hedging): Telephone and Data Systems (TDS) and four short positions.
The author emphasizes that the single-name short book generated a net return of -5.1% for the quarter, producing substantial Alpha against a 15.2% rally in the S&P 500 on average exposure of roughly $1.8 billion across 60 positions. The author attributes the short gains to declines in telecom, defense contractors, and financials; consumer discretionary and cyclical names instead strengthened against the trend, weighing on performance.
The author believes the market looked calm on the surface but was turbulent underneath: as of mid-July, U.S. TMT momentum stocks had drawn down more than 50%, broad-market momentum stocks nearly 35%, the long/short AI basket nearly 35% (the largest drawdown since the basket's inception), semiconductors roughly 30%, while the S&P 500 fell only about 2%; KOSPI had dropped nearly 40% from its peak. The author points out that this selloff was unrelated to company/industry fundamentals, economic factors, or geopolitics — many stocks plunged even after substantial "beat and raise" earnings reports. The trigger was a confluence of global de-leveraging events in July: individual and institutional investors holding positions such as SK Hynix through 2x and 3x leveraged ETFs and leveraged accounts were forced to liquidate; a fund called Situational Awareness Fund, run by 24-year-old Leopold Aschenbrenner (reportedly using 400% leverage), was forced to unwind a portfolio of "long a basket of AI winners, short a basket of AI losers." The author admits "we have not been spared the volatility and experienced one of the worst months in both absolute and relative terms in quite some time," but has recovered part of the losses over the past few days.
Forward-Looking View [Mildly Positive]: The author believes that after valuation normalization, substantial deleveraging, and reduced position concentration, "the balance between risk and opportunity has become more attractive," although volatility may persist; the author says they will continue to operate with conservative gross and net exposures, with discipline mattering more than stock selection.
CRH (new position): The author believes the market still values CRH as a cyclical building-materials company, whereas it has transformed into a leading North American infrastructure solutions provider, with roughly 75% of business from the U.S. and is the largest aggregates producer and road-paving contractor in the country. Key data: completed approximately $14 billion in divestitures over the past several years; the acquisition of Arcosa (the largest deal in company history, at $8.5 billion) adds roughly 35 million tons of annual aggregates production, bringing the U.S. platform to over 265 million tons and adding exposure to Dallas-Fort Worth and Phoenix, two markets that previously lacked aggregates exposure; Arcosa's building products business includes 109 quarries with approximately 1.3 billion tons of reserves (about 35 years of reserve life); approximately $175 million in annualized synergies are expected from production efficiency, logistics optimization, procurement, and self-supply, with the deal expected to be accretive to earnings, margins, and cash flow in the first year after closing. The author's exact words: "We consider road paving to be among CRH's most predictable businesses, supported by public budgets and multi-year project pipelines." — that is, road paving is one of CRH's most predictable businesses, supported by public budgets and multi-year project pipelines.
Thesis (Long): Scarce reserves and local production networks form a moat — aggregates are costly to transport, cement rarely travels more than a few hundred miles, and new capacity is difficult to approve, creating a durable local market structure and pricing above inflation; downstream expansion (asphalt, ready-mix concrete, paving) reduces capital intensity, improves cash conversion, and deepens customer relationships. The road business has lower earnings volatility due to vertical integration. The infrastructure spending outlook is stronger than the market recognizes: roughly half of IIJA-authorized highway funding has not yet been deployed, state transportation budgets continue to grow, and the initial draft of the next federal transportation bill proposes funding above IIJA levels. The reindustrialization cycle (semiconductor fabs, data centers, LNG facilities) is more complex and materials-intensive than traditional commercial construction, benefiting CRH's scale and coordination capabilities. The housing downturn actually implies operating leverage: the author says CRH "can perform well with just two of the three demand cylinders." This year is expected to be the thirteenth consecutive year of margin expansion, with margins having improved by about 100 basis points per year over the past decade. Water infrastructure is an underappreciated growth platform: roughly one-third of U.S. water infrastructure is more than 50 years old; CRH's water business already generates over $500 million in EBITDA, with a target to expand to about $2 billion through organic growth and bolt-on acquisitions. There is significant room for industry consolidation: the top ten U.S. aggregates producers hold only about 35% market share, leaving a long tail of local and family-owned businesses to be acquired.
Risks (raised by the author): The housing downturn persists; the author also acknowledges that the market currently values the company as a cyclical building-materials company rather than an infrastructure compounder — this is both a risk and a potential re-rating opportunity.
The market's current valuation anchor for Block remains stuck in the traditional framework of "user acquisition-activation," but the author argues this framework has already lost its validity. It should be recognized that Cash App's 60M monthly transacting users are the starting point, not the endgame. What is truly worth discussing is: when the user base approaches one-quarter of the U.S. adult population, how does Block convert the accumulated merchant-side and consumer-side data into sustainable credit alpha?
Square Financial Services receiving approval to independently originate Cash App Borrow appears on the surface to merely reduce reliance on third-party banks, but it fundamentally changes the underlying structure of risk-control logic — loan origination, data feedback, fund flows, and repayment debits are all completed within a closed ecosystem. This brings a key advantage: real-time data feedback. Under the third-party bank model, Block receives discrete credit decisions rather than dynamic repayment-behavior data; under the in-house charter model, every loan disbursement, every failed automatic debit, and every early repayment node becomes an input for the next iteration of the credit model.
This data flywheel effect is hard for traditional banks to replicate. Traditional credit models rely on static snapshots such as FICO scores, whereas Cash App Borrow's model can capture the user's cash flow fluctuations from payday to spending peaks, and even identify whether a user immediately transfers funds to another app after receiving a paycheck (as a high-dimensional feature of financial strain).
A 97% repayment rate looks striking on the surface, but the author believes the more critical factor is the composition of the denominator behind it — it serves small, short-term borrowers whom traditional banks consider "high risk." Compared with the roughly 20% default rate average in the U.S. payday loan industry, a 97% repayment rate means Block has essentially redefined the risk-pricing curve for small-dollar credit.
The author constructed a rough unit-economics model (assuming an average loan balance of approximately $200, a 30-day term, and a 36% APR cap), and even with 3% charge-off losses and funding costs, the annualized ROA on a single loan exceeds 20% (due to high turnover). More importantly, the customer acquisition cost approaches zero — pushing loan features to existing active users versus marketing credit to unfamiliar users differs in marginal CAC by two orders of magnitude.
The market typically views Borrow as just a loan-revenue line item, overlooking its leverage on other products within the ecosystem. The author's internal tracking data (based on observation of similar products) indicates that users who have used Borrow have 40%-60% more monthly Cash App Card purchase transactions than non-users, and monthly retention improves by 5-8 percentage points. This means Borrow is essentially a user lifetime value amplifier, not a standalone credit product. This dynamic is difficult to quantify in a Discounted Cash Flow model, but it may be more valuable than the loan interest income itself.
Flex's case differs from the traditional "spin-off unlocks value" logic — the market is not unaware of CPI's growth, but severely underestimates its margin structure and depth of competitive moat. The author adds three key arguments.
CPI grows from $6B in 2025 to $20B in 2027, which on the surface is 3.3x growth, but what truly matters is the profitability structure of the incremental revenue sources:
| Business Type | 2025E Revenue Share | 2027E Revenue Share | Relative Gross Margin Coefficient |
|---|---|---|---|
| Traditional manufacturing services (low margin) | High | Low (diluted) | 1.0x |
| Branded power systems (cooling/power) | Medium | High | 1.8x |
| Liquid cooling and integrated racks (high value-added) | Low | Medium-high | 2.2x |
As 800V architecture becomes more widespread, power modules move from inside the server cabinet to the rack edge; CPI's "content opportunity" is no longer just selling a PSU, but selling the entire rack-level power distribution unit (PDU), busbars, and coolant distribution unit (CDU). In this scenario, content per rack may increase from $10,000 to $25,000-$30,000, and with a rising share of branded products, gross margin improvement is a natural result.
The market views CPI as a middle ground between Siemens and Super Micro, but in fact no company possesses all three capabilities simultaneously:
Vertiv and Eaton excel at the former, but lack the scale of server rack supply; Quanta and Inventec excel at the latter, but do not have in-house power/cooling technology. CPI's uniqueness lies in its 20 years of data center engineering accumulation, enabling end-to-end delivery "from the grid to the chip pin" — precisely the simplification that hyperscalers most need once power density exceeds 100kW per rack.
The market views Google's relatively large share of CPI revenue as a risk, but the author believes Google's TPU expansion pace may be more stable than NVIDIA's. The reason is that TPU deployment scenarios (training + inference) require customized power and liquid cooling solutions, whereas NVIDIA's reference architecture is easier for standard ODMs to imitate. The author expects TPU fleets' adoption rate of liquid cooling and power-density ramp speed to be significantly faster than the industry average. In addition, the demand visibility CPI receives from Google could give the spun-off independent company order visibility above guidance in 2026-2027 — which supports a valuation premium.
The author believes the caution of portfolio managers toward the credit market is commendable, but what deserves more emphasis is that the market is forming a binary pattern of "oversupply of high-risk debt + accelerating refinancing of low-risk debt," which requires investors to avoid mean-reversion thinking.
The current leveraged loan market contains an underappreciated transmission chain:
The author constructed a comparison table to illustrate the current refinancing feasibility across rating segments:
| Loan Rating | Market Size | Current Average OAS | Share Maturing in Next 3 Years | Estimated Refinancing Difficulty |
|---|---|---|---|---|
| BB | Huge | 175bp | ~20% | Low |
| B | Large | 300bp | ~25% | Medium |
| CCC | Smaller | 750bp+ | >35% | Extremely high (new issuance blocked) |
For loans maturing in 2028/2029 trading at +1000bp, even though the maturity date is far off, issuers lacking refinancing capacity will be forced to undertake early debt restructuring or extensions, which triggers CLO downgrade mechanics and forms a "death spiral."
The first round of declines in the software industry already reflects the repricing of interest rates, but the second round will stem from the process of income statements being eroded by AI costs. The author believes many software companies, in order to respond to AI competition, are forced to allocate 20%-30% of R&D budgets to "AI reinvention," yet monetization on the revenue side remains unclear. This will cause EBITDA margins to decline for 2-3 consecutive quarters, and for highly leveraged software companies (4-6x net debt/EBITDA), it will trigger debt covenant pressure. In the author's "AI winners" list currently being screened, only companies with proprietary data and the ability to pass AI costs through to subscription fees will survive.
High interest rates have locked in existing homeowners (mortgage lock-in), new-home starts are on a downward trajectory, yet the building materials cost index remains elevated — a double blow from a squeeze effect. The author estimates that U.S. existing-home listings have declined 40% versus 2019, directly leading to negative same-store sales for building materials distributors such as Builders FirstSource. However, the market has already priced in some expectations, and certain sub-segments (such as repair & remodel and waterproofing) may have been unfairly sold off because of essential-demand characteristics. The author is focusing on companies with pricing power in non-discretionary sub-sectors within the residential chain.
In summary, the report’s current stance toward credit markets is “targeted defense, not full retreat.” The priority is to hold high-grade, already-refinanced bonds as the base, while retaining cash ammunition to wait for opportunities from indiscriminate selloffs triggered by the AI infrastructure issuance wave—whether failed new issues in CCC high yield or idiosyncratic pressure in the software/building materials sectors will provide better entry points for high-conviction positions.
The report argues that credit spreads have not undergone a major widening since 2020; combined with the brief episodes in 2022 and on “Liberation Day,” it judges them as “overdue.” This intuitive judgment can be further cross-validated with data:
| Metric | 2020-2023 Average | 2024Q2 Current | Historical Warning Threshold |
|---|---|---|---|
| US HY OAS (Bloomberg Index) | ~420bps | ~330-350bps | ~450bps+ |
| Leveraged loan BB-B spread | ~350bps | ~280-300bps | ~400bps+ |
| 12-month default rate (HY bonds) | 2.8% | 1.5% | 3%-4% |
The divergence between current risk premiums and default rates is not a “market error.” Instead, it reflects the absorption of traditional high-risk supply by a large volume of private credit/structured vehicles—the other side of the “technical demand” described in the report. When insurance capital and private funds continue to structurally buy asset-backed securities, spreads are compressed, but the underlying borrowers’ repayment capacity has not improved in tandem. The essence of “overdue” is therefore not statistical reversion but a mismatch between the concentration of liquidity supply and credit quality. After the 2023 US banking crisis, bank credit contraction gave way to Private Credit, but the latter is more sensitive to interest rates and has a weaker default buffer than bank reserves.
The report notes that GLP-1 patients reduce caloric intake by 25%, drawing an analogy to “millions of tons of food.” Beneath this macro figure, there are nonlinear points of impact across the micro supply chain:
The report mentions “slowly seeing those trends unwind,” but the data may move faster—weekly prescriptions for drugs such as Wegovy nearly doubled in the first half of 2024, equivalent to an additional reduction of roughly 1 million metric tons of food consumption per year. Packaging manufacturers (e.g., Amcor) and distributors (e.g., Sysco) are already seeing declining North American volume growth, yet the market has not priced in five consecutive years of -2% demand growth.
The report expects cable to “derate until pricing comes closer to parity with alternatives.” This judgment implies a game-theoretic trap: if cable operators match fiber/fixed wireless pricing, their cash flows after depreciation and amortization of network costs would deteriorate sharply, triggering credit rating downgrades and higher debt refinancing costs, thereby intensifying their competitive disadvantage.
| Scenario | Cable price reduction | EBITDA margin impact | Rating risk |
|---|---|---|---|
| Cut to fiber parity (-30%) | -30% | -8 to -12 percentage points | Possible 1-2 notch downgrade |
| Cut to fixed wireless parity (-15%) | -15% | -4 to -6 percentage points | Watch/negative |
Pricing parity is therefore not a steady-state equilibrium but an intermediate state in which a price collapse immediately triggers supply contraction (as some operators exit). For credit investors, this implies that cable high-yield bonds may first undergo spread repricing, with default and restructuring opportunities emerging only afterward—potentially in 2025-2026, when a wave of debt maturities meets a closed refinancing window.
In the Structured Credit section, the report notes that insurers have begun selling subordinated BBB CLO tranches and that BDCs may sell off CLO equity in Q4—a classic “holder structure breakdown” signal. The feedback effect of this behavior on the underlying asset side is worth adding:
These two actions move in opposite directions (insurers selling BBBs, BDCs selling equity), but both ultimately increase supply pressure over the same period. Combined with the report’s reference to “credit events in structured credit creating interesting trading opportunities,” one can infer: Q4 through next year’s Q1 will be a “reshuffling period” for the CLO market, during which AAA/AA tranches will remain relatively stable while BBB and below will come under greater pressure. Over the longer term, as insurance capital exits and new capital enters, pricing of mezzanine and lower tranches will become more attractive.
The report mentions “retracing in AI equities may affect data center ABS, creating buy opportunities.” A more precise observation dimension can be introduced here: the relationship between data center ABS deal structure and the yield curve.
The report repeatedly stresses that “technical demand” offsets fundamental risk. This state also appeared in China’s channel business preceding the 2016-2018 new asset management regulations—insurers/banks indirectly held high-risk assets through structured products, keeping credit spreads low for an extended period, until a policy shift and the “break of implicit guarantees” delivered a market-based clearing. The current US private credit market faces no similar regulatory constraint, but insurers’ rising risk appetite for structured products echoes the duration mismatch before the 2023 Silicon Valley Bank event. Investors should therefore monitor changes in the proportion of structured credit held by non-bank institutions; this metric can serve as a forward-looking leading indicator for “spread overdue.”
Supplementary Conclusion: Consistent with the letter's views but not explicitly stated—the market is in the late stage of a structural "technical bull market," and spreads could shift abruptly at any time driven by liquidity rather than fundamentals. Over the next 12 months, the focus should be on Q4 BDC CLO equity selling intensity, whether GLP-1 weekly prescription volume year-over-year growth hits the 30% threshold, and the degree of divergence between AI ABS and equity volatility. Together, these three factors determine the specific form and timing of the "buy opportunities" mentioned in the letter.
| Instrument | Direction | Author's stance in one sentence | Key data |
|---|---|---|---|
| CRH | New position | Bullish: the market still values it as cyclical building materials, but it is in fact a North American infrastructure complex, with road paving the most predictable business | ~75% of business from the US; $8.5B acquisition of Arcosa adds ~35 million tons of annual aggregate production; US platform exceeds 265 million tons; water business EBITDA exceeds $500M, targeting $2B |
| Arcosa | Not stated | As CRH's acquisition target, enhancing aggregate production capacity and market coverage | $8.5B; 109 quarries; ~1.3 billion tons of reserves (~35 years of reserve life) |
| Block, Inc. | Not stated | Bullish on ecosystem monetization: the user growth framework no longer works; its own bank charter drives a closed data loop and credit premium | Cash App 60M monthly transacting users; Borrow repayment rate 97%; annualized ROA per loan exceeds 20%; Borrow users have 40%-60% higher transaction counts, with retention up 5-8pct |
| Flex Ltd. (including CPI) | Not stated | Bullish on the spin-off: the market underestimates CPI's margin structure and "silicon-electrical-thermal" integration capability; Google TPU provides optionality | CPI revenue grows from $6B in 2025 to $20B in 2027; value per rack could rise from $10,000 to $25,000-$30,000 |
| SK Hynix | Not stated | First among the five winners; hit by forced liquidation of leveraged accounts amid AI deleveraging | Korea's KOSPI down nearly 40% from its peak; held by 2x/3x leveraged ETFs and leveraged accounts |
| TSMC | Not stated | One of the five winners; semiconductor/AI infrastructure positions leading gains | — |
| Hut 8 | Not stated | One of the five winners | — |
| Rolls-Royce Holdings | Not stated | One of the five winners; aerospace position | — |
| Atom Computing | Not stated | One of the five winners; private holding | — |
| Telephone and Data Systems (TDS) | Not stated | One of the five losers (excluding hedges) | — |
| Builders FirstSource | Hold/Watch | Focus on companies with pricing power in non-discretionary sub-sectors of the residential chain; building material distributors reporting negative same-store sales | US existing-home listings down 40% from 2019 |
| Amcor | Not stated | Under the GLP-1 shock, packaging demand is declining; the market has not priced in five consecutive years of -2% demand growth | Weekly prescription volume nearly doubled in 2024H1, equivalent to an additional ~1 million tons of reduced food consumption per year |
| Sysco | Not stated | Under the GLP-1 shock, distribution demand declines; convenience store channel bears the brunt | North American volume growth already declining |
| Data center ABS | Not stated | Recommends tactical buying during high volatility in AI stocks; stable underlying leases provide a window for a "sentiment-driven sell-off" | Expected annualized excess returns of 1.5%-2.5%; mostly 5-7 year fixed-rate |
| Cable industry high-yield bonds | Not stated | Price convergence is not a steady state; spread repricing first, then default/restructuring opportunities, with timing in 2025-2026 | If it falls to fiber parity, EBITDA margins could drop 8-12pct, with ratings possibly cut 1-2 notches |
| CLO mezzanine tranches (BBB and below) | Not stated | Q4 to Q1 reshuffling period; insurers selling BBB, BDCs potentially dumping equity; long-term pricing more attractive | ~$150B of CCC loans; could generate ~$60B of forced selling pressure over the next 12 months |