Sequoia, Lightspeed, and Dragoneer hold stakes in four or more competing foundation model labs simultaneously
Apr 15, 2025
Key Points
- Sequoia, Lightspeed, and Dragoneer now hold stakes in four or more competing foundation model labs each, abandoning venture capital's traditional single-winner concentration strategy.
- The portfolio approach mirrors how venture firms treat commoditized platforms like short-form video, suggesting foundation model capability itself may be becoming undifferentiated.
- Whether this reflects genuine model commoditization or simply venture's inability to pick winners early remains unresolved; backed companies may diverge into distinct products rather than stay generic inference engines.
Summary
Top-tier venture firms are now backing competing foundation model labs simultaneously, a striking reversal of traditional venture practice. Sequoia holds stakes in OpenAI, Thinking Machines, xAI, and SSI. Lightspeed is in four model companies. Dragoneer has invested in five: OpenAI, Thinking Machines, xAI, Mistral, and SSI.
This pattern suggests the model layer itself is commoditizing. When the core product becomes undifferentiated—much like short-form video feeds across TikTok, Instagram, and YouTube—venture firms diversify across competitors rather than picking winners. The reasoning mirrors Meta's FTC defense: if every platform ships the same core features (algorithmic feeds, vertical video, comments, shares), no single player owns the category.
That logic may not hold for foundation models. The companies backed by these same firms may diverge sharply in what they become. OpenAI is exploring a social network. Thinking Machines might become a B2B SaaS play. xAI and Mistral pursue different training and commercialization strategies. If model labs eventually specialize into distinct products rather than remain generic commodity inference engines, then portfolio diversification looks prudent rather than chaotic.
Venture's oldest rule—concentration of capital in a single winning bet—has bent. Whether that reflects genuine commoditization of model capability or merely the inability to call a winner this early remains unclear.