The US wants to block China from renting AI power abroad but China already built a parallel supply chain from US AI output
First reported by The Decoder ·
Access to leading US AI models for training student models now requires navigating licensing and potentially violating terms of service.
The US is moving to close a loophole that allowed Chinese companies to access advanced AI computing power abroad, circumventing export bans. The Bureau of Industry and Security (BIS) at the US Commerce Department plans to implement an interim rule requiring a US government license for data center operators in countries like Singapore, Malaysia, and Brazil to purchase US-made AI chips. This measure aims to align remote-access rules with existing physical export restrictions. The announcement follows reports of Nvidia CEO Jensen Huang's affair with a former employee suspected of espionage related to banned Nvidia chips found in a Chinese data center. Concurrently, a Reuters investigation reveals China's extensive use of "model distillation" to train its own AI models, effectively creating a parallel supply chain by extracting outputs from powerful US AI models. This technique reduces the need for high-end domestic hardware, though it may violate the terms of service of US AI developers.
The US move to license third-country data centers signals a heightened geopolitical struggle over AI development, extending export controls beyond physical chip shipments to encompass the remote use of AI capabilities. This broadens the scope of US sanctions, impacting global cloud infrastructure providers and potentially fragmenting the AI ecosystem. The Treasury Department and National Security Council's direct involvement indicates this is a high-priority national security issue, suggesting further aggressive actions may follow if loopholes persist.
China's sophisticated "model distillation" industry, developed to circumvent chip embargoes, highlights a strategic adaptation to hardware limitations, creating a parallel AI supply chain reliant on US model outputs. While this allows China to train smaller, more cost-effective AI systems for domestic deployment, it also raises concerns about intellectual property and violates US AI developers' terms of service. The effectiveness of US countermeasures against distillation, as noted by Anthropic, will be crucial in determining the future of AI development and the enforcement of US export controls.
AI-written summary. May contain errors.