AI Giants Back Open Access to Chinese Models Amid US Tension

A Surprising Show of Unity on Open Models

Dozens of the world's largest AI companies, including Nvidia, Meta, OpenAI, and Microsoft, have signed an open letter arguing that access to the most capable AI models should remain available regardless of where they originate — explicitly including China 1. The letter frames unrestricted model access as essential to innovation and competition, a notable stance given the geopolitical friction surrounding Chinese AI development and the increasingly tangled relationship between US and Chinese firms building on similar underlying technology.

Whether the sentiment is sincere is the central question raised by the letter's coverage. Skeptics note that many signatories have direct commercial interests in keeping the door open to Chinese models — Nvidia depends on global chip demand, and US labs benefit from a competitive, fast-moving open ecosystem that keeps pressure on pricing and capability. The letter arrives at a moment when the US government is simultaneously scrutinizing Chinese AI firms for allegedly appropriating American technology, making the industry's public embrace of openness look, to some, like a hedge rather than a principle 1.

Allegations of Theft Complicate the Picture

That tension is sharpened by a separate, high-profile dispute: US officials have accused China's Moonshot AI of stealing from Anthropic's advanced model, internally referred to as Fable, to help build its newly released K3 model 34. A senior White House technology official said Moonshot also acquired advanced Nvidia chips despite export controls meant to limit China's access to cutting-edge hardware 3. Reuters reporting on the episode underscores how quickly allegations of intellectual-property theft can undercut the collaborative language of the open letter, revealing a US-China AI relationship defined as much by suspicion and enforcement as by shared research norms 4.

Regulatory Pressure Builds at Home

Domestic politics add another layer of complexity. Congress is separately considering legislation that would give the Department of Homeland Security a so-called AI "kill switch," empowering regulators to shut down models judged to pose a risk to human life or the economy 5. That proposal, prompted partly by concerns tied to a recent OpenAI incident, signals that lawmakers are increasingly comfortable intervening directly in how AI systems are deployed — a posture that could just as easily be turned toward foreign models as domestic ones.

Practical Realities for Businesses and Users

Away from the geopolitics, the commercial AI market is being shaped by more mundane pressures. Analysts argue that most everyday users don't need the newest, priciest models at all, since good prompting on older systems often suffices outside of specialized tasks like coding or media generation 2. Enterprise leaders are meanwhile focused on operational integration rather than raw model superiority: Box CEO Aaron Levie says AI's next phase hinges on embedding models into real business workflows 6, while Microsoft CEO Satya Nadella has warned that companies betting everything on a single model risk losing control over their data and independence 7. Together, these threads suggest an industry publicly championing openness while privately navigating competition, security, and dependency concerns on multiple fronts.