Chinese AI Models Overtake U.S. Rivals in Corporate Usage

Chinese artificial intelligence models have overtaken American rivals in usage among U.S. firms on the OpenRouter platform, now accounting for 58% of processed tokens, up from less than 10% in early 2025. DeepSeek has emerged as the most widely used Chinese model within these firms.

During the World Artificial Intelligence Conference in Shanghai, Moonshot AI unveiled Kimi K3, a 2.8-trillion-parameter open-weight multimodal model. Kimi K3 benchmarks at or above Anthropic's Claude Opus 4.8 and possesses front-end coding capabilities that allow it to fix critical security bugs that U.S. models refuse to address due to safety guardrails. Moonshot AI plans a full open-source release of the model's weights on July 27. Alibaba Group also previewed its trillion-parameter Qwen 3.8 Max system during the conference.

These developments coincided with a 19-day suspension of Anthropic's Fable 5 model following a U.S. government export control directive regarding national security. In response to the rise of Chinese models, the Trump administration is considering restrictions or a ban on these systems. However, David Sacks, Chair of the President's Council of Advisors on Science and Technology, warned that such limits could stifle American innovation and competitiveness.

The shift has impacted global semiconductor stocks as investors question if low-cost open-weight models will reduce demand for expensive AI computing power. To demonstrate domestic hardware capabilities amid U.S. export controls, Huawei showcased its Atlas 950 SuperPoD computing system.