Silicon Valley split on open AI models as China gains ground

The emerging split over open‑weight models is the clearest ideological fault line in frontier AI since the early OpenAI board fights. On one side sit Nvidia, Microsoft, major VCs and a roster of startups arguing that open models are essential for innovation, security and national competitiveness. On the other are Anthropic and parts of OpenAI, warning that powerful open‑weights are harder to monitor and would hand a ready‑made accelerator to Chinese labs and other authoritarian projects.

This is not an abstract philosophical disagreement; it is about who controls the levers as we approach systems with genuinely dangerous capabilities. If the pro‑open camp wins in Washington and Brussels, we are more likely to see a fragmented ecosystem of strong, modifiable models embedded everywhere, with safety depending heavily on downstream integrators. If the cautious camp wins, regulation could converge on licensing regimes for a relatively small number of closed‑weight providers.

Either outcome has deep implications for the race to AGI. A wide proliferation of strong open‑weights might accelerate timelines by enabling more experimentation outside the big labs, but at the cost of more diffuse risk. A tightly regulated, closed‑weight world might slow the raw pace of capability diffusion but concentrate power and raise the stakes if alignment fails inside a small cartel.