Venture Capital Firms in AI

A venture fund answers to limited partners and is judged on financial returns over a roughly ten-year fund life; a corporate arm answers to a parent's strategy. When Samsung led Mistral AI's US$3.5B Series D in September 2026, it announced a chipmaking partnership the same day — a two-sided transaction a purely financial fund cannot structure.

Pooled LP capital committed for a decade lets a partnership underwrite research risk without redemption pressure. Sequoia holds an early position in Safe Superintelligence — a lab that explicitly declines to ship a product — because a ten-year fund can wait out years with no revenue, while a hedge fund marking positions quarterly generally cannot.

That one firm now writes both ends of the size distribution. Sequoia committed over US$2B into Anthropic's US$65B H round at a US$965B valuation while making agent infrastructure its densest 2026 category — ten deals, roughly a third of its AI transactions — at far smaller tickets. Multi-stage reach is now the norm, not the exception.

Andreessen Horowitz runs the largest dedicated AI capital pool in venture and led Cognition's US$2B Series E at a US$48B valuation in September 2026; Index Ventures backed Mistral AI, Cohere and Perplexity from early cheques; Lightspeed led Harvey's US$550M round at a ~US$15.6B valuation that same month.

Most run both sleeves with different ownership math: a16z holds OpenAI, xAI and Mistral at the model layer while also backing applications like Cursor and Replit, whereas a specialist such as Radical Ventures concentrates on labs and infrastructure — Cohere, Crusoe — where model progress itself is the product. Position sizing differs accordingly.

No. A fund announcement concerns capital available under a mandate; actual investments occur over time and may include non-AI companies, reserves and fees. Company financing totals should count documented company transactions, not the entire size of a VC fund.