Moonshot AI signed a contract in August this year to lease 32 Super Micro servers equipped with Nvidia B300s
According to Bloomberg, documents submitted to Beijing regulators this summer show that Chinese AI company Moonshot AI (Moonshot, the developer of Kimi) has obtained the right to use Nvidia’s most advanced chips. The company signed a contract in August 2026 to lease 32 Super Micro Computer servers, which are equipped with Nvidia B300 GPUs.
The B300 is the most powerful AI accelerator in Nvidia’s Blackwell lineup and is subject to U.S. export controls on shipments to China. The report noted that 32 refers to the number of servers. The information in the documents is currently insufficient to confirm the actual delivery status of the equipment or its final deployment location. Bloomberg made the above report based on regulatory filing documents it reviewed, underscoring that even under U.S. export restrictions, this leading Chinese AI model company still relies on U.S. hardware.
This is not the first time Moonshot has been mentioned in connection with seeking restricted Nvidia computing power. In July this year, White House officials publicly accused the company of using Blackwell chips through a data center in Thailand. Last week, CNBC reported that Moonshot had discussed with a European cloud service provider the possibility of leasing a cluster of at least 8,000 Blackwell GPUs, mainly for inference, with training needs also involved, but the relevant talks have not yet been confirmed as having resulted in a deal.
Original source:
Title: Moonshot Rents Nvidia’s Top AI Chips in China, Documents Show Publisher: Bloomberg Publication date: 2026-11-09 URL: https://www.bloomberg.com/news/articles/2026-11-10/moonshot-rents-nvidia-s-top-ai-chips-in-china-documents-show Nature of source: media report (based on reviewed documents from Beijing regulators) Verification status: Bloomberg’s publicly available summary and multiple secondary reports were read and the core facts were cross-checked; the actual delivery and deployment location of the equipment have not been independently verified. Supporting content: signing in August to lease 32 Super Micro servers equipped with B300s, source documents, export-control background.
ABAB AI Insight
The leasing structure rather than direct purchase allows a Chinese AI company to obtain usage rights without holding ownership of restricted hardware. U.S. export controls mainly target the export and re-export of chips and servers, and the extent to which they cover remote leasing of compute resources hosted overseas is limited. The arrangement in Moonshot’s filing points to deployment in China (some secondary reports mentioned equipment valued at about RMB 432 million and located in China). If ultimately confirmed and if no corresponding license exists, it could touch the boundary of the controls, but at present it remains at the stage of contract signing and filing.
From the perspective of compute acquisition channels, Moonshot has been reported multiple times as trying different avenues: the July accusations regarding Thailand, the recent European talks, and this domestic server leasing contract. These paths do not replace one another; they are parallel attempts. The European option is larger in scale (the discussion involved 8,000+ GPUs in a single instance), but it is still at the negotiation stage; this batch of 32 servers is more limited in scale and is more likely to be used for specific inference or development needs rather than large-scale training.
Beijing continues to promote domestic compute substitution, yet leading model companies are still seeking Blackwell-class GPUs worldwide, reflecting the fact that domestic chips still lag in energy efficiency, software ecosystem and cluster stability in the competition for frontier model performance. The leasing model turns “chips” into “services,” extending the object of controls from customs cargo to cloud contracts and data center operators. If regulators want to tighten further, they would need to cover overseas hosting, third-party leasing and remote access, rather than only hardware exports themselves.
The United States is trying to limit the spread of AI capabilities by controlling the supply of advanced chips, but when compute is repackaged as a service, the control point has shifted from the port to the contract and compliance chain. Moonshot’s continued moves show that, at least in the short term, China’s top AI labs still see Nvidia’s most advanced GPUs as an option that is difficult to fully replace.