All news

US outbound rules make cap tables part of national security

Closed graphite steel gate in a long warm-white wall with a small cobalt padlock

From January, US investors cannot back Chinese labs training large AI models. Training compute is now a legal threshold, and diligence on it a standard step.

A training run of more than 10²⁵ computational operations is now a line US money may not cross. The final rule for the US Treasury's outbound investment program, issued on 28 October and in force from 2 January 2025, bars US persons from investing in Chinese companies, including those in Hong Kong and Macau, that develop AI systems above that threshold or built exclusively for military, intelligence or mass-surveillance use.

National security review has moved from the chip to the cap table. Export controls on advanced chips already govern what can be shipped to China. Now the same logic applies to capital, and the compute used to train a model becomes a legal threshold for investors, as it did for labs under last year's AI order. Any fund that counts as a US person must screen its targets and co-investments for Chinese AI exposure.

The rule reaches well below the frontier. Deals with companies training above 10²³ operations, or building AI for cybersecurity, digital forensics or robot control, must be reported to Treasury within 30 days, and semiconductors and quantum technologies are covered too. Publicly traded securities are exempt, as are limited partner commitments of $2 million or less in a fund, so larger limited partners will need contractual assurances to stay outside the rule. Violations carry civil and criminal penalties under US sanctions law. Global investors with US ties will price that compliance into every cross-border deal, and Chinese AI startups will rely more on domestic capital.

The diligence lands on founders. Anyone raising from US investors should expect sharper questions about where the team sits, where subsidiaries are registered and who the training partners are. A development team or joint venture in China can trigger a notification or a prohibition for the investor, and that can stall a round, so the exposure is worth mapping before fundraising, with any Chinese operations clearly structured and documented. Companies building in the US, the EU and Latin America, the three regions in Fanaura's mandate, gain on the other side of the ledger: some investors will now stay away from Chinese AI deals altogether, which narrows the competition for US capital.

How durable all this is remains open. The administration that takes office in January decides how it treats the program, and Congress can write it into law with a wider scope. Treasury guidance and the first enforcement cases will show how investors are expected to verify the training compute of a target, and the test beyond Washington is whether the EU, the UK and Japan adopt similar screens on outbound capital.

In a structure like Fanaura's, a separate SPV for each deal holding its own and LP capital, geopolitics now shapes an AI cap table as much as valuation does. Diligence on jurisdiction, data and training compute is becoming a standard part of every AI round, and a cap table that passes it is now part of what a founder sells.

For founders

Building AI in the US, the EU or Latin America with a clean cap table? Send us the project.

Fanaura invests its own capital and the capital of its limited partners (LPs), usually through a separate SPV for each deal. If we see a fit, we get in touch.

Apply in five minutes
Subscribe

Get the news by email

One short note each weekday. No marketing, and you can unsubscribe at any time.