Roughly 40% of every new dollar Andreessen Horowitz deploys in 2026 goes to AI — funded from a ~$45B AUM base and spread deliberately across competing foundation labs, infrastructure, and AI-native apps. That's the short answer. The longer answer is more interesting.
a16z isn't making one AI bet. It's making forty, structured so that no single model winner has to be correct for the strategy to pay. Below is the actual shape of the portfolio — by stage, by sector, and by check size.
The a16z AI Fund Portfolio in 2026
The a16z AI fund portfolio in 2026 spans three layers: foundation models, AI infrastructure, and AI-native applications. Andreessen Horowitz manages roughly $45B across its apps, infrastructure, growth, bio, crypto, and American Dynamism vehicles, and now routes about 40% of new deployment into AI. Its largest AI checks — OpenAI, Mistral, xAI — run into the hundreds of millions, while early application bets start at $1M-5M.
The defining feature isn't size. It's spread. a16z holds positions in multiple directly competing labs at once, a hedge against the single hardest question in the market: which model company actually wins enterprise share. You can track how these private valuations move on the AI Valuations dashboard.
a16z AI Portfolio by Sector and Stage
| Company | Layer | Stage | Est. Valuation |
|---|---|---|---|
| OpenAI | Foundation model | Late / growth | ~$300B |
| xAI | Foundation model | Growth | ~$50B |
| Mistral | Foundation model | Growth | ~€6B |
| Databricks | AI infrastructure | Late | ~$62B |
| Cursor (Anysphere) | AI application | Growth | ~$9B |
| ElevenLabs | AI application | Growth | ~$3.3B |
| Character.AI | Application / model | Growth | ~$5B |
| Together AI | AI infrastructure | Early growth | ~$3.3B |
Valuations are last-round estimates as of mid-2026 and move frequently; figures are approximate.
How a16z's AI Fund Portfolio Is Structured
a16z doesn't run a single "AI fund." AI checks come out of whichever vehicle fits the stage — and that structure is the strategy. The firm raised a reported $7.2B across funds in early 2024 and has continued raising AI-weighted growth capital since.
Apps Fund
Early AI-native application bets — $1M-15M checks into companies like Cursor and ElevenLabs at seed and Series A.
Infrastructure Fund
AI tooling, data, and compute layers — Databricks, Together AI, and the picks-and-shovels of the model economy.
Growth Fund
The $50M-350M+ late-stage checks into OpenAI, xAI, and Mistral that anchor the portfolio.
American Dynamism
Defense and national-security AI — dual-use autonomy, sensing, and infrastructure aligned with government demand.
Why a16z Spreads Its AI Bets
Most funds with $45B in AUM would concentrate. a16z does the opposite at the model layer, and the logic is straightforward: the cost of missing the eventual winner dwarfs the cost of backing a loser. A $300M check into a lab that goes to zero is survivable. Missing a 50x compounder is not.
What the spread buys
- ✓ Exposure to whichever lab wins enterprise share
- ✓ Pro-rata rights to double down on breakouts
- ✓ Signal flow across competing roadmaps
- ✓ Distribution leverage with portfolio app companies
What it risks
- ✕ Capital tied up at peak entry valuations
- ✕ Markups that may not survive a correction
- ✕ Conflicts across competing portfolio labs
- ✕ Returns diluted if no single bet 50x's
What a16z's AI Portfolio Tells You About the Market
When the largest venture firm in the world routes 40% of new capital to one theme, it's a signal about conviction and about return math. a16z needs fund-returning outcomes at $45B AUM, and only AI currently produces the $50B-300B private valuations capable of moving a fund that size. Track those vintage returns on the VC Performance dashboard.
The risk is concentration disguised as diversification. Eight AI positions across three layers still all depend on one macro thesis: that enterprise AI spend keeps compounding. If model margins compress and applications commoditize, spreading bets across labs won't help — they'd all derate together.
a16z's AI portfolio isn't a stock pick. It's a structural bet on a category.
Spread the entry, win the layer, follow on into the breakout — and let one of forty bets carry the fund.
Track AI valuations and fund performance on the AI Valuations and VC Performance dashboards at Value Add VC. Originally published in the Trace Cohen newsletter.