Altimeter Capital
Brad Gerstner (founder, chairman, CEO)
Ultra-concentrated tech crossover fund and one of the most vocal AI bulls; pairs a ~13-name public book with marquee private AI stakes (Anthropic, OpenAI, SpaceX (xAI)/SpaceX).
- Not open — Not an open-ended vehicle — analysis only, not an available allocation.
- Partial 13F — The disclosed book is longs-only and ~45 days lagged — a slice, not the strategy.
Top holdings
Largest position, ~8.1M shares (~$1.5B); approx weight of a ~13-holding book
Top-5; AI capex
Top-5; autonomy/AI platform thesis
Top-5; AI foundry
Top-5; OpenAI-aligned hyperscaler
Recent moves
Co-led Anthropic's $65B Series H at a ~$965B valuation (May 2026) alongside Dragoneer, Greenoaks and Sequoia; Gerstner is also a public OpenAI champion and an SpaceX (xAI)/SpaceX backer, and publicly argued Anthropic's run-rate could triple in 2026.
Loading disclosed positions…
Our take
Altimeter is the purest 'AI is the trade of the decade' expression on this list — extreme public concentration in NVDA plus direct private stakes in the leading model labs; the positioning is a non-zero-sum bet that the intelligence TAM lifts multiple winners at once.
The question is not whether Gerstner is right about AI but whether the book is as diversified as the framing implies.
Long Nvidia and TSM (picks and shovels) plus long the labs privately is not two independent bets — the labs' capex IS Nvidia's revenue line. A deceleration in training and inference spend marks down both legs the same day; calling that 'the TAM lifts multiple winners' converts a largely single-factor position into a story about breadth. Second, the Anthropic mark carries a mild self-reference: the valuation revaluing Altimeter's stake was set by a round Altimeter co-led. The syndicate blunts this — four leads (Altimeter, Dragoneer, Greenoaks, Sequoia) plus six co-leads including GIC and Capital Group — so it is a negotiated print among arm's-length parties, not a self-set mark. It is still not a clearing price, and stays untested until an IPO or secondary. Third, a reader inferring exposure from the 13F is studying the shop window, not the warehouse.
- 2026-07Even the picks-and-shovels leg is not one bet: NVDA and TSM diverged ~15.7pp in July (+0.3% vs -15.4%), and Meta raised 2026 capex guidance to $130–145B on 29 Jul while its shares fell ~1.2% for the month.
- 2026-06The Anthropic mark's test is now a dated process rather than an open contingency: a confidential draft Form S-1 went to the SEC on 1 Jun 2026.
Thesis
Altimeter's visible public book is now its least important part. The Q1 2026 13F shows 13 names worth $5.70B, with the top five — NVDA (28.6%), META (19.6%), UBER (10.1%), TSM (8.1%), MSFT (7.7%) — at ~74%. The decisive exposure is private: co-leading Anthropic's $65B Series H at a $965B post-money (May 2026). The critique that matters is not stock picking but that the public and private books are largely two expressions of one factor: sustained AI capex, framed as a diversified 'many winners' thesis.
Concentrated tech crossover: a dozen public names at conviction weights alongside late-stage private stakes in model labs and AI infrastructure. Q1 2026 narrowed from 18 holdings to 13 — added NVDA (+$216M), UBER (+$171M), CoreWeave (+$100M), TSM (+$50M), opened Arm (+$260M) and Axon; exited seven names including Alphabet, Coupang, Shopify and MercadoLibre. Public advocacy (the 2022 'Time to Get Fit' letter to Meta) is an active tool, not incidental.
Assessment
- Willing to hold real concentration and act on it — Q1 2026 exited seven of 18 names (Alphabet, Coupang, Shopify, MercadoLibre, Zillow), showing pruning, not just accretion.
- Crossover structure lets one view be expressed across the stack — private labs, silicon, foundry, neoclouds — rather than through a single entry point.
- The Axon open cuts against a pure-thematic read: it is not an AI-capex name, evidence the book is not wholly one trade.
- Public engagement (the 2022 Meta letter, which preceded Meta's efficiency pivot) shows willingness to underwrite a view in the open.
- Correlation dressed as diversification: the semis, neoclouds and private labs largely resolve on one variable — AI capex persistence — so the book has less breadth than its name count implies.
- Altimeter co-led the round that marks its own largest private stake. The ten-firm lead syndicate makes this weakly self-referential rather than improper, but it is still not an independent clearing price.
- The widely-cited ~29.5% annualised figure is a third-party BACKTEST of top-20 13F names rebalanced quarterly, not audited net-of-fee fund returns.
- NVDA at 28.6% of the disclosed book: in 2023-25 nearly every long-tech fund was carried by the same name, making skill hard to separate from a shared factor.
- Firm identity is one person. There is no publicly disclosed succession plan or co-CIO structure.
- 2026-07July tested the one-factor read and the legs split: MSFT +24.6% precisely for holding capex flat, NVDA +0.3%, against TSM -15.4% and SOX -20.6%. The Q1 top five, 74% of the book, netted roughly flat.
- 2026-07July's dispersion came from inside the AI complex, not from the small non-AI line: MSFT +24.6% against TSM -15.4% and CoreWeave -27.9%. Crediting the breadth to Axon understates where the offset came from.
Record
Actual net-of-fee returns are not disclosed — Altimeter files 13Fs but has no public RETURN-reporting obligation, so no verified ytd, 1y or since-inception figure exists. What IS observable: the disclosed 13F book fell from $6.66B (Q4 2025) to $5.70B (Q1 2026), a decline mixing marks with net selling and seven exits — it is not a return. Aggregators publish 13F-derived proxies (hedgefollow shows -12.29% for 26Q1) but these reconstruct longs-only US equities from a ~45-day-lagged snapshot: no shorts, no hedges, no private book, no intra-quarter trading, no fees. The fund reportedly had a severe 2022 and a strong 2023 rebound, but neither is disclosed with a number. The Grab/AGC SPAC (~$40B, then the largest ever) is the cleanest public test of turning a private mark into a public one, and it went badly for post-deal holders.
Risks & fit
- A single capex deceleration at hyperscalers or labs hits the private stakes, the semis and the neoclouds simultaneously — one factor, not five independent positions.
- Private marks are illiquid and stale; a down round or a soft IPO print at any lab resets a large share of value with no offsetting liquid hedge.
- Key-person concentration in Gerstner, with no disclosed succession.
- 13F opacity: shorts, options, non-US names and the private book are all invisible, so the disclosed 74%-in-five figure may understate true concentration.
- Public bullish commentary and owned marks point the same direction; treat the commentary as position-consistent, not independent.
The 'multiple independent winners' framing fails if one shock moves the whole book together: capex guidance decelerates and, in the same window, semis multiples compress while private lab marks reset or a lab IPO prints below its last round — showing the positions were one factor. It is validated if a lab realises DPI at or above the last private mark while the infrastructure leg holds, or if the two legs measurably diverge through a capex scare. The nearest test is Anthropic's mooted IPO: a realised price against the $965B May-2026 post-money would be the first non-syndicate mark on the largest driver.
Useful as a study in crossover-fund construction and in how private marks interact with a public book — how a single macro factor can hide inside a name-diverse portfolio, and how to discount a manager's public commentary when it is position-consistent. It is not accessible: Altimeter is a private fund, and its lab stakes are late-stage private rounds open only to institutional co-investors. Nothing here is a view on whether to hold any named security.
Fee terms are not publicly disclosed. Crossover funds of this shape typically charge a management plus performance fee on the liquid sleeve and carry on a locked private sleeve, but Altimeter's terms appear in no public filing we located.