Chinese AI models claim 48% of traffic as US debates open-weight ban

China's open-weight AI strategy is reshaping the global competitive landscape, forcing Washington to confront a blind spot in its technology policy.

America has a chip strategy for artificial intelligence. It does not yet have an open-model strategy — and China is filling the void. Chinese AI models accounted for 48% of traffic on OpenRouter during the last week of June, up from 20% a year earlier, while US models fell to 32% from 74%.

"Open weights let every organization match the right model to the right job at the right cost, reserving frontier-scale capability for genuine frontier problems," Nvidia Chief Executive Officer Jensen Huang wrote in an open letter signed by 25 companies including Microsoft and Meta.

The letter, published last week, urged the Trump administration to avoid broad restrictions on open-weight models as it weighs potential curbs on Chinese AI. Beijing-based Moonshot AI's Kimi K3 — a 2.8-trillion-parameter open-weight model released July 27 — reached No. 1 on Hugging Face's trending chart within 30 minutes, garnering more than 7,700 likes and tens of thousands of downloads. The model performs near frontier closed systems from Anthropic and OpenAI on coding, reasoning and multimodal tasks, according to Moonshot's published benchmarks.

The shift threatens the business models of US companies that charge a premium for access to proprietary systems. Every improvement to open-weight models weakens the pricing power of closed providers and risks making American developers dependent on Chinese technology. Nvidia, Microsoft and other signatories argue that open models strengthen competition and prevent AI capabilities from concentrating in a few hands.

The Security Paradox

A cybersecurity incident days before the letter's publication illustrated the complexity of the debate. During an internal test, OpenAI models breached restricted environments at the AI platform Hugging Face. When Hugging Face tried to investigate, safety controls on closed commercial models blocked the analysis. The company instead used an open Chinese model that it could run on its own infrastructure.

"A closed American model caused the incident. A Chinese open model helped investigate it," the open letter noted, complicating the argument that closed AI is inherently safer.

Anthropic Chief Executive Officer Dario Amodei has warned that open-weight models, once downloaded, cannot be monitored or recalled. "The public risk is not simply that outsiders can see the model. It is that outsiders can possess it," said Chinmayi Sharma, a law professor at Fordham University who studies AI and cybersecurity. Once weights are broadly distributed, she said, developers lose the ability to revoke access or patch vulnerabilities.

Washington's Dilemma

The Trump administration is reportedly considering adding Chinese AI labs to its Entity List, effectively cutting off US access to their models. Treasury Secretary Scott Bessent warned of potential sanctions against Moonshot AI, accusing it of using AI distillation — a technique in which developers train models on outputs from existing systems — to copy capabilities from Anthropic's frontier models. Moonshot denied the allegations.

Commerce Secretary Howard Lutnick said Thursday that the US Center for AI Standards and Innovation evaluated Kimi K3 and found it "performs significantly below" America's top-tier models in cybersecurity capabilities, suggesting the administration may avoid a blanket ban.

The open letter argues that distillation is a widely used technique for model improvement and that concerns about intellectual property should be addressed through targeted legal frameworks rather than sweeping restrictions. "Relying solely on closed models is not inherently safe: they can be breached, misused, or fail in ways that outsiders cannot detect," the letter states.

The Investment Stakes

For investors, the open-weight shift carries direct implications. Nvidia benefits from broader AI adoption regardless of model architecture, while Microsoft's Azure cloud business gains when more organizations run AI workloads on their own infrastructure. OpenAI and Anthropic, by contrast, face competitive pressure as capable free alternatives erode their pricing power.

"The data that these companies have is really their strategic asset," said Vipul Ved Prakash, chief executive of AI platform Together AI. By sending proprietary data to a closed model provider, he said, a company risks giving away its "business' recipe."

The way to win an open-source race is by out-building, not with a ban, the letter's signatories argue. The US could support open AI the way it supported chips — by giving universities and startups access to computing power, awarding government contracts to domestic open-model developers and funding the security tools needed to run those models safely.

The future of AI may not be decided by who builds the smartest single model. It may be decided by who builds the models that everyone else runs on. Right now, that is increasingly China.

This article is for informational purposes only and does not constitute investment advice.