a16z
Marc Andreessen, Ben Horowitz (general partners); registered investment adviser
Full-stack AI ownership plus crypto and American Dynamism (defense/space). Has deployed $10B+ across the AI value chain — the single most active major VC in AI. Edge is access to nearly every frontier-model and AI-infra round at scale.
- Not open — Not an open-ended vehicle — analysis only, not an available allocation.
- Self-reported — Headline returns are manager- or firm-reported and unaudited.
Key positions
Frontier-model stake; last private round ~$852B post-money (~$122B raise, Mar 2026)
Stake in the now-most-valuable private AI lab (~$965B Series H, May 2026)
Elon Musk frontier-model bet
AI-data platform; among a16z's most valuable infra positions
European open-weight frontier lab
Application-layer leaders (AI coding; voice)
Recent moves
Raised a record $15B+ across five funds in Jan 2026 (incl. a $6.75B growth fund), explicitly to double down on AI, crypto, and defense — the largest VC haul to date, ~18% of all 2025 US VC dollars.
Loading disclosed positions…
Our take
The closest thing to an index fund on the private AI frontier — it owns a piece of nearly every consequential model and infra company, which is both the appeal and the concentration risk if AI valuations compress. Scale is its moat and its constraint.
The marketing and the cash are two different portfolios. a16z's realized record is overwhelmingly pre-2015 vintages plus crypto — Fund III (2012) carried a net 11.3x TVPI after fees at 30 Sep 2025 (9.1x including parallel funds) on Coinbase, GitHub and Databricks; Coinbase alone produced ~$7B of gross distributions across a16z vehicles, and the first crypto fund returned ~5.4x net DPI. Nothing in the reported cash record yet attributes a dollar to the AI book being sold today. Second, run the arithmetic on their own numbers: a16z has returned at least $25B to LPs since 2009 (reported Sept–Oct 2025). It raised $15B in a single January 2026 vintage. For that one vintage to clear a 3x net it must return roughly twice the disclosed floor of lifetime distributions — from a starting position where the marquee names were already entered at three-comma valuations. Scale is the moat and the ceiling in the same sentence.
Thesis
a16z has stopped being a venture firm in the classic sense and become a diversified private-markets asset manager wearing venture's clothes: >$90B AUM (Jan 2026), an RIA registration that frees it from the 20% non-qualifying basket, six vertical fund families (apps, infrastructure, growth, bio+health, American Dynamism, crypto) and a media arm. The strategy is ownership-of-the-frontier at index scale. The open question is not whether it can access the deals — it plainly can — but whether index-scale access still produces venture-scale multiples on the $15B raised in January 2026.
Multi-strategy private capital across AI apps/infra, crypto, bio+health, American Dynamism (defense/space) and growth. Jan 9 2026: >$15B across five funds — Growth $6.75B, Apps $1.7B, Infrastructure $1.7B, American Dynamism $1.176B, Bio+Health $700M, plus ~$3B other venture; crypto Fund V added $2.2B on 5 May 2026 (crypto total $9.8B). RIA status permits crypto, secondary and public-market positions a standard VC partnership could not hold.
Assessment
- Structural access: at this size it can write into essentially every consequential frontier-model and AI-infra round — an advantage that compounds rather than decays.
- Fund III's 11.3x net TVPI (30 Sep 2025) is a real, fee-adjusted result, not a mark-up narrative — among the strongest large VC funds on record.
- RIA conversion was correct years before it was obvious: it enabled the crypto book (first fund ~5.4x net DPI) that non-RIA peers structurally could not hold.
- Vertical fund architecture sizes each sleeve to its own opportunity rather than forcing $15B through one generalist pipe.
- The AI book is entirely paper. OpenAI and Anthropic marks rest on round prices set by a small number of participants — not a liquid clearing price, and a16z is itself one of those participants.
- The crypto sleeve that carries much of the realized record is repricing: Fund V raised $2.2B in May 2026, well below Fund IV, and crypto-VC marks at a16z, Paradigm and Pantera have compressed since early 2026.
- Fee-and-scale drift: on $90B+ AUM, management fees alone are a large, certain annual business independent of carry — that changes the incentive to raise versus to return.
- Entry valuation, not access, is the binding constraint. Owning a piece of everything at 2025-26 AI prices produces index-like outcomes, and venture economics do not work on index outcomes.
- Vintage attribution is unflattering: pre-2015 funds and crypto carry the record. A 2016 WSJ review put Fund I top-5%, Fund II top-quartile, Fund III then outside top-quartile — early marks mislead both ways.
- 2026-07Anthropic's mark is no longer round-price-only: secondaries printed a ~$1.2T implied valuation on Caplight and via Rainmaker (9 Jul 2026), 24% above the $965B insider round — thin, with almost no sellers.
Record
Disclosed selectively — and the strongest numbers are not firm disclosure at all: they come from leaked LP materials reported by Newcomer and Not Boring. No fund-level net IRRs are public. Reported: at least $25B distributed to LPs since 2009 (reported Sept–Oct 2025); Fund III net TVPI 11.3x after fees, 9.1x with parallel funds (30 Sep 2025), with ~$7B net already distributed from that family; first crypto vehicle ~5.4x net DPI. Absent: DPI on the 2019-2023 vintages, precisely the window that would test whether the mega-fund era works. The headline AI stakes contribute TVPI, not DPI. Read the record as attributable to a 2010-2014 sourcing edge in a low-valuation environment plus one exceptional crypto call — both real skill — and treat the current AI book as unproven until a 2020s vintage prints cash.
- 2026-07The wait for a 2020s vintage to print cash now has a date: Anthropic filed confidentially 1 Jun 2026 for an October listing. Reported pricing talk of $400–500B sits far below the $1.2T secondary.
Risks & fit
- AI private-market repricing: a broad compression in frontier-model valuations marks down the concentrated core of the newest vintages simultaneously — the sleeves are not independent.
- Denominator/liquidity squeeze: LPs already over-allocated to illiquid AI paper slow re-ups, and the growth fund's exit path depends on an IPO window that has been narrow.
- Reflexive marks: a16z participates in the rounds that set the valuations it reports, so mark-to-market credibility is structurally weaker than in a liquid book.
- Power-law dependence at scale: a $6.75B growth fund needs several multi-tens-of-billions outcomes, not one, to clear a venture hurdle.
- Policy/regulatory exposure concentrated in crypto and defense — two sleeves whose outcomes turn on political regimes rather than product quality.
The read that scale has broken the returns fails if a 2019-2023 vintage prints DPI above ~1.5x within the next three to four years — that would show the mega-fund structure converting AI-era entry prices into cash, not just marks. Concretely: a realized OpenAI or Anthropic secondary sale, an Anthropic or Databricks listing, or an a16z-reported DPI schedule showing post-2018 funds tracking the pre-2015 curve. Conversely, the read is confirmed if 2028 reporting still attributes the bulk of distributions to Fund III, Coinbase and crypto vehicles while the AI funds remain carried at unrealized marks.
Relevant as a read on how the private AI market is being priced and who sets those prices — a16z is the largest single data point on frontier-AI entry valuations. Fund access is restricted to institutional and qualifying LPs and is not open to the general public, so this is analysis of a market-structure actor, not something a reader participates in. Useful as context for judging AI-exposed public equities and why private AI marks behave as they do.
Fund-level management fee and carry terms are not publicly disclosed and we do not assert them. Structurally, at >$90B AUM the management-fee stream is large and certain regardless of carry — a standing incentive question for any mega-fund manager.