Moonshot AI secured a computing-power agreement with Alibaba Group Holding Ltd. to use approximately 20,000 Nvidia H200 chips [1].
The deal allows the Chinese AI startup to train its latest Kimi K3 model using high-end hardware that is otherwise difficult to acquire. This partnership highlights the critical role of cloud infrastructure in the race for artificial intelligence supremacy within China.
According to reports published late last month, the cluster of 20,000 chips [1] was supplied by Alibaba to provide the massive compute required for the development of Kimi K3 [2]. The arrangement ensures that Moonshot AI has the necessary processing power to compete with global AI leaders, a feat that requires immense hardware scaling.
Industry analysts said the move is particularly significant given the current geopolitical climate. The agreement enables Moonshot AI to advance its capabilities despite ongoing U.S. export curbs on advanced semiconductors [3]. By leveraging Alibaba's existing infrastructure, the startup can bypass some of the hurdles associated with procuring chips directly from the U.S.
Alibaba has emerged as a primary provider of compute for other Chinese AI champions. This specific deal for the Kimi K3 model underscores a trend where large conglomerates bankroll the infrastructure for smaller, agile AI firms [4]. However, some reports said that this shift in compute allocation may impact Alibaba's own AI projects, such as its Qwen model [2].
Moonshot AI has not issued a public statement regarding the specific terms of the lease, but the deployment of the H200 cluster marks one of the largest single-model training efforts in the region [1].
“Moonshot AI secured a computing-power agreement with Alibaba Group Holding Ltd. to use approximately 20,000 Nvidia H200 chips”
This partnership illustrates a strategic pivot in China's AI ecosystem, where the bottleneck is no longer just algorithmic design but physical hardware access. By utilizing Alibaba as a proxy for compute, Moonshot AI demonstrates how domestic cloud giants can mitigate the impact of U.S. trade restrictions, effectively creating a shared resource pool to maintain national competitiveness in large language model development.
