NVIDIA CEO Jensen Huang made his X debut on July 24, posting a three-page policy letter co-signed by 25 companies and organizations urging Washington to avoid 'premature restrictions' on open-weight AI models. The coalition spans chipmakers (NVIDIA, Dell, IBM), cloud providers (Microsoft), open-weight champions (Meta, Mistral, Black Forest Labs), security firms (CrowdStrike, Palantir), and open-source institutions (Mozilla, Linux Foundation, Hugging Face), plus venture capital (a16z, Y Combinator).
Notably absent: OpenAI, Anthropic, and Google—the three largest closed-model vendors. Both OpenAI and Anthropic are worth roughly $1 trillion each and are preparing massive IPOs potentially in 2026. The coalition's central argument: open models strengthen safety by distributing audit and patch responsibility across many researchers, rather than concentrating control in "single points of failure" at a few labs. They defend distillation (training one model on another's outputs) as legitimate research technique, not theft.
The letter landed exactly one year after the Trump administration made open-weight support official U.S. policy (July 23, 2025). It was triggered by Moonshot AI's Kimi K3, released July 16, which achieved near-Fable 5 performance at a third of the cost and open-weights distribution. The Trump administration has since weighed restricting U.S. access to Chinese models and drafted rules making Americans liable for running them.
For architects evaluating the open/closed divide in production AI: the July 24 letter signals a structural fracture. Huang's thesis is that commodity open models + custom enterprise fine-tuning will dominate inference spend, while closed shops lose margin to API erosion and compliance costs. OpenAI/Anthropic's absence is a tell: they cannot endorse a position that validates cost-competitive Chinese open models without undermining their own valuation narratives.