Moonshot AI Reportedly in Talks to Rent Nvidia Blackwell Compute in Europe

CNBC reported on November 5, 2026, that Moonshot AI had discussed leasing Nvidia Blackwell computing capacity from neocloud operators in Europe, citing four people familiar with the private talks. One source described potential clusters of 8,000 or more GPUs.

Three of the sources said Moonshot was seeking Blackwell-powered systems. Two identified interest in European inference clusters, which serve trained models to end users, while one was also aware of interest in training infrastructure. One source characterized the pursuit as an aggressive push for more computing power.

One source placed some European discussions after July 2026, when the White House said Moonshot had accessed Blackwell chips through a facility in Thailand. Moonshot did not comment at the time and did not respond to a request for comment on the new report.

Under the arrangements described in the report, a Chinese AI company would rent capacity from a non-Chinese neocloud or data-center operator able to purchase advanced chips. Customers pay for capacity and make an upfront payment calculated as a percentage of the contract’s lifetime value. One source said a proposal called for Moonshot to pay 30% to 50% more upfront than a non-Chinese customer.

Neoclouds and financing firms can seek more lucrative upfront terms from Chinese customers to hedge against a possible future prohibition on remote-access agreements. Receiving more cash at signing reduces their exposure to revenue later in the contract if the legal framework changes.

Kimi K3’s Capacity Limits

Moonshot, founded in 2023 and based in Beijing, released Kimi K3 in July 2026. The model’s technical blog describes a 2.8-trillion-parameter system with native vision and a one-million-token context window, available through Kimi.com, Kimi Work, Kimi Code and the Kimi API. It recommends supernode configurations with 64 or more accelerators because larger high-bandwidth communication domains benefit inference efficiency.

That statement documents immediate serving pressure around K3’s launch and gives the reported interest in European inference clusters a concrete commercial context.

Remote Access and Export Controls

The regulatory distinction is between physical shipments and remote use. U.S. export controls effectively prohibit Blackwell imports into China, but current law does not give regulators broad authority over access to controlled chips through cloud services hosted overseas.

On January 13, 2026, the Commerce Department’s Bureau of Industry and Security revised its license-review policy for Nvidia H200, AMD MI325X and similar chips, moving applications to case-by-case review when specified security requirements are met. Applicants must demonstrate that exports will not reduce global semiconductor production capacity available to U.S. customers, that the Chinese purchaser has adopted compliance procedures including customer screening, and that the product has undergone independent third-party testing in the United States to verify its performance and security.

The Remote Access Security Act would apply export controls directly to remote access by a foreign person through a network connection, including the internet or a cloud service, from somewhere other than the item’s physical location. It would allow BIS to issue licenses and impose penalties involving controlled items. The House passed the bill on January 12, 2026. The Senate received it on January 13, 2026, and referred it to the Banking, Housing, and Urban Affairs Committee; the official tracker remains at Passed House.

An Nvidia spokesperson said U.S. AI infrastructure must win worldwide, arguing that global cloud dependence on foreign-made chips would mean models and applications are also built abroad.

The reported talks expose the gap between controlling chips and controlling the computation they make available. European leases could give Moonshot access without moving Blackwell hardware into China, but the proposed upfront premium and the pending bill show that remote capacity carries commercial and regulatory exposure of its own.