China plans to block US investment in its top AI firms without government approval

China Plans to Block US Investment in Top AI Firms

China plans to restrict its leading technology companies, including top AI startups, from accepting US capital without first obtaining government approval, according to Bloomberg News, citing people familiar with the matter. No official confirmation has been made by the Chinese government.

This move would represent a significant shift, effectively placing US venture capital under the same approval framework already governing technology exports, data flows, and foreign acquisitions in China.

The timing is notable. This comes after the Trump administration's announcement on Wednesday of cracking down on foreign tech companies, particularly China, accused of "exploiting" US AI models through a practice known as model distillation.

White House Director of Science and Technology Policy Michael Kratsios framed this as the first response to complaints from Silicon Valley AI companies about Chinese developers using open-source or commercially accessible US AI models to build rival systems. Bloomberg characterized the US move as targeting Chinese firms "improperly" using American AI models.

Together, these announcements signal a 24-hour escalation where both governments simultaneously moved to sever AI technology and capital transfer channels. The US aims to prevent its models from being used for Chinese training, while China seeks to control American investment flows into its AI national champions.

The reported Chinese capital controls echo existing US outbound investment rules, which prohibit US persons from making equity investments in certain Chinese tech sectors without approval or notification from the Treasury Department. The Chinese plan mirrors this by requiring government approval before Chinese AI companies accept capital from the United States, which has also been restricting chip exports to China since 2022.

The model distillation question is a technical novelty. Chinese developers have used accessible US models like DeepSeek-R1 (Meta’s Llama models) as training signals for their own systems, currently legal under open-source licenses but now subject to potential regulation.