China intends to prohibit US investments in its leading AI companies unless there is government approval.

      Two parallel actions within 24 hours signify a notable escalation in the US-China AI conflict, shifting focus from chips and exports to capital and models.

      According to a Bloomberg News report on Friday, China intends to impose restrictions on its leading tech firms, including prominent AI startups, requiring them to seek government approval before accepting US investments, as per sources familiar with the situation. No official from the Chinese government has verified this report. Should this plan be realized, it would represent a fundamental change in how Chinese AI businesses secure foreign funding, effectively aligning US venture capital with the approval regulations that already apply to specific technology exports, data transfers, and foreign acquisitions of Chinese assets.

      The timing of this development is not coincidental. On Wednesday, the Trump administration announced intentions to tighten regulations on foreign tech firms, specifically targeting China, that are allegedly “exploiting” US AI models, a practice referred to as model distillation.

      Michael Kratsios, the White House Director of Science and Technology Policy, described this initiative as the first significant response from the US government to concerns from Silicon Valley AI companies regarding Chinese developers leveraging open-source or commercially available US AI models as training data for building competing systems, thereby narrowing the capability gap without starting from scratch.

      Bloomberg characterized the US initiative as aimed at Chinese companies that are “improperly” utilizing American AI models. Together, the two announcements illustrate a 24-hour escalation where both governments acted simultaneously to disrupt the remaining avenues for AI technology and capital exchange. The US is attempting to prevent its models from being used to train Chinese competitors, while China is seeking to prevent American capital— which comes with intangible advantages such as managerial expertise, talent networks, and strategic access—from reaching its AI national champions without state oversight. Each action is a response to the other, establishing conditions for further retaliation.

      The context for China's reported investment restrictions is the existing US outbound investment regulation effective from January 2, 2025, which prohibits US individuals from making equity investments in Chinese firms involved in advanced semiconductors, quantum computing, or certain AI technologies without the approval or notification of the Treasury Department. The proposed plan from China structurally mirrors this US regulation by requiring government consent before Chinese AI companies can accept capital from a country that has also been imposing chip export restrictions to China since 2022.

      The issue of model distillation represents a more technically novel aspect of the two initiatives. Chinese developers have utilized DeepSeek-R1, Meta’s open-source Llama models, and other accessible US models as training signals for their systems—a practice that is currently permitted under open-source licenses, yet US AI companies argue that it provides Chinese labs an unfair structural advantage. DeepSeek V4-Pro, which was just released and reported separately by TNW, was trained using Huawei chips and claimed near-frontier performance; whether it incorporated distillation from US models is a matter that the new administration framework would directly address.

      The specific enforcement mechanism for the crackdown on distillation has not been publicly defined, and the means by which a government would restrict the flow of training data across borders remains technically and legally unresolved. The commercial ramifications for Chinese AI startups are substantial but uncertain. Companies such as Moonshot AI, Zhipu AI, MiniMax, and the entity previously known as Manus AI are already navigating a capital environment constrained by US regulatory signals.

      If the approval requirement is enforced, it would introduce a formal layer of Chinese government oversight for any US VC investments in these companies, potentially discouraging further investment or steering more of China’s AI capital formation through domestic channels. The Chinese government has been enhancing state investment in AI infrastructure and has openly favored domestic AI leaders over those funded internationally. Establishing a formal approval process for US investments aligns with that preference.

      Neither of these measures, however, addresses the underlying dynamic that drives them: China's AI capabilities are progressing more rapidly than the export controls are impeding them. DeepSeek V4, released today, claims near-frontier performance in coding and mathematics using Huawei chips instead of Nvidia chips. The US controls on chip exports and investments were predicated on a growing capability gap; however, that gap is narrowing. The current question for both governments is not “how do we preserve the existing technological order,” but “how do we define the terms of a competition that is already fully engaged.”