Chinese AI startup Moonshot is training its Kimi AI model on a cluster of 20,000 Nvidia chips, supplied through Alibaba's cloud infrastructure. The move underscores how Chinese companies are still finding ways to access advanced hardware despite ongoing U.S. export restrictions.
How the cluster came together
Moonshot, a Beijing-based company founded in 2023, secured the Nvidia chips through Alibaba Cloud. The cluster is one of the largest known AI training setups in China. Nvidia's high-end chips, including the A100 and H100, have been subject to U.S. export controls aimed at limiting China's access to cutting-edge semiconductor technology. But companies like Alibaba have maintained access through existing stockpiles and cloud services.
The deployment signals that China's AI sector continues to scale up despite geopolitical headwinds. With a 20,000-chip cluster, Moonshot can train large language models comparable in size to those built by American rivals. The company's Kimi AI is designed to compete with models from OpenAI and Google. Chinese regulators have encouraged domestic AI development as part of a broader push for technological self-reliance.
Global implications
The development could shift the balance in the global AI race. While U.S. restrictions aim to slow China's progress, the ability to assemble large clusters through cloud providers suggests the gap may not widen as quickly as some policymakers hoped. At the same time, the reliance on Nvidia hardware — even via third parties — means China's AI sector remains dependent on American chip designs, a vulnerability that Beijing is trying to address through domestic chipmaking efforts.
Moonshot has not disclosed the exact cost or timeline of the project. The company declined to comment on whether it had sought or received any special export licenses for the chips.
