Moonshot AI Seeks 8,000-Plus Nvidia Blackwell GPUs in Europe as US Curbs Leave Cloud Loophole
Updated
Updated · CNBC · Nov 5
Moonshot AI Seeks 8,000-Plus Nvidia Blackwell GPUs in Europe as US Curbs Leave Cloud Loophole
3 articles · Updated · CNBC · Nov 5
Summary
Moonshot AI has discussed renting European data-center capacity built on Nvidia Blackwell chips, with talks covering clusters of 8,000 or more GPUs for inference and possibly training.
Those discussions exploit a gap in U.S. export controls: China cannot import Blackwell systems, but current law does not broadly bar overseas cloud access to the chips.
The push comes after the White House said in July that Moonshot had remotely accessed Blackwell chips through a facility in Thailand, and sources said Europe talks continued after that disclosure.
Moonshot's demand for compute has risen as its Kimi models gained traction in 2026; one source said Chinese customers may pay 30% to 50% more upfront for capacity.
Washington is weighing tighter rules through the Remote Access Security Act, while Nvidia has also told cloud providers to keep at least 51% of services for non-Chinese customers, the Financial Times reported.
Will the massive premiums paid by Chinese AI labs tempt European data centers to risk US regulatory backlash over Blackwell GPUs?
How long can Chinese AI giants exploit European cloud loopholes to access banned Nvidia chips before US lawmakers shut the door?
If cloud computing replaces hardware ownership, can any export control truly stop a trillion-parameter AI model from training across borders?
The $1.5 Trillion Cloud Loophole: How Chinese AI Firms Bypass U.S. Chip Controls via Global Compute Arbitrage
Overview
In 2026, U.S. export controls blocked physical shipments of Nvidia Blackwell GPUs to China, but did not prevent Chinese firms like Moonshot AI from remotely accessing these chips in Europe through cloud networks. After Moonshot’s Kimi K3 model launch caused a surge in demand that overwhelmed its infrastructure, the company negotiated with European neocloud operators to rent large GPU clusters, paying high risk premiums due to looming U.S. legislative efforts to close this loophole. As U.S. regulators and Nvidia tightened operational rules, Chinese AI developers continued to rely on overseas compute, highlighting the ongoing tension between technological ambition and regulatory barriers.