Moonshot AI Seeks Nvidia Blackwell Chips in Europe Amid U.S. Export Controls

Moonshot AI, a Beijing-based artificial intelligence firm, has sought to rent access to Nvidia’s Blackwell graphics processing units through data centers in Europe. Four people familiar with the matter said these discussions took place earlier this year. The talks involved neocloud providers that host Nvidia hardware at European facilities. The sources requested anonymity because the negotiations were private.

Nvidia holds a dominant position in the market for the most powerful AI chips. This monopoly has made U.S. export controls a central tool in Washington’s effort to maintain an advantage over China. American companies like OpenAI and Anthropic are currently competing with Chinese developers, including Moonshot, which builds the Kimi family of models. The competition centers on model training and pricing.

Three of the four sources stated that Moonshot specifically wanted to lease capacity powered by Blackwell systems. These chips are subject to export controls that effectively ban their import into China. In July, the White House claimed the Chinese lab accessed Blackwell chips via a facility in Thailand. Current U.S. law does not grant regulators broad authority over cloud services hosted abroad. This creates a loophole for remote access through overseas data centers.

Two sources indicated that Moonshot was interested in European clusters for inference, a process that serves models to end users. One source also noted interest in training clusters. One source described the lab as very aggressive in its search for computing power. The arrangements under discussion would provide access to clusters containing 8,000 or more Blackwell GPUs in the region. Another source confirmed that negotiations on European compute began after the White House made its claim regarding Thailand. Moonshot declined to comment on the Thailand issue at that time.

Moonshot was founded in 2023 and is based in Beijing. It released a new model in July that the company said closed the performance gap with leading systems from Anthropic and OpenAI. The model saw strong adoption as a lower-cost alternative to U.S. offerings. Around the time its Kimi K3 model launched, Moonshot said demand exceeded its available capacity. The company temporarily paused new subscriptions to manage this load. In a LinkedIn post, the firm stated it was adding capacity as fast as possible and would reopen subscription spots in batches.

Leasing AI compute is a growing business where customers typically pay an upfront fee. This fee represents a percentage of the total contract value. One source said Moonshot was proposed to pay 30% to 50% more upfront than a non-Chinese customer would. These arrangements can be more profitable for neoclouds and their financing partners. The partners use this premium to hedge against the risk that remote access agreements might be banned in the future.

In Washington, lawmakers have proposed the Remote Access Security Act to extend export controls to cloud-based access to critical hardware and software. The bill passed the House of Representatives in January but has not yet cleared the Senate. The Financial Times reported earlier this week that Nvidia told cloud providers selling access to its chips that they must sell at least 51% of services to non-Chinese customers. An Nvidia spokesperson said, America cannot win the AI race unless America’s AI infrastructure wins worldwide.

Nvidia announced in August that Washington had licensed some sales of its older-generation H200 AI chips. The company noted that such sales were restricted by the Chinese government, which has limited access to Nvidia products to boost domestic chipmakers. Moonshot, Nvidia, the White House, the U.S. Department of Commerce, the Chinese embassy in the U.K., and China’s Ministry of Foreign Affairs did not respond to requests for comment.

The discussions between Moonshot and European providers remain in the negotiation phase. No final contract has been publicly announced. The lab continues to seek additional computing power to support its growing user base.