Sam Altman (personal portfolio + Hydrazine Capital)
Sam Altman — OpenAI CEO, ex-Y Combinator president; one of the most active angel investors of the era (100+ personal bets since 2010, plus the Hydrazine Capital funds he co-founded with brother Jack Altman in 2012).
The single most consequential personal AI-wave portfolio that is NOT primarily about its namesake company — Altman holds NO direct OpenAI equity (May 2026 court filings; only a since-sold fractional indirect YC/Sequoia-fund stake) and draws a ~$76k salary, so his wealth is his side bets. The thesis is energy-for-compute: his ~one-third stake in fusion startup Helion (chairman) and his role at fission startup Oklo are direct plays on the power bottleneck behind AI. Layered with biotech (Retro Biosciences) and a deep angel book (Stripe, Reddit, Instacart).
- Personal book — An individual's own capital, not a vehicle outside investors can access.
Top holdings
$0 confirmed — no direct Altman equity per the 2026 for-profit cap table and the May 2026 Musk-litigation court filings; he takes a ~$76k salary and his only OpenAI exposure was a since-sold indirect YC/Sequoia-fund stake. Speculative: a first-time equity grant has been periodically floated during the restructuring (unconfirmed, none vested as of Jul 2026) — at OpenAI's ~$852B valuation even a low-single-digit % would exceed his entire book, but it is not priced into this portfolio.
~$1.65-1.7B — Largest position. Invested since 2015; committed $375M in 2021; owns ~1/3, chairman. Helion last valued >$5.4B. Direct power-for-AI bet.
Chairman post-SPAC (May 2024) until stepping down Apr 2025 to clear OpenAI-Oklo conflict. Only listed name in his book.
$180M — Altman's largest disclosed biotech bet; cellular-reprogramming anti-aging. Not an AI position but a flagship personal be
long — Early angel; payments infra, AI-adjacent.
long — Former chairman + large angel; AI-data-licensing relevance. Now public.
AI Pin maker; assets sold to HP 2025 — a marquee loss in the book.
Recent moves
Stepped down as Oklo chairman (Apr 2025) to remove the OpenAI-Oklo conflict; Humane's assets sold to HP (2025), writing down that position. Helion stake reaffirmed at >$5.4B valuation (2025 press). Court filings in the Musk-OpenAI litigation (May 2026) disclosed >$2B in OpenAI-tied holdings and confirmed no direct OpenAI equity. A first-time equity grant for Altman has been periodically floated during the for-profit restructuring but is unconfirmed and unvested as of Jul 2026 — not priced into the book.
Our take
Why it matters: Altman's personal book is the purest individual expression of the 'AI is bottlenecked on power' thesis — fusion (Helion) and fission (Oklo) are his two biggest energy bets, exactly the compute-power nexus this page tracks. Edge: unmatched signal into which infra the OpenAI ecosystem will need. Caveats: almost entirely private and disclosed-not-market-weighted (no 13F — he holds little US-listed equity); 'positions' are press/court estimates, not audited marks; the conflict-of-interest overhang (Oklo, Helion potentially supplying OpenAI compute) is real and is exactly why he keeps recusing.
The line everyone repeats — he owns no OpenAI equity — is the least useful fact here.
Zero direct equity alongside billions in assets whose demand curve runs through OpenAI is economically a synthetic long on OpenAI, with worse liquidity and no cap-table protection. Second, this book is knowable only because adversarial discovery compelled it: the exhibit covered nine companies that had done business with OpenAI, so it is a litigation-shaped sample selected for the positions raising conflict questions, marked as of Dec 31, 2025 and roughly seven months stale. Third, the independence question is narrower than the headlines but real: Helion's $15.5B June 2026 round was led by Thrive Capital, and its signed anchor offtake is Microsoft's May 2023 PPA, not OpenAI. Altman testified OpenAI has never bought or received power from Helion; what exists is a 2024 option and exploratory talks. The mark is not circular today — but any future OpenAI offtake would be a customer chosen by the company its largest shareholder runs.
Thesis
A personal book organized around the idea that AI is bottlenecked on electricity, expressed through two nuclear bets — roughly one-third of fusion startup Helion, marked at ~$1.65B as of Dec 31, 2025 per Altman's May 2026 trial testimony, and an undisclosed but still-significant stake in fission SMR developer Oklo (OKLO) — layered on a YC-era angel book (Stripe, Reddit, Instacart) and biotech (Retro Biosciences). Court exhibits in the Musk–OpenAI litigation covered nine companies that had done business with OpenAI, and confirmed he holds no direct OpenAI equity.
Concentrated, permanent-capital angel investing with no fund structure, no LPs on the personal book, and no exit discipline — positions are held indefinitely and sized by conviction rather than portfolio construction. The energy positions were governance-adjacent (chairman/board seats, since vacated) rather than passive. Hydrazine Capital, co-founded 2012, is the separate LP-funded vehicle; the personal book is the larger and more consequential of the two.
- 2026-06Helion's 4 Jun 2026 Series G ($465M led by Thrive, $15.5B post) re-marks the roughly one-third stake at ~$5.2B undiluted, or $4.1B per Finance Monthly (9 Jun) — some 2.5-3x the ~$1.65B Dec-2025 figure.
Assessment
- Identified the power constraint early — first Helion check ~2015, well before the 2023-26 datacenter energy scramble (though most dollars came via the Nov 2021 Series E he led).
- Structural information edge: sits at the demand node of the AI buildout, so he sees which infrastructure is needed before the market prices it.
- Permanent capital with no redemption clock suits fusion's 15+ year physics timeline — a fund structure could not have held Helion this long.
- Recusals were executed, not just promised — Oklo chair Apr 2025, Helion board Mar 2026 — a real cost paid for governance hygiene.
- Concentration is extreme — Helion is the largest single position by a wide margin, though its exact share of the disclosed total is not derivable from the exhibit.
- The ~$1.65B Helion figure is testimony against a Dec 31, 2025 private mark, not audited or exit-tested. Private marks in hot sectors reprice last.
- The two energy bets are not independent — both monetize the same AI-power demand curve, so they correlate to one, not zero.
- Helion has not delivered grid power. The thesis rests on a 2028 target that is a contractual Microsoft PPA deadline no fusion company has ever met on schedule.
- There is no disclosure cadence for the personal book, so every figure here can go stale silently — these marks are already ~7 months old.
- 2026-07Schedule slippage is no longer hypothetical or fusion-only: Oklo missed the DOE's 4 Jul 2026 criticality deadline for Aurora-INL while Antares, Valar Atomics, Deployable Energy and Aalo Atomics reached it.
Record
No audited track record exists and no fund-level return series is public for either the personal book or Hydrazine — this is a private portfolio, not a reporting vehicle, so no IRR, DPI or TVPI can be attributed. What can be said: the record is dominated by one position. Helion, valued at ~$1.65B in May 2026 testimony against a Dec 31, 2025 mark (~one-third of a company later valued at $15.5B in June 2026), is the largest disclosed holding by a wide margin — a power-law outcome carried by one 2010s-vintage relationship. The only liquid, marked position is Oklo, which closed $44.13 on July 21, 2026, down roughly 31% year-to-date — a reminder that when these bets get a public price, they get a volatile one. Humane went the other way, its assets sold to HP in 2025. Forbes put net worth above $4B in May 2026 after the trial disclosures. The return traces to early access and long holding, not to an observable repeatable process.
- 2026-07Oklo closed $38.83 on 31 Jul 2026: -45.89% YTD and -49.30% over one year (Yahoo), some 15pp worse in ten sessions. The -31% quoted at the $44.13 close read light — other providers had -38.5% that day.
- 2026-06Net-worth marks now disagree in both directions: Forbes Real-Time ~$3.4B (Jun 2026), the Forbes list $6.5B (May 2026), Finance Monthly $6.1B (9 Jun). 'Above $4B' sits inside a $3.4B-$6.5B spread.
Risks & fit
- Key-person and attention risk — the same person is CEO of OpenAI, and attention is the scarce input across both the company and the book.
- Litigation and congressional scrutiny of the OpenAI-adjacent holdings is unresolved; outcomes could force divestiture or restructure counterparty relationships.
- Fusion technical risk is close to binary. If Helion's approach does not reach net grid power, the position is not marked down — it approaches zero.
- Oklo's public price is sentiment-driven and pre-revenue; SMR licensing timelines slip, and the stock is down roughly 31% year-to-date.
- Private marks reset only at the next round. A funding-market cooling would surface losses that carrying value currently hides.
The independence concern fails if Helion delivers grid power near its contractual 2028 Microsoft deadline and no OpenAI offtake ever materializes — the mark would then rest on physics and an arm's-length customer, and the concentration would read as conviction that paid. Conversely the energy thesis fails even if AI power demand is real: if hyperscalers procure at scale from incumbent gas, grid, and utility nuclear on 2027-2030 timelines, both bets are right about the problem and wrong about the supplier.
Readers studying how the AI buildout's power constraint is being financed at the earliest stage, and how personal capital, corporate strategy, and governance intersect at the top of the AI industry. Also a case study in disclosure asymmetry — what litigation reveals versus what voluntary reporting would. It is not a trackable or replicable book: the holdings are private, unfiled, marked only at irregular intervals, and closed.
No fee structure applies — the personal portfolio is Altman's own capital, no outside investors, no management or performance fee. Hydrazine Capital is a separate LP vehicle whose terms and fund sizes are undisclosed. Neither is open to outside capital.