All news

US blacklisting of Chinese AI firms makes end use a valuation risk

A row of small dark chips on pale stone stops at a glass partition, beyond which stands a graphite-blue dome camera with a cobalt lens

Washington cut eight Chinese AI and surveillance companies off from US technology over abuses in Xinjiang. Who uses a model now shapes what it is worth.

US suppliers now need a license to ship any item subject to US export rules to eight companies at the center of China's AI industry, and for most items the government will presume to deny it. The US Commerce Department added them to its Entity List on 7 October among 28 Chinese organizations cited for their role in human rights abuses against Uighurs and other Muslim minorities in Xinjiang, and the rule took effect on 9 October.

Washington is now using export controls against AI companies because of how their technology is used, not only because of what it is. That draws a new line through AI valuations. The end use of a model and the customer behind it carry a political risk that can cut a company off from its suppliers overnight, and that risk has to be priced before a round, not after.

The list reads like a directory of the sector. Twenty entries are public security bodies in the region. The eight companies are the computer vision start-ups SenseTime, Megvii and Yitu, the speech recognition company iFlytek, the camera makers Hikvision and Dahua, and two smaller firms, Xiamen Meiya Pico and Yixin Science and Technology. They train and run their models on processors that come largely from US chip makers, and several of them raise money from international investors and plan public listings, as Megvii's Hong Kong filing in August showed. The license requirement therefore reaches both the hardware under their models and their path to public markets.

Government security work was a growth engine for Chinese computer vision companies. It is now a source of sanctions risk for them and a warning for everyone else. Founders who sell surveillance, biometric or predictive tools to public agencies anywhere should expect investors to ask who the end users are, what the contracts allow and what the company does when a customer misuses its product. Clear use policies, audit trails and a customer base that does not hinge on one state buyer are becoming part of a fundable company, and Fanaura raises them early in its talks with founders who sell to governments.

Chip substitution is the test that could blunt the measure. If the listed companies replace US processors with domestic or other foreign chips without losing ground, the license costs them a supplier rather than a business. Capital markets give the second reading: how far the listing slows the planned share offerings and fundraising of the Chinese start-ups named, and whether international investors step back. Europe and other allies are the third variable, depending on whether they add their own limits on exports of surveillance technology.

AI now goes through the same geopolitical sorting as telecom equipment did earlier this year. Value will build in companies whose markets and supply chains hold up under export rules on every side, a filter at the core of our thesis on AI markets.

For founders

Building AI for public-sector buyers with use policies investors can audit? 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.