Chinese AI startup Moonshot built its Kimi language model on a cluster of roughly 20,000 Nvidia chips supplied by Alibaba, according to Bloomberg.
The arrangement was described by Stocktwits as a reported power agreement between Moonshot and Alibaba covering 20,000 Nvidia chips, and by Yahoo Finance's UK site as Alibaba supplying Moonshot with a major Nvidia AI cluster. Alibaba is an investor-adjacent heavyweight in China's AI scene, and Stocktwits noted that its U.S.-listed shares (BABA) were trading near two-month highs as the report circulated.
The detail matters because computing capacity, not clever code alone, is the main bottleneck in building frontier AI models. Twenty thousand advanced chips is a serious cluster — the kind of hardware base that, until recently, was associated mostly with the largest American labs. That a Chinese startup is training on that scale, with the chips coming by way of Alibaba rather than a direct purchase, says something about how compute is being pooled and shared inside China's AI industry.
Moonshot's Kimi models have meanwhile become a fixture outside China too. Amazon Web Services has published a guide to deploying Kimi K3 on Amazon SageMaker HyperPod and Amazon EKS, meaning the model can be run on standard AWS infrastructure by enterprise customers.
The sources here are news summaries rather than confirmations from the companies; the chip agreement is described as reported, and neither Moonshot nor Alibaba is quoted directly in the available items.
It matters because the story ties together the three forces shaping AI right now — who controls the chips, who can assemble them at scale, and whose models end up running on the world's cloud platforms.