Moonshot AI Sought Nvidia Blackwell Capacity in Europe, Sources Say

Moonshot AI, the Beijing-based Chinese AI lab that the White House accused in July of accessing Nvidia's advanced Blackwell chips remotely through a facility in Thailand, has since pursued deals to rent the same restricted chips from data centers in Europe, four people familiar with the matter told CNBC.

The talks involved so-called neoclouds — smaller cloud providers that rent out raw GPU computing power rather than full cloud platforms — hosting Nvidia graphics processing units at European facilities. Three of the sources said Moonshot wanted capacity powered by Nvidia's Blackwell systems, the chip generation currently covered by U.S. export controls that effectively ban its import into China.

The company, which builds the Kimi family of models, did not respond to a request for comment. Nvidia, the White House, the U.S. Department of Commerce, China's embassy in the U.K. and China's Ministry of Foreign Affairs also did not respond.

Why does remote access matter for U.S. export controls?

Nvidia holds an effective monopoly over the most powerful AI chips. That dominance has made U.S. export controls a central instrument in Washington's effort to preserve its advantage over China in the AI race, at a moment when American labs such as OpenAI and Anthropic are locked in a training and pricing battle with Chinese developers.

Current law, however, does not give regulators broad authority to control how companies access those chips through cloud services hosted overseas. Remote access from overseas data centers is the loophole Moonshot appears to be exploiting. Beijing has done little to close it, and Washington so far cannot.

To rent Nvidia computing power in Europe, a Chinese AI company would typically lease capacity from a non-Chinese neocloud or data center operator that is legally able to purchase advanced chips. That structure keeps the transaction formally outside U.S. export rules — at least for now.

The gap in enforcement is the story here: chips Washington says China cannot have are reachable by anyone with a cloud contract.

How aggressive was Moonshot's pursuit of compute?

One source told CNBC that talks covered arrangements for Moonshot to rent access to clusters of 8,000 or more Blackwell GPUs hosted in Europe. The source described the Chinese AI lab as "very aggressive" in its pursuit of more computing power. A cluster of that size represents industrial-scale capacity, enough to serve or train frontier-class models.

Two sources said Moonshot was interested in leasing capacity for inference — running trained models to serve end users — at European sites. One of those sources said they were also aware of interest in clusters for training, the far more compute-intensive process of building models from scratch. A training ambition would signal more than commercial deployment; it would mean Blackwell capacity feeding directly into Moonshot's next generation of models.

One source said the European talks took place after the White House said in July that Moonshot had accessed Blackwell chips via Thailand. Moonshot did not comment at the time.

Take the picture as a whole: even after a public White House accusation, the lab kept shopping for restricted silicon abroad.

What does the pricing tell us about the risk?

Leasing AI compute is a booming industry. Customers — known as offtakers — typically take capacity for a fee and pay an upfront cost to secure the deal, calculated as a percentage of the lifetime contract value.

One source said it had been proposed that Moonshot pay 30% to 50% more upfront than a non-Chinese offtaker would. That premium is not arbitrary. Neoclouds and the financing firms behind them are hedging against a potential future ban on remote access agreements with Chinese customers, and they want compensation for that risk now, in cash, rather than in monthly payments that could be cut off by new legislation.

The premium is, in effect, a market-priced estimate of how likely providers think Washington is to close the loophole — and it is high enough to suggest providers see a ban as a live possibility.

Why does Moonshot need the compute so badly?

Moonshot AI was founded in 2023 and has become one of China's leading model builders. In July it released a new model that it said closed the gap with leading systems built by Anthropic and OpenAI. The model saw significant uptake as a cheaper alternative to U.S. offerings, and Chinese AI models broadly have surged in global adoption in 2026 on the strength of capability gains and lower costs than their American rivals.

Demand strained the company's own infrastructure. Around the time it released Kimi K3, Moonshot said demand had stretched its capacity, and it temporarily paused new subscriptions to prioritize compute for existing members. "We're adding capacity as fast as we can and will reopen new subscription spots in batches," the company said in a LinkedIn post.

That supply crunch explains the aggression. Chinese developers have made notable capability gains in the past six months, and greater access to Nvidia chips could put even more pressure on their American rivals — the direct competitive stake behind the export-control fight.

Is Washington moving to close the loophole?

Lawmakers have proposed the Remote Access Security Act, or RASA, which would expand U.S. export controls to include cloud-based access to critical hardware and software. The bill passed the House of Representatives in January but has not passed the Senate. Until the Senate acts, remote access remains legal in practice for overseas-hosted capacity.

Nvidia, for its part, is applying commercial pressure of its own. The Financial Times reported on Tuesday that the chip giant told cloud providers selling access to its chips that they must sell at least 51% of services to non-Chinese customers — a contractual cap that would constrain, though not eliminate, how much capacity a Chinese lab could lease.

Asked about the reports earlier this week, an Nvidia spokesperson said: "America cannot win the AI race unless America's AI infrastructure wins worldwide — if the world's clouds rely on foreign-made chips, the model and application layers will be foreign-built as well."

The company is caught between two governments. Nvidia said in August that Washington had licensed some sales of its H200 AI chips, an older generation, but that "such sales were restricted" by the Chinese government, which has sought to boost domestic chipmakers by limiting access to Nvidia chips. Beijing is pushing its market away from Nvidia even as Chinese AI labs chase Nvidia silicon abroad.

Neither the House-passed bill nor Nvidia's 51% rule has the force of an enforceable export control over foreign-hosted cloud capacity — which is precisely the space where Moonshot has been operating.

What happens next?

The unanswered question is enforcement. If the Senate passes RASA in some form, the remote-access arrangements now being priced at a 30% to 50% upfront premium would become legally exposed, and neoclouds holding Chinese offtake contracts would face losses they have already been hedging against. Until then, the market signals its own verdict: providers will rent to Chinese customers, at a price that reflects the risk that the loophole closes.

For Moonshot, the calculus is equally clear. A lab that paused subscriptions because it ran out of compute, and that kept negotiating for 8,000-GPU Blackwell clusters after a White House accusation, is telling investors and rivals that compute — not model architecture — is its binding constraint.