Overview
- In an FT op-ed, 01.AI CEO Kai-Fu Lee argues China’s open-source strategy is enabling rapid progress, pointing to DeepSeek’s R1, which he says matched top U.S. models on performance using far less compute and was released for free use and modification.
- Lee cites widespread open-sourcing by Chinese players such as Alibaba’s Qwen, Baidu, Zhipu, Moonshot and Meituan, contrasting it with U.S. developers that keep frontier systems closed despite Meta’s Llama push.
- He highlights developer momentum by noting that, within days of R1’s release, more than 500 derivative models appeared on Hugging Face with about 2.5 million downloads, and he claims newer DeepSeek models rival GPT-5 and Gemini 3 Pro on reasoning tasks.
- The essay frames China’s approach as efficiency-driven under GPU export constraints, while acknowledging that U.S. firms retain a research lead, unfettered access to top chips and paying enterprise customers for closed models.
- Lee likens the landscape to Android versus iOS, suggesting openness can win broad market share, and he references Eric Schmidt’s warning that U.S. companies risk ceding the open-source AI arena to China.