The Two-Way Street of Chips and Talent
A strange few weeks in the US-China AI race.
Last Wednesday, January 14th, the Trump administration formally approved exports of Nvidia’s H200 chips to China. The H200 is Nvidia’s second-most-powerful AI processor. Not the bleeding-edge Blackwell or the upcoming Rubin, but serious hardware nonetheless. Chinese companies have already placed orders for more than two million units.
The decision came with conditions: a 50% volume cap relative to US domestic sales, third-party testing, no military use, and a 25% tariff that the White House frames as the US “getting a piece of the sale.” The Commerce Department calls it a “thoughtful balance.” Critics call it a capitulation. The truth is probably somewhere in between.
Either way, it’s a stark reversal. The Biden administration spent its final year tightening the screws on chip exports to China, arguing that advanced semiconductors were too strategically sensitive to let flow freely. The H200 was explicitly restricted under those rules. Now it’s not.
The official logic, championed by White House AI czar David Sacks, goes like this: controlled sales to China discourage Huawei and other domestic players from accelerating their own chip development. If Chinese firms can buy American, they won’t invest as heavily in building Chinese alternatives.
Whether this logic holds is an open question. China has historically used access to foreign technology as an input to domestic development, not a substitute for it. The two approaches tend to run in parallel: acquire Western products, study them, and build local alternatives, often with state support. Access and indigenous development aren’t either/or. They can be sequential.
Two million H200s will train a lot of models. They may also inform the next generation of domestic Chinese chip designs. Or they may simply meet commercial demand without much strategic spillover. It depends on factors that are difficult to assess from the outside.
Three weeks before the chip decision, a different kind of transfer occurred.
On December 29th, Meta announced it was acquiring Manus, a Singapore-based AI startup, for somewhere north of $2 billion. The deal closed in roughly ten days. Fast, even by tech standards.
Manus had burst onto the scene last March with demos of AI agents that could actually do things: screen resumes, research stocks, book travel, build prototypes. Not chatbot tricks, but autonomous task completion. The company hit $100 million in annual recurring revenue within eight months of launch. By December, it was one of the most impressive AI agent platforms in the world.
It was also, originally, Chinese.
Manus was founded in China in 2024 by the team behind Monica.im, a popular AI tool. It raised money from Tencent, ZhenFund, and other Chinese investors. But as US-China tensions escalated, the company relocated its headquarters to Singapore, laid off its Beijing staff, and made its product unavailable in China. By the time Meta came calling, Manus had scrubbed most of its Chinese roots. Enough to satisfy US regulators, anyway.
The acquisition included an explicit condition: “no continuing Chinese ownership interests.”
Beijing noticed. Ten days ago, on January 9th, China’s Ministry of Commerce announced it would investigate the deal for potential export control violations. The investigation is ongoing, and its outcome is uncertain. But the message was clear: China is now asserting that AI talent and intellectual property developed on its soil are strategic assets, not freely exportable goods.
It’s a familiar argument. The US has made similar claims about semiconductors for years. Now both sides are reaching for the same tools.
Zoom out, and the two stories form a strange picture.
In the span of three weeks, the US approved a massive flow of advanced chips into China while facilitating a major flow of AI talent and capability out of China. One move puts compute in Beijing’s hands. The other puts agentic AI expertise in Silicon Valley’s.
Both are framed as wins for the US. Nvidia gets access to a market worth potentially $50 billion a year. Meta gets technology it couldn’t build fast enough internally. From Beijing’s perspective, the picture looks different: chips are coming in (good for near-term AI development), but talent and companies are leaving (bad for long-term ecosystem building). Hence the investigation.
What makes this interesting is that neither side has a fully coherent position.