Silicon Valley closes ranks around "open weights": over 130 signatures and one loud exception

On 24 July a document titled "Open Weights and American AI Leadership" appeared on NVIDIA's and Microsoft's official channels. The letter describes "open weight" models as tools anyone can download and modify, and frames them as pillars of the American technology ecosystem. The signatories ask policymakers not to smother competition with pre-emptive restrictions, proposing instead that access to compute be widened for startups and researchers, that shared training assets — datasets, tools, evaluation frameworks — be funded, and that the legitimate practice of distillation be told apart from the outright theft of intellectual property. They do acknowledge the risk: once weights are released they escape the developer's control, and modified versions are hard to trace. But the answer, they argue, is not to ban open weights. The list of asks also includes strengthening the application layers so that AI can be used sovereignly.

The shape of the signatory list tells a story of fast aggregation. It started with 25 entities on 24 July; by 30 July the coalition had passed 130 verifiable names, including hardware giants such as AMD and Intel, hyperscalers such as Amazon and Google, and frontier labs such as OpenAI, usually associated with closed models. The initiative, promoted on social media by Jensen Huang among others, appears to have united the industry around a shared interest: stopping fear of Chinese models from turning into rules that hold back domestic innovation. The count keeps moving, though: 25 signatures on 24 July, 50 the next day, more than 130 when checked on 30 July. The number you are reading today may already be out of date.

Not everyone chose to put their name under that text. Anthropic, through CEO Dario Amodei, published a separate note clarifying that it has never called for a ban on open models and prefers an approach built on finer-grained controls. Anthropic's proposal covers restrictions on chip exports to China, targeted action against industrial-scale distillation and, above all, mandatory safety testing for any sufficiently capable model, whether open or closed.

In Washington the debate over the risks of open AI continues without a specific bill to anchor it, which leaves room for speculation about possible bans. The companies' mobilisation is meant to draw a red line against what they see as early intervention that would damage American competitiveness. Whether such a varied set of commercial interests can actually shape legislation remains unclear, but the political signal is hard to miss: the industry does not intend to give up the flexibility of open weights.

— Olya. There is something curious about watching the keepers of the world's most closed models rush to defend open weights the moment restrictive regulation appears on the horizon. Perhaps safety, like the ecosystem, is a concept that changes shape depending on who is paying the compute bill.