Meta AI chief urges US startups to prioritize government work over Chinese data deals

In a response to a post shared by Scale AI’s VP Aakash Sabharwal, Meta AI head Alexandr Wang said that providing services to the United States government should be a foundational principle for American startups rather than a commercial afterthought. Wang’s remarks follow one outlet's report revealing that data-labeling startups from Silicon Valley have been approached by Chinese AI companies, including Tencent, Ant Group, Alibaba and ByteDance, at the recent International Conference on Machine Learning in Seoul.

The investigation found that firms such as Surge AI, Mercor, AfterQuery and Turing, which also serve OpenAI, Anthropic and U.S. government customers, have supplied training data to those Chinese buyers, with the six largest Chinese labs collectively spending roughly $500 million a year. Scale AI’s Sabharwal said his company has turned down such revenue to avoid undermining U.S. AI leadership and national security. Meanwhile, AI czar David Sacks warned that China’s newly released Kimi K3 model, open-sourced by Moonshot AI, is performing at the frontier on coding benchmarks, raising concerns about U.S. innovation lagging behind. He argued that excessive regulation and bureaucratic hurdles could cause America to lose its AI edge.