Meta AI head Alexandr Wang told American startups that serving the US government should be a "bedrock principle," not a commercial convenience. His warning follows a Forbes investigation showing Chinese AI labs spend about $500 million a year with Silicon Valley data companies - including vendors that also work with US agencies.
"Serving the US government should not be a commercial convenience, it must be a bedrock principle for startups," Wang said in response to a post from Aakash Sabharwal, VP of engineering at Scale AI.
Sabharwal warned that American companies should not sell data to Chinese AI labs. "American companies should not be selling data to Chinese AI labs. @scale_AI doesn't do this work, and we have turned down revenue because of it. The companies that do are undermining American AI leadership and risking national security," he wrote.
The data pipeline to China
The Forbes report found that sales teams from Silicon Valley data-labeling startups met eager Chinese buyers at this year's International Conference on Machine Learning in Seoul. Tencent was among the companies circulating detailed requests for training data covering finance, cybersecurity, and self-improving AI systems.
The same American vendors that serve OpenAI, Anthropic, and in some cases the US government - including Surge AI, Mercor, AfterQuery, and Turing - have also supplied Chinese buyers such as Tencent, Ant Group, Alibaba, and ByteDance. Citing data-labeling entrepreneurs, Forbes estimated that the top six Chinese AI labs collectively spend around $500 million a year with American data-labeling companies.
The US has restricted China's access to advanced AI chips but has not placed similar limits on training data or the human expertise that shapes how AI models learn. That gap is the pressure point Wang and Sabharwal are pointing to.
Sacks warns on regulation
Separately, AI Czar David Sacks called China's new Kimi K3 model "concerning" for American technological dominance. "This is concerning. For the first time, a Chinese model Kimi K3 has taken #1 in the Frontend Code Arena and is scoring at or near the frontier on other benchmarks," Sacks said on X.
Beijing-backed startup Moonshot AI open-sourced a 2.5-trillion-parameter system that is now outranking top American models on global coding benchmarks. Sacks argued that US regulation is the bigger threat. "Meanwhile America is tying itself in knots: politicians and bureaucrats are banning new data centers, piling on state regulations, and pushing for new federal agencies to pre-approve frontier models," he said, pointing to the government's decision to restrict Anthropic's Fable 5 model to non-Americans.
"This is how you lose the AI race. The rest of the world won't play by our rules if we bog ourselves down. Permissionless innovation is how America won the internet and became the technological envy of the world. We can do it again with AI -- while addressing risks in a targeted way -- or we'll watch our lead evaporate," Sacks added.
Why this matters for government professionals
The stakes are concrete for federal employees. The same data-labeling vendors that train models for US agencies may also be training models for Chinese labs, and current export controls don't cover data the way they cover chips. That means procurement and security teams need to ask vendors directly about their customer lists and data-handling practices.
Government professionals who oversee AI contracts should understand the difference between chip restrictions and data restrictions - and push for the same scrutiny on data that chips already receive. For those building AI policy or managing vendor relationships, courses like AI for Government and AI for Policy Makers can help clarify how these risks map to procurement and governance decisions.
Your membership also unlocks:
