China's AI development relies on Western chips, with Moonshot reportedly using 20,000 Nvidia GPUs

Chinese artificial intelligence startup Moonshot is reportedly leveraging a significant computing power deal involving 20,000 Nvidia chips, a move that underscores China's continued dependence on Western semiconductor technology to fuel its burgeoning AI sector. The arrangement, which reportedly includes access to Alibaba Cloud's infrastructure, positions Moonshot's Kimi model as a formidable competitor in the AI landscape. Moonshot's Kimi K3 model, launched in early July, has garnered global attention for its performance, with some benchmarks rivaling advanced models from OpenAI and Anthropic. The deal with Alibaba, one of Moonshot's major investors, grants the startup crucial processing capabilities. However, the partnership is complex, as Moonshot and Alibaba are also competitors. Reports suggest that Kimi's superior performance on key metrics has caused disappointment within Alibaba, as its own AI product, Tongyi Qianwen (Qwen), has been surpassed. Adding to the controversy, Michael Kratsios, head of the White House Office of Science and Technology Policy, recently accused Moonshot of illegally acquiring Nvidia's GB300 Blackwell chips. These chips are considered among the most powerful AI accelerators available and are reportedly being used in Thailand, potentially for training AI models in violation of U.S. export control regulations. The exact number of these advanced accelerators possessed by Moonshot remains unclear. This reliance on foreign technology, particularly high-end GPUs from Nvidia, is a critical factor in China's AI development. While Chinese companies are investing heavily in AI research and development, the U.S. government's export controls on advanced semiconductors aim to limit China's access to the most cutting-edge technology, creating a complex geopolitical dynamic in the global AI race.