U.S. VC Discovers China’s AI Models Integrated into the American AI Supply Chain

Dongcha Beating AI News: U.S. venture capital firm Dimension spent a week in China conducting research, visiting AI labs, investors, and entrepreneurs, and later wrote a five-page internal letter to its LPs. Dimension had previously visited Shanghai last year; this time, they also traveled to Beijing and Shanghai to recalibrate their assessment of China’s AI landscape. This trip, Dimension highlighted that China’s open-weight models have already entered the production pipelines of U.S. AI companies. Cursor’s Composer 2 was further trained on Kimi K2.5, and the legal AI company Harvey’s Tenet underwent post-training on Kimi K3. They summarized this workflow as “two crossings of the Pacific”: U.S. labs first train cutting-edge models; Chinese labs absorb these capabilities and release open-weight models; U.S. application companies then further train on Chinese models and ultimately deploy them as products for American enterprises. However, while usage has surged, revenue has not proportionally remained with Chinese model companies. Dimension summed it up in one sentence: China is receiving “Western workload, not Western revenue.” Open-weight models can be deployed across multiple platforms, with U.S. inference companies like Fireworks, Baseten, and Modal capturing the majority of service fees. Chinese models drive down costs, and the ultimate beneficiaries include U.S. cloud providers, inference platforms, and AI application companies. Dimension’s final conclusion is that the U.S.-China AI ecosystems can no longer be simply viewed as two independent systems. Hardware such as chips is decoupling—the U.S. is constrained by power and grid capacity, while China lacks advanced chips—but models, data, inference services, and software frameworks continue to flow across borders. Today, it is no longer meaningful to simplisticly ask “Who is winning in U.S.-China AI?”—the real industrial chain is far more interconnected.