Moonshot AI sought access to Nvidia Blackwell chips in Europe to get around US controls

According to four sources familiar with the matter cited by CNBC, Chinese artificial intelligence start-up Moonshot AI, maker of the Kimi family of models, has sought access to advanced Nvidia chips housed in European data centers. Deals to rent capacity from "neoclouds" (GPU-specialized cloud providers) with facilities in Europe were reportedly discussed this year. The sources requested anonymity because the information is private. Moonshot, Nvidia, the White House, the U.S. Department of Commerce, the Chinese embassy in the UK and China's Foreign Ministry did not respond to requests for comment.

The backdrop is the tech rivalry between Washington and Beijing. Nvidia's near-monopoly on the most powerful chips has made export controls a key U.S. tool for preserving its edge in the AI race. The controls restrict shipments of Nvidia's most advanced chips to China, and Blackwell systems are effectively barred from import. However, remote access through overseas data centers works as a loophole: current law does not give regulators broad authority to control companies' access to those chips via cloud services hosted abroad.

Three of the sources said Moonshot wanted to lease capacity based on Blackwell systems. Two of them specified that the interest centered on clusters for inference, meaning serving models to end users, and one source was also aware of interest in training clusters. One source said talks had been held on rental deals for clusters of 8,000 or more Blackwell GPUs hosted in the region, and described the Chinese company as "very aggressive" in its pursuit of more computing power. Another source said the talks to secure compute in Europe had taken place since the White House claimed Moonshot had already done so through Thailand.

That precedent is relevant: in July, the White House alleged that the Chinese lab had accessed Nvidia Blackwell chips through a facility in Thailand. Moonshot did not comment at the time.

Moonshot, founded in 2023 and based in Beijing, is one of China's leading model developers. In July it launched a new model, Kimi K3 according to the article, which the company said closed the gap with leading systems from Anthropic and OpenAI, and which saw notable adoption as a cheaper alternative to U.S. offerings. According to CNBC, Chinese AI models have gained global adoption in 2026 thanks to capability improvements and lower costs than their U.S. rivals, and greater access to Nvidia chips could increase the pressure on those rivals even further. With the launch, Moonshot said demand had strained its capacity and temporarily paused new subscriptions to prioritize compute for existing members; the company said on LinkedIn it was adding capacity "as fast as we can" and would reopen spots in batches.

The article also describes the economics of these deals. Leasing AI compute is a booming business in which customers pay a fee plus an upfront payment, which is a percentage of the contract's value over its full lifetime. According to one source, it was proposed that Moonshot pay between 30% and 50% more upfront than a non-Chinese customer. Deals with Chinese customers can be more lucrative at first because neoclouds and financial firms seek to hedge against a possible future ban on remote-access arrangements.

On the policy front, U.S. lawmakers have proposed the Remote Access Security Act (RASA), which would extend export controls to cloud access to critical hardware and software. It passed the House of Representatives in January but not the Senate. In addition, the Financial Times reported on Tuesday that Nvidia had told cloud providers selling access to its chips that they must allocate at least 51% of their services to non-Chinese customers. An Nvidia spokesperson, asked about those reports, said the U.S. cannot win the AI race unless its AI infrastructure succeeds around the world, because if the world's clouds depend on foreign chips, the model and application layers will be foreign too.

Finally, Nvidia said in August that Washington had authorized certain sales of its previous-generation H200 chips, but that the Chinese government restricted them, as it seeks to boost domestic chipmakers by limiting access to Nvidia's.

It is worth stressing that the account relies on anonymous sources and there is no indication that any deal has been closed: the article refers to talks and interest, not confirmed contracts.