Latest Assessment by US VC After Visiting China: AI Capabilities Need to Span Two Round Trips Across the Pacific to Sell to US Customers

Insightful Beating AI News Flash: U.S. venture capital firm Dimension conducted a week-long research trip to China, visiting AI labs, investors, and entrepreneurs. Upon returning, they wrote a 5-page internal letter to their Limited Partners (LPs). Dimension had conducted a similar research trip to Shanghai last year and this time, they expanded to Beijing and Shanghai, aiming to reassess their outlook on China's AI landscape.

This time, Dimension emphasized that China's deployment of open-weight models has entered the production chain of U.S. AI companies. Cursor's Composer 2 continued training from Kimi K2.5, and the legal AI firm Harvey's Tenet also underwent post-training on Kimi K3. They summarized this chain as a "cross-Pacific round trip": U.S. labs initially train cutting-edge models, Chinese labs absorb these capabilities and release open-weight models, U.S. application companies further train the Chinese models, and ultimately commercialize the products back to U.S. enterprises.

However, despite the increased usage, the revenue has not stayed proportionally with the Chinese model companies. Dimension succinctly stated: China receives "Western effort, but not Western revenue." Open-weight models can be deployed on various platforms, with U.S. inference companies such as Fireworks, Baseten, and Modal charging service fees. Chinese models drive down costs, benefiting U.S. cloud providers, inference platforms, and AI application companies in the end.

Dimension's final assessment is that differentiating between Chinese and American AI systems is becoming increasingly challenging. Hardware like chips is undergoing decoupling, with the U.S. constrained by power and the grid, and China facing a shortage of advanced chips. However, models, data, inference services, and software frameworks continue to flow across borders. Today, discussing simply "who is winning in U.S.-China AI" no longer accurately depicts the complex industry chain.