NVIDIA Leads 77-Company Coalition Pushing US to Permit Open-Weight AI Models
Key Takeaways
- ▸NVIDIA leads a 77-company coalition pushing the US government to permit open-weight AI models, framing openness as compatible with national security and competitiveness
- ▸The coalition's composition reveals industry divides: hardware and infrastructure companies leading, while major frontier model labs (OpenAI, Google, notably Anthropic) initially absent or reluctant
- ▸The policy debate centers on US-China AI competition, with the US threatening sanctions against Chinese companies for allegedly distilling American models
Summary
An open letter titled "Open Weights and American AI Leadership" has circulated through Washington with rapidly growing support, attracting 77+ signatories in just days. NVIDIA CEO Jensen Huang, posting about the initiative in his first-ever X post, is among the most prominent advocates. The coalition asks the US government to avoid broad preventive restrictions on open-weight (downloadable) AI models, to expand computing access for startups and researchers, to invest in shared evaluation tools, and to focus enforcement on demonstrated IP theft rather than sweeping sanctions.
The composition of signatories reveals significant industry tensions. Initial supporters included hardware makers (NVIDIA, AMD), cloud providers, software companies, and open-source organizations—but notably excluded the largest frontier model labs. OpenAI and Google only added their names after public pressure, while Anthropic remains notably absent. The letter explicitly defends distillation—the practice of training smaller models on larger systems' outputs—as a legitimate development technique, distinguishing it from outright IP theft.
The coalition's emergence reflects escalating US-China AI competition. The Biden administration has threatened sanctions against Chinese companies like Moonshot AI, which allegedly distilled Anthropic's Claude model to build its Kimi K3, and has considered broader bans on Chinese open-source models. The industry is essentially arguing that overly restrictive US policies would handicap American innovators more than they would contain Chinese competitors, and that these restrictions could chill legitimate development practices used across the industry.
- The letter proposes distinguishing between distillation as a legitimate development practice versus IP theft—a distinction that will be critical to any US regulatory framework
Editorial Opinion
This coalition reveals a fundamental contradiction in proposed US AI policy: policymakers cannot simultaneously contain Chinese competition while handicapping the open development practices that drive American innovation. Even the industry itself remains divided—with Anthropic notably absent—suggesting legitimate concerns about how these policies might work in practice. The critical question for Washington is whether it will recognize the distinction between distillation-as-development and outright IP theft, or whether the pursuit of containment will ultimately weaken American technological leadership.


