How China's AI Industry Rapidly Closed the Gap with US Competitors
A Wall Street Journal report highlights that the rapid advancement of Chinese AI models is driven by long-term academic talent networks centered around Tsinghua University, open-source collaboration, returning overseas talent, and highly efficient computing techniques. Key figures driving this progress include Tang Jie of Zhipu AI, Yang Zhilin of Moonshot AI, and Liang Wenfeng of DeepSeek. Despite facing severe US chip sanctions and having less than one-fifth of the funding of US tech giants, Chinese AI startups have narrowed the capability gap with top US models to just a few months. This demonstrates that architectural innovations and collaborative knowledge sharing can compensate for hardware limitations in the global AI race. Companies like DeepSeek bypassed compute constraints by adopting Multi-head Latent Attention (MLA) and Mixture-of-Experts (MoE) architectures, which significantly reduce memory and compute requirements. Furthermore, Chinese AI firms actively learn from each other's open-source models and research papers, creating a rapid, decentralized cycle of technological iteration.
## BACKGROUND
The foundation of China's AI talent pool was laid over two decades ago, notably with Turing Award winner Andrew Yao founding the "Yao Class" at Tsinghua University in 2005. Major Chinese AI startups like Zhipu AI, Moonshot AI, and DeepSeek have since emerged as key players, often releasing high-performing open-weights models. However, access to advanced semiconductors remains a critical bottleneck, with Chinese labs reportedly having only a fraction of the high-end chips available to US counterparts like OpenAI or Google.