Is VC cooked? Axios calls DeepSeek an 'extinction level event' for venture capital
Jan 27, 2025
Key Points
- DeepSeek's open-source model threatens venture capital's bet that proprietary foundation models would dominate before cheaper alternatives emerged, with Axios arguing the shift could be an 'extinction level event' for firms that deployed unprecedented capital on that thesis.
- Y Combinator CEO Garry Tan counters that cheaper models accelerate startup formation rather than kill venture, pointing to vertical SaaS companies reaching $10 million ARR with under 10 people as proof the market expands downmarket.
- Investors are concerned but not panicking, though deal closures may slow as the venture case tilts from foundation model winners toward infrastructure and applications in a more competitive, lower-margin landscape.
Summary
DeepSeek's Open-Source Model Threatens VC's Foundation Model Thesis
Dan Primack's Axios piece argues that DeepSeek could be "an extinction level event for venture capital firms," particularly those betting heavily on proprietary foundation models. The claim, while incendiary, rests on a narrow but real tension: venture firms have deployed unprecedented capital on the assumption that foundation model companies would achieve distribution and defensibility before open-source competitors caught up. DeepSeek, by releasing a capable model as free, open-source software developed by a Chinese hedge fund, may have broken that timeline.
The thesis under pressure
The Davos consensus entering this week was that the US held a structural lead in AI—chip access, compute scale, regulatory clarity—with the primary risk being whether enough infrastructure could be built in time. DeepSeek shifts that framing. If the model genuinely achieved competitive performance on smaller budgets and fewer chips than OpenAI, and if it remains open-source, the venture case for betting on a closed foundation model company weakens significantly. Primack frames the problem as one of misdirected capital: the "quantums of capital are just so much more than anything VC has ever before dispersed based on what might be suddenly a stale thesis."
That math is real. A16Z's Stargate consortium is committing $500 billion in CapEx toward data center buildout. If the scaling laws that justified those commitments no longer hold—if you can train competitive models more cheaply—those bets look speculative rather than inevitable.
The pushback from Tan
Y Combinator CEO Garry Tan counters that the environment is not collapsing but expanding. He points to vertical SaaS companies reaching $10 million annual recurring revenue with fewer than 10 people, suggesting that cheaper, open-source models accelerate the long tail of startup formation rather than kill venture. His framing is optimistic: "a thousand flowers will bloom." More startups, more distribution plays, more opportunities for venture to own customer relationships even if foundation models become commoditized.
Investors interviewed for the Axios piece are "not panicking, but clearly concerned," particularly that they were caught off-guard. Primack notes: "Don't be surprised if some deals in process get paused."
What remains uncertain
The segment does not resolve whether DeepSeek's cost claims are real or whether the model truly rivals OpenAI's capabilities across the board. If export controls or geopolitical friction prevent US companies from using the model, or if it turns out to require similar capital despite appearances, the VC thesis survives. If, conversely, the model is genuinely cheaper and capable, the venture case tilts toward infrastructure, applications, and services—a more competitive, lower-margin landscape than the foundation model winner-take-all narrative that drew capital in the first place.