In this article
← All articles

David Tepper's Appaloosa in Q3 2026: more chips, not fewer, and a filing that leaves out the cash


Executive summary

Appaloosa's 13F for 30 September 2026 shows David Tepper adding to chips, not leaving them: chips and chip equipment rose from 36.2% to 39.0% of the long shares reported, $7.56bn in September. The filing of 16 November closed AMD, Broadcom and Qualcomm but added Micron, Nvidia, Lam Research, ASML and a new Applied Materials stake. SpaceX grew from 0.5% to 7.4% of reported value.

The performance finding rests on press figures, because Appaloosa publishes no returns. On Institutional Investor's Rich List figures, the fund compounded at least 14.3% a year over 2022-2025, against 9.85% for MSCI World net in dollars. Most of that lead comes from 2022, when the fund reportedly gained 8.25% or 12.5% while the index fell 18.1%. Whether the index compounded faster over 2023-2025 depends on which 2024 figure is right, 8.25% or almost 26%, so no Sharpe or Sortino ratio is defensible.

Copio's assessment: the reported record does not show that Tepper's mix of cash and concentrated stocks cost long-run growth, but the disclosed stock book is a concentrated bet on one spending cycle. Held hypothetically at fixed share counts, its 24 positions fell 16.9% from 30 September to 17 November 2026, and the six chip positions fell 29.8%, while MSCI World fell 3.7%. That is a hypothetical, not the fund's result: the reported 40% cash, any hedges and the fund's non-US shares are outside the filing.

What is in the fund?

Appaloosa LP advises two private funds, and a 13F covers only part of what they own. Its Form ADV annual amendment of 25 March 2026 reports $20.40bn of regulatory assets: Azteca Partners LLC with gross assets of $15.71bn and Palomino Master Ltd with $4.68bn. The adviser and its related persons own about 100% of Azteca and 73% of Palomino, about 94% of the total on our weighting.

That ownership matters for what follows. Investing.com, citing Bloomberg on 6 July 2026, reported that about 90% of the capital belongs to Tepper and other insiders. The ADV figures support that order of magnitude. Appaloosa is close to a family office: few outside investors can ask for their money back in a fall.

The 30 September 2026 filing

The 13F-HR for 30 September 2026 lists 24 positions worth $7,559,628,593, all long shares, with no options. Micron is the largest at $1,331.4m (17.6%), then Amazon $1,114.9m (14.7%), TSMC $718.5m (9.5%), SpaceX $556.5m (7.4%), Uber $501.2m (6.6%) and Nvidia $456.8m (6.0%). The five largest make up 55.9% of reported value and the ten largest 80.4%.

Appaloosa 13F reported value by theme, 30 June and 30 September 2026, in millions of dollars, with the share of long shares in brackets. Chips and chip equipment $2,701.4m to $2,944.8m (36.2% to 39.0%); internet, software and cloud $3,228.1m to $2,910.1m (43.2% to 38.5%); SpaceX $38.4m to $556.5m (0.5% to 7.4%); iShares MSCI South Korea ETF $489.6m to $385.0m (6.6% to 5.1%); power, utilities and pipelines $666.9m to $516.4m (8.9% to 6.8%); industrials, travel and consumer $346.8m to $246.9m (4.6% to 3.3%); Apple and Berkshire puts, value of the shares covered, $254.1m to zero.

30 June to 30 September 2026, share of long shares: chips $2,701.4m to $2,944.8m (36.2% to 39.0%); internet, software and cloud 43.2% to 38.5%; SpaceX 0.5% to 7.4%; power and utilities 8.9% to 6.8%; puts covering $254.1m of shares to none. Source: 13F-HR information table, 30 September 2026 and 30 June 2026; every line from December 2025 to September 2026 (CSV). The themes are our grouping of the filing's lines.

The June filing values its puts at the price of the shares they cover, not at what the options cost, so we leave them out of the percentages; including them would put June's chip share at 35.0%.

The chip increase is not only a price effect. Valued at 30 September prices, the June chip share counts would have been worth $2,575.2m; the September holdings were worth $369.6m more, so the share-count changes added chip exposure. The long shares reported rose 1.2%, from $7,471.3m to $7,559.6m.

What changed since 30 June

CNBC's 16 November roundup listed the four new names correctly: Oracle ($172.2m, 1,253,932 shares in the filing), Applied Materials ($115.1m), Modine ($53.0m) and PG&E ($30.9m). The filing also shows larger stakes in Micron (975,000 to 1,250,000 shares), Nvidia (1,525,000 to 2,000,000), Lam Research (+63.4%), ASML (+30.0%) and SpaceX (225,000 to 3,883,329 shares).

The cuts were concentrated in internet names and travel: Alphabet -45.9% of shares, American Airlines -47.6%, Meta -22.2%, NRG -19.0% and Amazon -10.5%. AMD, Broadcom, Qualcomm, Whirlpool and Goodyear are absent, as are the June put options on 835,000 Apple shares and 25,000 Berkshire Hathaway shares. CNBC's 14 August article described the Apple line as a position; the June filing records it as a put, a right to sell that gains when the share falls.

What the filing cannot show

The SEC's 13F guidance excludes short positions and options the manager writes, and admits foreign shares only when they trade on a US exchange. So the filing shows no cash, no Treasury bills, no credit, no shorts and no shares that trade only abroad. That gap is material here: Bloomberg, as reported by Investing.com, named Samsung Electronics, SK Hynix and Kioxia among the first-half winners, and none of them appears in any of the four 13Fs from December 2025 to September 2026.

The ADV shows $20.40bn of regulatory assets, while the 13F for 31 December 2025 listed $6.93bn, about 34%. The two figures may be measured on slightly different dates. The gap is consistent with large holdings outside the filing, including the reported cash, but cannot measure them. One SpaceX detail is also unexplained: the 13F value implies $143.32 a share against a $150.86 close on 30 September.

Why does Tepper use this structure?

We found no 2026 letter or interview in which Tepper explains the September book, so his reasoning comes from earlier interviews and reported fund information. The structure itself is explained by ownership: Investing.com, citing Bloomberg, wrote that Tepper "returned most outside capital in 2019". Institutional Investor wrote in March 2022 that he "deploys an eclectic multistrategy-like approach".

On valuations and cash, Tepper's last fetched interview is from 18 September 2025. He told CNBC: "I don't love the multiples, but how do I not own it?" He added: "I'm constructive because of the easing right now, but I'm also miserable because of the levels," and "Nothing's cheap anymore." That is a stated reason to hold stocks and caution together, though not a statement about a cash target.

He also described trading rather than buy-and-hold on Nvidia: "I do own Nvidia, but I go back and forth a little bit ... trade a little bit." In September 2024 he asked of Nvidia's prospects: "Do you have enough power for the growth?" The power holdings in the 2026 filings, Vistra, NRG and now PG&E, may reflect the same view, but that is our guess; no source links them.

His view of position limits is explicitly flexible. Discussing Chinese shares, he said in September 2024: "I have limits, historic limits. I probably said a long time ago, I don't go above 10% or 15%. Well, that's probably not true anymore."

CNBC reported in August 2026, from a person familiar with the matter, that Appaloosa "bought a bigger position in memory stocks than what it sold in the second quarter" after June. The US filing does not show it: Appaloosa sold 690,000 Micron shares and all its Sandisk shares in the second quarter, then added only 275,000 Micron shares in the third.

Any larger memory purchase would have to sit outside the filing, for example in Korean or Japanese shares.

Historical performance versus MSCI World

On Institutional Investor's Rich List figures, Appaloosa compounded at least 14.3% a year over 2022-2025, against 9.85% for MSCI World net in dollars and 4.02% for three-month Treasury bills. These are press figures, not fund reports, and the basis is not stated except where marked. On those figures the fund beat the index in 2022 and 2025 and lagged it in 2023 and 2024; on Bloomberg's 2024 figure it beat the index that year too.

Press-reported Appaloosa returns against MSCI World net in US dollars and the average three-month Treasury bill yield. 2022: fund 8.25%, MSCI World minus 18.14%, bills 2.02%. 2023: fund more than 18% (shown at 18%), MSCI World 23.79%, bills 5.07%. 2024: fund 8.25%, MSCI World 18.67%, bills 4.97%. 2025: fund 23.3%, MSCI World 21.09%, bills 4.07%. January to June 2026: fund 32% gross, MSCI World 9.69%, bills 1.80%.

2022: fund 8.25% (12.5% in another report), MSCI World -18.14%; 2023: more than 18% against 23.79%; 2024: 8.25% against 18.67%; 2025: 23.3% against 21.09%; first half of 2026: 32% gross against 9.69%. Source: Institutional Investor Rich Lists of March 2023, March 2024, March 2025 and March 2026; MSCI 990100 net USD; FRED TB3MS; annual inputs (CSV).

The path matters more than the endpoint. Over 2023-2025 alone, the Rich List returns compound to 16.35% a year and MSCI World to 21.17%; with Bloomberg's almost 26% for 2024 the fund compounds at 22.39%. On the Rich List figures the four-year lead rests on one year, 2022, when the index lost 18.1% and the fund reportedly gained 8.25%. That year shows a loss avoided; the sources do not say how, so we do not attribute it to cash, hedges or stock selection.

Why we do not report Sharpe or Sortino ratios

A Sharpe ratio divides return above cash by volatility; a Sortino ratio counts only the volatility of returns below a target. Neither is defensible from this series, for four reasons. 2021 has no figure, only Institutional Investor's "mid-teens". 2023 is a range: "between 18 percent and 19 percent" in Institutional Investor of 4 January 2024 and "better than 18 percent" in its Rich List. The same January 2024 article gives "Appaloosa's 12.5 percent return in 2022", against 8.25% in the Rich List.

The outlets also disagree. For 2024, Institutional Investor reported 8.25%, while Investing.com, citing Bloomberg, reported "almost 26% in 2024" and "double-digit returns every year since at least 2021"; the sources do not settle which figures are right. With 12.5% for 2022 and 26% for 2024, the 2022-2025 rate would be 19.84% a year instead of 14.27%. And the first-half 2026 figure is gross of fees while the basis of the others is not stated.

Four annual observations with a range, a disputed year and mixed fee bases would produce ratios that look precise and are not. We therefore show the yearly series, the compound rates above with their bounds, and no risk-return or frontier chart. The decisive missing input is an audited annual or monthly net return series from the fund, which Appaloosa does not publish.

Copio's reading

Is it optimal for geometric growth?

The reported record does not show that Tepper's structure costs long-run growth, but optimality cannot be established, and the disclosed stock book takes concentrated risk. We judge expected log wealth, meaning growth that compounds through gains and losses, over a horizon of a decade or more. The constraint is light: with about 94% owned by the adviser and related persons, few outside redemptions can force sales. The comparison set is the same capital in a world index fund, or the same stock book held without the reported cash.

First, is the concentration paid for? Chips and the South Korea ETF make up 44.0% of reported value, and Bloomberg, as reported by Investing.com, named memory-chip makers as the first half's biggest winners. The same concentration lost 29.8% in the chip positions over the next seven weeks. Neither result supplies an expected return; a single winning half-year cannot show that concentration raises long-run compounding.

Second, does the cash improve whole-portfolio growth after its cost? The reported 40% cash would have earned three-month bill yields of 3.57% to 3.66% a year in the first half of 2026, about 1.8% for the half-year, against 9.69% for MSCI World.

Cash can add to long-run growth in two ways: by being redeployed at lower prices after a fall, or by cutting the volatility drag of a stock book that would otherwise be too large. The 2022 result fits either pattern, but one year without attribution cannot separate cash, hedges and selection, and neither route is shown here.

Third, do fees, financing and cash flows change the result? Fee terms, borrowing and Appaloosa's own returns net of costs are not public, and the press figures mix gross and unstated bases. Our verdict: the observed 2022-2025 path is consistent with a growth-friendly structure; the possible 2023-2025 shortfall against the index, which appears only on the Rich List's 2024 figure, is the weakening evidence. The decisive comparison is a net monthly return series against the same capital in an index fund over a full cycle.

Is it crash aware and protected?

The disclosed stock book is not protected, and the fund's protection, if any, sits outside the filing. The September filing lists no puts: the June puts on Apple and Berkshire Hathaway are gone, though a 13F need not show every option, and index options are not reportable at all. At fixed share counts, the 24 positions fell 16.9% from 30 September to 17 November 2026, with a low of -18.6% on 5 November; MSCI World net fell 3.7%.

Hypothetical buy-and-hold of the 24 positions disclosed for 30 September 2026. All positions: minus 6.6% at 16 October, minus 11.3% at 30 October, minus 18.6% at 5 November and minus 16.9% at 17 November. Six chip positions: minus 13.7%, minus 15.3%, minus 31.6% and minus 29.8%. MSCI World net: plus 0.1%, minus 0.8%, minus 2.8% and minus 3.7%.

Hypothetical, 30 September to 17 November 2026: all 24 positions -16.9%; six chip positions -29.8%; MSCI World net -3.7%. Source: 13F-HR for 30 September 2026, Yahoo Finance daily closes and MSCI World net USD; daily path (CSV), price inputs (CSV).

CoreWeave fell 42.7%, Lam Research 35.9%, Micron 33.1%, Nvidia 31.9% and Applied Materials 31.7%; only PG&E, Amazon, MPLX and Uber were higher. Recovering from the 17 November level needs a 20.3% rise for the whole book and 42.4% for the chip positions; our recovery maths article explains why.

CNBC's 16 November roundup put Nvidia's fall since 30 September at nearly 32%, "amid questions around the sustainability of the artificial intelligence trade" and a scandal that led to its chief executive's departure, so not all of the chip decline was a spending-cycle event. The 13F cannot show whether Appaloosa held these positions through the fall.

The fund may have fared better than the book. In a rough scenario where the reported 40% cash still applied and the other 60% moved like the 13F book, the fund would have lost roughly 10%, before any hedges; that is our arithmetic, not a reported figure.

The undisclosed holdings may include Korean and Japanese memory shares that fell like the chips, so the true figure could be worse. The near-absence of outside investors also lowers the risk of selling at the bottom to meet redemptions. Lower volatility from cash alone does not earn a positive verdict; what earns credit is the 2022 result, and that is one reported year.

Is it diversified without sacrificing growth?

The disclosed stocks divide into a chip cluster that moves together and platform names that move more independently; the fund's main diversifier is reported cash. A correlation of 1 means two daily returns move in lockstep and 0 means no linear relation. Over the 12 months to 30 September 2026, the ten largest positions excluding SpaceX, which listed only in June, had pairwise correlations from 0.06 to 0.77, averaging 0.33.

Correlation matrix of daily price returns, 1 October 2025 to 30 September 2026, for Micron, Amazon, TSMC, Uber, Nvidia, the iShares MSCI South Korea ETF, Meta, Alphabet, Vistra and Lam Research. Pairs range from 0.06 to 0.77, average 0.33. Micron with Lam Research 0.77 and with the Korea ETF 0.77; TSMC with Lam Research 0.75; Uber with Vistra 0.06; Micron with Meta 0.08.

1 October 2025-30 September 2026, 251 days: pairs 0.06 to 0.77, average 0.33; the four chip names average 0.62 with each other; Micron with Lam Research and with the Korea ETF 0.77; Micron with Meta 0.08. Source: Yahoo Finance daily closes; correlation matrices for both windows (CSV). These are holdings correlations, not a measure of the whole fund's diversification.

In the 34 trading days after 30 September, the average among the four chip names rose to 0.76, and the Korea ETF correlated 0.89 with Micron. Amazon, Uber, Meta and Alphabet stayed less tied to the chips, and Amazon and Uber ended the period higher. Two holdings can diversify one another and still fall together, as Alphabet (-18.5%) and Meta (-14.7%) did; the useful test is the combination's growth after costs. Our study of falling together shows how correlations rise in bad months.

On the reported numbers, the mix did not sacrifice growth over 2022-2025, because the 2022 result outweighed the lag in 2023 and 2024. Copio does not grade Appaloosa in its checker, because a book of single stocks, an unverified cash share and undisclosed credit, shorts and non-US shares cannot be represented by an ETF replica.

What a household can take from it

A 13F shows the risky part of a portfolio, not the whole of it, so copying the filing is the weakest use of it. It arrived 47 days after the date it describes; by its filing day the six chip positions were already 28.6% lower at fixed share counts. It also omits the cash, any memory shares held abroad and any hedges that shape the fund's actual risk.

  • Group holdings by what drives them. A chip maker, a chip-equipment supplier, a Korea ETF and a cloud company that rents out chips can depend on the same cycle; here CoreWeave, filed under internet, fell most, 42.7%.
  • Give cash a job. Separate money with a dated use from long-term capital, then compare the cash's yield with what the same money compounds to in a fall and a rise.
  • Test the joint loss. Apply a 30% fall to your largest theme together and check that you could keep contributing without selling.
  • Judge the path, not one year. Look at compounded growth across falling and rising years, after fees, before crediting a strategy.

Our study of Seth Klarman's cash allocation examines a different manager's case for holding cash, and the Situational Awareness filing shows a far more concentrated AI book.


Sources and methodology

Checked on 18 November 2026. Filings come from SEC EDGAR for Appaloosa LP, CIK 1656456: 13F-HRs for 30 September 2026 (accession 0001656456-26-000004, filed 16 November 2026), 30 June 2026 (0001656456-26-000003), 31 March 2026 (0001656456-26-000002) and 31 December 2025 (0001656456-26-000001). The Form ADV is the annual amendment of 25 March 2026, CRD 281909, Item 5.F and Schedule D Section 7.B.(1).

Quotations come from CNBC articles of 26 September 2024 (two), 18 September 2025, 14 August 2026 and 16 November 2026. Press returns come from Institutional Investor (Rich Lists of 10 March 2022, 7 March 2023, 27 March 2024, 21 March 2025 and 26 March 2026, and an article of 4 January 2024) and from Investing.com of 6 July 2026, syndicated on Yahoo Finance, which cites Bloomberg; the Bloomberg original returned HTTP 403 to our request.

MSCI World is the net total return index in US dollars (990100), year-end and daily levels from MSCI. The cash benchmark is the three-month Treasury bill, FRED series TB3MS, as the calendar-year average of monthly yields, and half the first-half average for 2026. Compound rates are (product of (1 + annual return))^(1 / years) - 1; for 2023 the fund's lower bound of 18% is used.

Theme totals sum the 13F value column by our classification; the total equals the cover page, and theme shares are of long shares, excluding the value of shares covered by puts. Every share line's value divided by shares equals the quarter-end close except SpaceX in September.

The hypothetical holds the 13F share counts fixed from 30 September, weights by reported value and uses Yahoo Finance closes without dividends, while MSCI World includes net dividends, a small mismatch over seven weeks; Modine's 1,086-for-1,000 share adjustment on 2 October 2026 is applied. Correlations are Pearson coefficients of daily price returns over 251 and 34 days.

The 13F describes positions at quarter-ends, not returns, cash or investor capital. The fund's net returns, fees, borrowing, shorts and non-US holdings are not public. Every calculated figure can be reproduced from the CSVs linked above with the formulas stated here. This article is educational, not personalised investment advice.

FAQ

What did David Tepper's Appaloosa buy and sell in the third quarter of 2026?

Its 13F for 30 September 2026, filed on 16 November, shows new positions in Oracle ($172.2m), Applied Materials ($115.1m), Modine ($53.0m) and PG&E ($30.9m). AMD, Broadcom, Qualcomm, Whirlpool, Goodyear and the put options on Apple and Berkshire Hathaway were no longer listed. The filing also shows more shares of Micron, Nvidia, Lam Research, ASML and SpaceX, and 45.9% fewer Alphabet shares.

Did Appaloosa sell out of chip stocks in Q3 2026?

No. It closed AMD, Broadcom and Qualcomm, but added 28.2% more Micron shares, 31.1% more Nvidia shares, 63.4% more Lam Research shares and a new Applied Materials stake. Chips and chip equipment rose from 36.2% of the long shares in the 13F at 30 June to 39.0% at 30 September 2026.

How much cash does Appaloosa hold?

No filing shows it, because a 13F lists only certain US-listed long positions. Investing.com, citing Bloomberg on 6 July 2026, reported an average cash position of about 40% "throughout the year". The Form ADV filed on 25 March 2026 reports $20.4bn of regulatory assets, against $6.9bn of 13F value at 31 December 2025.

What return has Appaloosa made?

Appaloosa publishes no returns, and press figures conflict. Institutional Investor's Rich Lists reported 8.25% for 2022, more than 18% for 2023, 8.25% for 2024 and 23.3% for 2025, but another Institutional Investor article gave 12.5% for 2022. Investing.com, citing Bloomberg, reported almost 26% for 2024 and 32% gross for the first half of 2026. On the Rich List figures, Appaloosa compounded at least 14.3% a year over 2022-2025, against 9.85% for MSCI World net in dollars; on the higher figures, 19.8%.

Can a household copy Appaloosa from its 13F?

Not usefully. The filing arrived 47 days after the date it describes, and by then the six chip positions had fallen 28.6% at fixed share counts. It also omits the reported cash, any shorts, credit and any shares that trade only outside the US, which can include memory makers such as Samsung Electronics.

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.