Potential US ban on Chinese AI models could cost American businesses US$12b a year: report

A US academic estimates that a potential ban on Chinese open-weight artificial intelligence models could cost American businesses as much as US$12 billion annually, according to calculations by Daniel Yue from the Georgia Institute of Technology. If OpenRouter users are compelled to switch from Chinese models to proprietary alternatives, they might face an extra US$2 billion annual expense, based on token usage and price gaps observed from July 21 to 27.

Extrapolated to the broader US economy, the cost hike could range between US$3 and US$12 billion, depending on the nation's reliance on Chinese open-weight models. The researcher emphasizes these figures as rough approximations rather than definitive projections due to the challenge of tracking usage outside centralized platforms.

The US administration's concerns over Chinese open-weight models surged following the release of Moonshot AI's Kimi K3, a model with capabilities comparable to top US proprietary offerings from Anthropic and OpenAI. This has sparked resistance from major American tech firms, including Nvidia, Palantir, and Meta Platforms, who signed an open letter urging the government against restrictions, warning of potential "stifling of competition or overseas innovation."

However, the economic ramifications remain uncertain as it is difficult to predict if US firms will opt for closed models or abandon specific AI workflows. Venture capitalist Jaya Gupta suggests that many American systems and industries already heavily rely on open-weight models, and an outright ban could lead to a significant drop in AI demand, potentially collapsing the AI infrastructure market.

While some argue that a ban would cause economic pain, others maintain that it would have minimal impact, citing the limited scale of Chinese open-weight model usage by US enterprises and the price drop of proprietary US models.