According to people familiar with the matter who revealed to CNBC, after the White House accused China's artificial intelligence laboratory, Moon's Dark Side (Moonshot AI), of remotely obtaining advanced NVIDIA chips from East Asia, the company sought to obtain the rights to use these chips through data centers in Europe.
Four people familiar with the matter stated that this year, there were discussions regarding the possibility of a “new cloud” (neocloud) service provider, which currently hosts NVIDIA graphics processors (GPU) at its European facilities, renting out computing power. As these discussions involved undisclosed information, these individuals have requested to remain anonymous.
NVIDIA is virtually monopolistic in the field of the most powerful chips, making U.S. export controls a key tool for Washington to maintain its advantage over China in the race for artificial intelligence. American laboratories such as OpenAI, Anthropic, and AI are competing with Chinese developers, including those from "The Dark Side of the Moon"; "The Dark Side of the Moon" develops the Kimi series of models.
Three sources told CNBC that the 'Dark Side of the Moon' is seeking to rent computing power driven by the NVIDIA Blackwell system. Blackwell involves advanced chips that are currently subject to U.S. export controls, effectively prohibiting their import into China.
In July this year, the White House stated that this Chinese AI laboratory had obtained NVIDIA Blackwell chips through a facility in Thailand.
U.S. export controls restrict the shipment of NVIDIA's most advanced chips to China, but remote access through overseas data centers provides a loophole. Current laws do not grant regulatory agencies broad powers to oversee companies' acquisition of these chips via overseas hosted cloud services.
Neither the dark side of the moon, NVIDIA, the White House, the US Department of Commerce, the Chinese Embassy in the UK, nor the Chinese Ministry of Foreign Affairs responded to requests for comments.
In the past six months, Chinese developers have made significant progress in their capabilities, and if they could obtain more NVIDIA chips, it would likely further increase their pressure on their American competitors.
If a Chinese AI company wants to rent NVIDIA computing power in Europe, it usually rents capacity from non-Chinese “new cloud” service providers or data center operators, as the latter can purchase advanced chips.
Two people familiar with the matter said that Dark Side of the Moon is interested in renting clusters in Europe for inference purposes. By inference, it refers to providing models for end-users to use. One of the sources also mentioned that they are aware that Dark Side of the Moon is interested in training clusters as well.
A source stated that the discussions involved renting access to a cluster hosted locally, consisting of 8,000 or more Blackwell GPU, which are used for certain purposes. The source described this Chinese AI laboratory as being "very aggressive" in its pursuit of additional computing power.
One of the sources told CNBC that after the White House claimed that the Dark Side of the Moon had obtained relevant computing power through Thailand, negotiations regarding its acquisition of computing power in Europe were still ongoing. The Dark Side of the Moon did not comment on this at the time.
Competing for computing power
Based in Beijing, Moon's Dark Side was established in 2023 and is one of China's leading large-scale model developers. The company released a new model in July, claiming that this model has narrowed the gap with the leading systems Anthropic and OpenAI.
This model has been widely adopted as a cheaper alternative to similar products in the United States. In 2026, with improved capabilities and lower costs compared to its American competitors, the adoption rate of the Chinese AI model surged globally.
Before and after the release of Kimi K3, the "Dark Side of the Moon" indicated that they had reached their maximum computing power capacity. As a result, the company temporarily suspended new subscriptions to prioritize ensuring the computing power for existing members. In a post on LinkedIn, the company stated, "We are increasing our capacity as quickly as possible and will gradually reopen new subscription slots."
Renting AI computing power has become a booming industry. Customers, namely the purchasers (offtakers), typically obtain capacity by paying a fee and also pay an upfront cost to lock in the transaction. This fee is usually a certain percentage of the total value of the contract over its lifetime.
A source said that it was once proposed that the upfront payment for the dark side of the moon should be 30% to 50% higher than that for non-Chinese buyers.
Arrangements reached with Chinese customers may be more profitable in the early stages, as "new cloud" service providers and financing institutions wish to hedge against the risk that future remote access protocols may be prohibited.
Political pressure
In Washington, lawmakers have proposed the Remote Access Security Act ( Remote Access Security Act, RASA ), which aims to extend U.S. export controls to cover cloud access for critical hardware and software. The bill was passed by the House of Representatives in January but has not yet been approved by the Senate.
The British "Financial Times" reported on Tuesday that this chip giant has informed cloud service providers that sell access to its chips that they must sell at least 51% of their services to non-Chinese customers.
Earlier this week, when asked about related reports, a spokesperson for NVIDIA said: "If the US's AI infrastructure cannot succeed globally, then the US cannot win the AI competition – if global cloud services rely on chips manufactured abroad, then the model layer and application layer will also be built by foreign entities."
In August, NVIDIA stated that Washington had approved the sales licenses for some of its H200 AI chips (an older generation of products), but that "such sales are restricted" as the Chinese government has been attempting to support local chip manufacturers by limiting access to NVIDIA chips.
