What happened
Beijing-based AI startup Moonshot AI has reportedly secured access to Nvidia’s advanced Blackwell GPUs by leasing capacity from European 'neocloud' providers. According to a report by The Tech Buzz, this arrangement allows the company to train large language models on hardware that is otherwise prohibited from being exported to China under U.S. semiconductor restrictions. These cloud-based GPU-as-a-service platforms, which have proliferated globally, currently operate in a regulatory gray zone where they are not subject to the same physical export bans that govern the direct sale of chips to Chinese entities.
Moonshot AI, the developer behind the Kimi chatbot, has reportedly turned to European neocloud providers to rent Nvidia Blackwell chips. These chips, which Nvidia claims offer up to 5x the training performance of previous generations, are strictly controlled under U.S. export regulations implemented in October 2023 and subsequently tightened.
The arrangement functions as a 'GPU-as-a-service' model. Because the hardware remains physically located in Europe, the transaction does not constitute a prohibited export of physical goods. This allows Moonshot AI to access high-performance resources that are otherwise unavailable to them within China.
Nvidia has declined to comment on specific customer arrangements involving third-party cloud providers. The company continues to navigate a difficult regulatory environment, balancing compliance with U.S. law while attempting to maintain its market presence in China through the sale of lower-performance, export-compliant chips.
Why it matters
This development exposes a critical vulnerability in U.S. export policy, which focuses primarily on the physical shipment of hardware rather than the remote consumption of power. By utilizing European infrastructure, Moonshot AI can bypass the performance limitations of export-compliant chips designed specifically for the Chinese market. This strategy allows Chinese labs to maintain competitiveness in the global AI race despite Washington's efforts to restrict their access to cutting-edge hardware. The situation creates a complex geopolitical dilemma for U.S. regulators, who must now decide whether to pursue politically sensitive extraterritorial enforcement or accept that cloud-based compute access effectively renders physical export controls incomplete.
The reliance on remote power highlights the limitations of traditional trade policy in the age of cloud-based AI. As long as advanced hardware is accessible via an internet connection, physical border controls may prove insufficient to contain the proliferation of AI capabilities.
For Moonshot AI, this strategy involves significant trade-offs, including higher operational costs, increased , and potential data sovereignty issues. Furthermore, the company faces substantial geopolitical risk; if U.S. pressure forces European providers to terminate these contracts, Moonshot could lose access to its training infrastructure abruptly.
The broader AI infrastructure sector is currently experiencing intense competition for resources. With major players like OpenAI, Google, and Meta aggressively securing Blackwell allocations, the demand from Chinese firms provides a lucrative, albeit risky, revenue stream for smaller, cash-strapped neocloud operators.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Which component of an AI application is the machine-learning model itself?
What to watch next
The primary concern is whether the U.S. government will attempt to close this 'cloud loophole' by pressuring European allies to restrict their cloud providers from serving Chinese customers. Such a move would likely trigger significant diplomatic friction and complicate the business models of emerging GPU-as-a-service providers, who rely on high utilization rates to remain profitable. Additionally, observers should monitor whether other Chinese AI labs adopt similar offshore strategies, potentially leading to a broader crackdown on international cloud access for entities currently under U.S. trade restrictions.
Watch for potential U.S. policy shifts regarding '-as-a-service' exports. Regulators may seek to extend export controls to cover cloud services, which would require unprecedented cooperation from European governments.
Monitor the financial stability of the neocloud sector. As Elon Musk and other industry leaders have suggested, the market may face a shakeout; if providers become overly reliant on Chinese customers, they may become primary targets for future U.S. sanctions.
Observe whether Moonshot AI or other Chinese labs announce new model architectures that specifically account for the and reliability challenges inherent in training on remote, cross-border infrastructure.