In this article
← All articles

Dan Loeb's Third Point: the Q3 2026 AI buys and three decades against MSCI World


Executive summary

Third Point's flagship Offshore Fund returned -6.8% net in the third quarter of 2026 while MSCI World net gained 1.88% in US dollars, according to Dan Loeb's letter of 2 November. In the same quarter the firm's 13F added Meta, Nvidia and Microsoft; Nvidia's share price then fell 31.9% from 30 September to 17 November.

The fund's reports show it had already cut net exposure by 31 October, so that price fall is not a measure of the fund's loss. These headline returns are for the original series, which holds legacy private investments and is closed to new investors; the New Series open to them returned -8.6% in the quarter.

The longer record is the main finding. By Third Point's own figures the Offshore Fund compounded at 13.2% a year net from December 1996 to June 2026, against 8.5% for MSCI World (TR). The yearly path we could verify, the investment return of the London-listed feeder from 2008 to 2024, compounded at 7.72% against 7.15% for MSCI World net, with the same Sharpe ratio of 0.41. That lead was earned before 2015; from 2015 to 2024 the feeder trailed by 3.99 points a year.

Copio's assessment: Third Point compounded strongly in its earlier years, but its record shows no growth advantage over a cost-free world index in the last decade. Its annual returns moved closely with world equities (correlation 0.89), and it lost more than the index in 2018 and 2022. On the evidence we have, Third Point is a concentrated equity-and-credit bet, not a crash shelter or a diversifier.

What is in the fund?

At 30 September 2026 the Third Point Master Fund held long positions worth 121.0% of its net asset value and short positions worth 27.8%: 64.8% long and 24.1% short in equities, 46.0% long in credit and 7.1% in private investments. Those exposures come from the September performance update, p. 1-2. The flagship Third Point Offshore Fund, Ltd. invests through this master fund.

Share of net asset value30 Sep 202631 Oct 2026
Equity long / short / net64.8% / -24.1% / 40.7%45.1% / -19.8% / 25.3%
Equity net, beta-adjusted to the S&P 50065.2%25.2%
Credit long / short (corporate and structured)46.0% / -3.6%43.0% / -6.5%
Private investments7.1%6.8%
Whole portfolio long / short / net121.0% / -27.8% / 93.2%98.1% / -27.6% / 70.6%
Equity positions long / short39 / 6729 / 51

Source: September and October 2026 performance updates, p. 1-2; interest-rate, currency and some macro positions are excluded from these exposure figures.

Credit is the second engine. At 31 October the book held 15.1% of net asset value in residential mortgages, 15.5% long in high-yield bonds and 6.0% in consumer asset-backed securities. Its top corporate credit longs included SoftBank Group, the neocloud Nscale, Bausch Health, Peraton and ION Platform Finance. On 30 June 2026, 15% of the master fund's assets were valued as Level III, meaning with unobservable inputs rather than market prices.

The Offshore Fund had $3.78bn of net assets at the end of 2024, according to its audited 2024 financial statements, p. 3 and note 3. Investors pay a management fee of 1.5% to 2.0% a year, depending on share class, and an incentive allocation of 20% of annual gains. Redemptions are monthly, quarterly or annual by class, and the board may cap them at 20% of the fund's net asset value in any quarter.

What the Q3 2026 13F shows

Third Point LLC's 13F for 30 September 2026, filed on 16 November, reports 39 rows worth $4.59bn; Meta is the largest at $475.0m, or 10.35% of the reported total. A 13F lists US-listed long equities, convertible notes and listed options across all accounts Third Point LLC manages with investment discretion. It leaves out short sales, most credit, private holdings and swaps, so its weights are not the fund's weights.

30 Sep 2026ValueShare of 13F30 Jun 2026
Meta Platforms$475.0m10.35%not reported
Unum Group$438.7m9.56%not reported
Nvidia$342.6m7.46%not reported
CRH$235.2m5.12%$234.9m
Block$233.8m5.10%$194.5m
Microsoft$230.8m5.03%not reported
Amazonnot reported
$417.1m
Warner Bros. Discoverynot reported
$533.2m

Source: 13F-HR filings, accessions 0001040273-26-000004 and 0001040273-26-000003; all rows for both quarters (CSV). The ten largest rows make up 61.0% of the September total, by our calculation.

Meta, Nvidia, Microsoft, TSMC and Alphabet together account for 29.5% of the reported September value. The June filing's equivalent group, then Amazon, Alphabet and TSMC, accounted for 23.1%. The filing also lists put options on 61,000 shares of the SPDR S&P 500 ETF, reported at $46.5m. That is $762.63 a share, the ETF's closing price on 30 September, so the row shows the value of the shares the puts cover, not the premium paid.

Two dated reports qualify the "new buys" reading. Nvidia and Microsoft were already among the fund's top five gross equity longs at 31 December 2025, according to the December 2025 update, so they were sold and repurchased, or held in a form the June 13F does not show.

Amazon was a top-five equity long again at 31 October 2026, a month after it was absent from the 13F.

Managers can also ask the SEC to keep a position confidential for a time, another reason a holding can be missing from a filing.

What happened after 30 September

Nvidia closed at $155.46 on 17 November, 31.9% below its 30 September close of $228.38, the price implied by the 13F row. Meta fell 14.7% and Microsoft 1.3% over the same dates, using Yahoo Finance closing prices. Applying those falls to the 13F values would be misleading. By 31 October neither Nvidia nor Meta was among the five largest equity longs, and Meta was among the month's five biggest losers.

Why does Loeb build the fund this way?

Third Point describes an "opportunistic, event-driven" approach across equities, corporate, structured and private credit and venture capital, and Loeb's Q3 letter says the fund has cut its AI exposure, including the largest cloud companies, and moved risk away from the index. The strategy page says: "We employ an opportunistic, event-driven approach to finding investments in select strategies while managing individual security and market risks." It adds: "We also have a long history of short-selling individual securities."

The Q3 2026 letter frames the AI question as one of financing.

"Investors are no longer underwriting AI revenue; they are underwriting who pays for it," Loeb writes on p. 4.

The same letter describes a sharp retreat. "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," it says on p. 5. It has cut its semiconductor and data-centre-adjacent positions and the hyperscalers, the largest cloud companies, "which we had earlier added to based on our assessment of their strong business models backed by recurring revenue, and reasonable valuations".

In the credit section (p. 11) the letter says that "hyperscalers (and certain related project finance structures) currently offer the most attractive way to maintain exposure to the theme". The context is lending: it argues that their extra financing needs are modest compared with their balance sheets and cash flow.

On the market as a whole, citing interest rates and a forward multiple of about 20 times earnings: "the index is not where we want our risk." Loeb also says the fund is "willing to miss the early stages of a market rally or a rotation back into tech in order to preserve capital".

Loeb treats the short book as a source of returns as well as a hedge. The letter puts gross equity short exposure at about 21% of net asset value, "entirely single-name", and says "dispersion will continue to present opportunities here". It explains why some shorts on AI-disrupted sectors were reduced: "those shorts were amplifying our AI longs and creating basis risk rather than hedging it."

In credit, Loeb suspects "the next wave of opportunities will come from credit markets", comparing the moment with March 2020 and the end of the financial crisis. He describes a convertible loan to Nscale made in the quarter as having "the downside protection of debt", converting into equity at a discount to the price at which Nscale goes public.

The letter attributes the Q3 loss to positions that had driven the second quarter. Its five largest Q3 losers, excluding hedges, were SK Hynix, Flex, MasTec, CRH and TTM Technologies; Meta was among the five largest winners. In the credit book, "AI infrastructure positions were the largest detractors", while the corporate credit book "gave back its strong second-quarter performance". These are Third Point's explanations; the monthly reports do not let us test them position by position.

Historical performance versus MSCI World

Third Point reports that its Offshore Fund compounded at 13.2% a year net from December 1996 to June 2026, against 8.5% for MSCI World (TR) and 10.0% for the S&P 500 (TR). That comes from the Q2 2026 letter, p. 1. The Q3 letter updates the fund to 12.8% through September 2026 but no longer shows MSCI World. Neither letter says whether the index includes dividends net or gross of withholding tax.

These since-inception figures are an endpoint, not a path. Third Point's public documents that we could retrieve do not give the Offshore Fund's year-by-year returns from 1997 to 2023. The longest official yearly record we found belongs to Third Point Investors Limited, the London-listed company that invested substantially all its assets in the Offshore Fund from its listing in July 2007 until it bought a reinsurer in September 2025.

Its audited annual reports give the return of its US-dollar shares after the fund's fees and incentive allocation. We use the "total return from operations" per share, the investment result, not its share price, which traded at a discount of 21.3% to net asset value at the end of 2024.

From 2018 the reported total return also includes gains from buying back the company's own shares below net asset value, which are not fund performance: in 2024 it reported 25.48%, of which 24.11% came from operations, close to the headline Offshore Fund's 24.2%.

We compare matched calendar years from 2008 to 2024 with MSCI World net in US dollars and 3-month Treasury bills.

Including those buyback gains would raise the feeder's 2008-2024 CAGR to 8.53% and its 2015-2024 CAGR to 7.32%.

Growth of $100 from end-2007 to end-2024: Third Point's London feeder, return from operations in US dollars, reaches $354.07; MSCI World net USD reaches $323.34. Calendar-year returns, feeder and index: 2008 -38.24% and -40.71%; 2009 +40.36% and +29.99%; 2010 +36.32% and +11.76%; 2011 +0.34% and -5.54%; 2018 -12.00% and -8.71%; 2022 -25.27% and -18.14%; 2023 +2.17% and +23.79%; 2024 +24.11% and +18.67%.

End-2007 to end-2024: $100 becomes $354.07 at the feeder's return from operations and $323.34 in MSCI World; the feeder was ahead in every year from 2008 to 2014 and in only three of the ten years from 2015. Sources: Third Point Investors Limited annual reports for 2008, 2010, 2012 and later years listed in the methodology; MSCI World net USD; annual returns (CSV).

Calendar years, US dollarsThird Point feederMSCI World net
CAGR 2008-20247.72%7.15%
Sharpe / Sortino 2008-2024 (annual)0.41 / 0.720.41 / 0.67
Excess-return risk 2008-202421.5%18.9%
CAGR 2008-20168.30%3.26%
CAGR 2015-20245.96%9.95%
Sharpe / Sortino 2015-20240.32 / 0.560.60 / 1.27
Worst year-38.24% (2008)-40.71% (2008)
Losing years4 of 175 of 17

CAGR is the compound annual growth rate of $100 left invested. Sharpe divides the average return above Treasury bills by the volatility of that excess; Sortino counts only shortfalls below the bill return. Both are computed from 17 annual observations, so they say little about losses within a year. Treasury bills compounded at 1.16% a year over the period.

The record splits in two. From 2008 to 2014 the feeder beat the index every year, including a 36.32% gain in 2010 against 11.76%. From 2015 to 2024 it compounded 3.99 points a year slower, with higher measured risk. Its 21.6-point shortfall in 2023, 2.17% against 23.79%, was about twice its leads in 2020, 2021 and 2024 added together, 11.0 points.

Risk-return map, calendar years 2008 to 2024 in US dollars. Horizontal axis: annual excess-return risk; vertical axis: mean annual return. Third Point feeder: 21.5% risk, 9.9% mean return, CAGR 7.72%, Sharpe 0.41, Sortino 0.72. MSCI World net: 18.9% risk, 9.0% mean return, CAGR 7.15%, Sharpe 0.41, Sortino 0.67. The gold curve of unlevered MSCI World and T-bill mixes ends at MSCI World; the feeder lies on the extension of the line through it.

2008-2024: the feeder took 21.5% annual excess-return risk for a 9.9% mean return, MSCI World 18.9% for 9.0%; Sharpe 0.41 for both and Sortino 0.72 against 0.67. Sources: as above, plus FRED TB3MS; curve data (CSV). The curve shows only cost-free mixes of MSCI World and bills, without leverage; it is a restricted benchmark, not the opportunity set of a fund that can short, borrow and buy credit.

Over the full period the feeder sits on the line through MSCI World, so it earned the same return per unit of measured risk; it took more risk for more return. No unlevered index-and-bills mix reached its level of risk, so the curve cannot show a same-risk alternative. Over 2015-2024 the feeder fell well short, with a Sharpe ratio of 0.32 against 0.60.

Since the feeder stopped tracking the fund

Since September 2025 the London company has held a reinsurer alongside its fund stake, so its net asset value no longer measures Third Point alone. For 2025 and 2026 we therefore use Third Point's headline Offshore Fund returns, which assume the highest fee class. They are for the original series, which includes legacy private investments and is closed to new investors; Atom Computing, a private holding, was the top winner of the third quarter.

PeriodOffshore Fund (headline)MSCI World net USD
2025+8.9%+21.09%
Q3 2026-6.8%+1.88%
January to September 2026-0.1%+11.75%
January to October 2026-2.5%+10.89%

Sources: December 2025 and October 2026 performance updates, p. 1; Q3 2026 letter, p. 1; MSCI month-end levels. The New Series open to new investors, which excludes legacy private investments from June 2023, returned -8.6% in Q3 2026 and -2.0% from January to September (letter, p. 16). These headline figures are Third Point's estimates until audited. The October update also shows the cost of fees in a weak year: the master fund's positions lost 0.7% before fees from January to October, and the Offshore Fund 2.5% after them.

Copio's reading

Is it optimal for geometric growth?

No growth advantage is visible in the last decade, so we cannot call Third Point's current structure growth-optimal; its earlier record shows that skill can outweigh high fees. The objective here is expected growth of log wealth over decades, after all fees, for money that can stay invested through quarterly or annual redemption terms. The feasible comparison for an investor in the fund is a cheap world equity fund, possibly combined with cash, under the same liquidity needs.

We cannot estimate Third Point's future expected growth from 17 annual returns. The question is whether the evidence supports paying for its concentration, shorts and credit book over that simpler alternative. Three tests point the same way for recent years and the other way for the early record.

Are concentration and joint losses compensated? At 30 September five large AI-linked companies made up 29.5% of the 13F value, and the fund's top ten equity longs 35% of net asset value. Loeb writes that names which drove the 7.7% second-quarter return, including SK Hynix, Flex and MasTec, were caught up in the third-quarter rout. Over January to October 2026 the fund returned -2.5% against +10.89% for MSCI World.

Do the shorts and credit book improve whole-portfolio growth after their costs? They did not lower its co-movement with equities in the annual record (below), and the fund lost more than the index in 2018 and 2022. The letter's own description of shorts that "were amplifying our AI longs" shows that offsetting positions can add to the same risk. The early outperformance coincides with the same toolkit, so the record does not say which part created or destroyed value.

What do fees and liquidity terms change? A 2% management fee plus 20% of gains is a large hurdle against a cost-free index.

In a losing year no incentive fee is due, so the gap between a 0.7% gross loss and a 2.5% net loss in 2026 to October is mainly the management fee and expenses. The comparison above already uses after-fee returns for the fund and a cost-free index, which flatters the index slightly. Redemption limits matter for a household-scale investor, who cannot assume access to the money in a crisis.

Copio's verdict: the observed record from 2015 favours the simpler world index on compounded growth, risk-adjusted return and worst year. The decisive missing evidence is the Offshore Fund's full yearly record since 1997 and an attribution of returns between equities, shorts and credit after fees. A strong record from 2008 to 2014 is a historical outcome, not evidence that the current portfolio will repeat it.

Is it crash aware and protected?

Third Point actively cuts risk, but its record does not show protection: in the years equities fell, it usually fell about as much or more. Its net exposure fell from 93.2% to 70.6% of net asset value during October 2026, and its beta-adjusted equity exposure from 65.2% to 25.2%. That is fast de-risking. It came after the third-quarter loss, and the October update still shows a -2.3% month.

The annual record shows five years in which MSCI World net fell. The feeder fell less in 2008 (-38.24% against -40.71%) and gained in 2011 (+0.34% against -5.54%). It fell more in 2015, 2018 (-12.00% against -8.71%) and 2022 (-25.27% against -18.14%). Year-end values hide falls within the year, so the true worst losses were probably deeper for both.

The protection the filings show is small. Portfolio hedges had zero exposure on 31 October, and the 13F's S&P 500 put options covered underlying shares worth about 1% of the reported total, a minor part of a fund that was 93.2% net long a month earlier. The 15% of assets valued as Level III may be hard to sell quickly, and the board's 20% quarterly redemption cap protects remaining investors at the cost of liquidity for leavers.

Copio's verdict: not crash-protected in the sense of limiting portfolio losses; crash-aware in its exposure management, with no measured payoff from that awareness in 2018 or 2022.

Is it properly diversified without sacrificing growth?

No: on the annual record its returns moved almost in step with world equities, despite a large credit book and single-stock shorts. The number of positions and asset types does not answer this question; joint losses do.

Correlation of calendar-year US-dollar returns, 2008 to 2024. Third Point feeder with MSCI World net USD 0.89, with gold 0.34, with 3-month Treasury bills -0.20. MSCI World with gold 0.23 and with bills 0.05. Gold with bills 0.30.

2008-2024: the feeder's annual returns correlate 0.89 with MSCI World net USD, 0.34 with gold and -0.20 with Treasury bills. Sources: Third Point Investors Limited, MSCI, World Gold Council gold prices and FRED; correlation matrix (CSV). These are observed correlations over 17 years, not an estimate for a future crash.

A correlation of 0.89 means the feeder's good and bad years lined up closely with the index's. A regression of its annual excess returns on the index's gives a beta of 1.02: a 10% index year came with roughly a 10% feeder year, on average. With only 17 points, a single year such as 2010 can move these numbers noticeably.

Credit and short-selling could still have changed the size of the gains and losses, and in 2008-2014 the combination beat the index. But in this record they did not produce returns that held up when equities fell. For an investor who already owns world equities, the fund added more of the same risk, with stock-picking on top.

Copio does not grade Third Point in its portfolio checker, because an ETF replica cannot reproduce its single stocks, shorts, credit and private positions.

What a household can take from it

Four checks transfer to a personal portfolio.

  1. Check which return you are comparing. Third Point's record moves by one to five points in some years depending on share class (original or New Series), fee class and buyback gains in its listed feeder; compare the bad years over identical dates.
  2. Read a 13F as a dated, partial snapshot. It shows US long positions 47 days before it is filed, without shorts or credit, and the fund's own report a month later showed different top longs.
  3. Add up one theme across all holdings. Five AI-linked companies made up 29.5% of Third Point's 13F; a world index fund, individual shares and a tech fund can build a similar cluster in a household portfolio.
  4. Test a diversifier on the bad years. Compare its returns with your main equity holding in 2018 and 2022, over identical dates and after fees, before treating it as protection.

Our analysis of Situational Awareness' Q3 2026 13F covers the fund whose unwind Loeb cites as one cause of the summer sell-off. The study of Carl Icahn's funds shows another activist record split between strong early years and a weak decade, and Bill Ackman's Pershing Square a concentrated book held through a listed vehicle. Why investments fall together explains why correlations matter most in down markets, and Sharpe ratio versus CAGR why compounded growth is the measure that counts.


Sources and methodology

Checked on 18 November 2026. Holdings are dated 30 September 2026 unless stated; exposures 30 September and 31 October 2026; the most recent fund return used is the year to 31 October 2026.

Third Point LLC filings on EDGAR (CIK 1040273): the 13F-HR for 30 September 2026, filed 16 November 2026, accession 0001040273-26-000004 (information table, cover page), and the 13F-HR for 30 June 2026, filed 14 August 2026, accession 0001040273-26-000003.

Third Point documents published by Malibu Life Holdings: the investor letters for Q3 2026 (2 November 2026, pp. 1-12 and 16), Q2 2026 (p. 1) and Q4 2024 (p. 1); the master fund performance updates for 31 December 2025, 30 September and 31 October 2026; and the Offshore Fund's audited 2024 financial statements.

Feeder returns come from the financial highlights note of Third Point Investors Limited's annual reports: 2020, 2022 and 2024 from Malibu Life's site, and 2008-2019 and 2023 from Internet Archive copies of the company's former site, each report covering one or two years. The source file for each year is named in the annual returns CSV.

Benchmarks: MSCI World net total return in US dollars (index 990100) from MSCI's month-end levels; 3-month Treasury bills from FRED TB3MS, each monthly yield divided by 12 and compounded, an approximation of a bill return; gold from the World Gold Council's price series, last December price each year. Share prices for Nvidia, Meta and Microsoft are Yahoo Finance daily closes.

CAGR is the product of yearly growth factors raised to one over the number of years, minus one. Sharpe is the mean annual excess return over bills divided by its sample standard deviation; Sortino divides the same mean by the root mean square of negative excess returns, counting every year. Beta is the slope of the feeder's annual excess returns on the index's; 13F shares, group shares and price changes are our calculations.

Feeder returns are each year's "total return from operations" per share divided by the opening net asset value per share; for 2018-2024 this excludes the share buyback and redemption accretion included in the reported total return, and for 2008-2017 the two agree.

Limits: the feeder's return reflects its own fee terms and company costs, so it differs from the Offshore Fund's headline class. Offshore Fund returns to 2019 use the fees actually paid across share classes; from 2020 they assume the highest fee class, according to Third Point's notes. Annual data hide falls within each year, Level III valuations are estimates, and no figure includes investor taxes or inflation. The source ledger and reproduction script are in docs/research/blog-pipeline/2026-11-18/loeb/.

This article is educational, not personalised investment advice.

FAQ

What did Dan Loeb's Third Point buy in Q3 2026?

Third Point's 13F for 30 September 2026, filed on 16 November, lists new rows for Meta ($475.0m, now the largest reported holding), Unum ($438.7m), Nvidia ($342.6m) and Microsoft ($230.8m). Amazon ($417.1m in June) and Warner Bros. Discovery ($533.2m) no longer appear. Nvidia and Microsoft were among the fund's top five equity longs at the end of 2025, so a new 13F row is not necessarily a first purchase.

How did Third Point perform in 2026?

The flagship Offshore Fund returned -6.8% net in the third quarter and -0.1% from January to September, according to the investor letter of 2 November 2026; MSCI World net in US dollars gained 1.88% and 11.75% over the same periods. Those figures are for the original series, which is closed to new investors; the New Series open to them returned -8.6% and -2.0%. Third Point's 31 October update put the fund at -2.5% for the year to date.

Has Third Point beaten MSCI World over the long run?

By Third Point's own figures, yes: the Offshore Fund compounded at 13.2% a year net from December 1996 to June 2026, against 8.5% for MSCI World (TR). The yearly path we can verify is shorter. From 2008 to 2024 the London-listed feeder's investment return compounded at 7.72% against 7.15%, but from 2015 to 2024 at 5.96% against 9.95%.

Is Third Point a diversifier for an equity portfolio?

Not on the annual record we could verify. From 2008 to 2024 the feeder's calendar-year returns had a correlation of 0.89 with MSCI World net USD, and it lost more than the index in three of the five years the index fell, including 2022 (-25.27% against -18.14%). Seventeen annual returns make each estimate imprecise.

Can I copy Third Point's portfolio from its 13F?

Not faithfully. The 13F shows US-listed long equities and listed options held at 30 September, 47 days before filing. It omits the equity shorts (24.1% of net asset value then), the credit book (46.0% long), private positions and swaps. By 31 October Third Point had cut net exposure from 93.2% to 70.6% of net asset value.

Optional cookies help us measure visits and advertising. Choose each purpose, or decline both. You can change your choice using Cookie settings in the footer. Cookie policy.