The US had decided earlier this week to let NVIDIA sell H200s to China in exchange for a 25% cut, but there might have been more to the deal than making money alone.
US Commerce Secretary Howard Lutnick has revealed the strategic thinking behind the decision in a recent statement, explaining that the move was the result of extensive discussions between NVIDIA CEO Jensen Huang and President Donald Trump. The revelation sheds light on a calculated attempt to fragment China’s AI development ecosystem by keeping Chinese developers tied to American technology infrastructure.
According to Lutnick, the decision came after more than a dozen conversations between the two leaders. “Well, that was a deal between Jensen Huang from NVIDIA and the President of the United States. They spoke together. They talked about it a lot. I mean, I would say more than a dozen times,” Lutnick said.
The Commerce Secretary outlined the strategic rationale behind the decision: “The idea from the President’s perspective is that the technology, the American technology stack, he wants the developers in China to also be using the American technology stack. If you let all the developers in China, and this was Jensen’s argument, if you let all the developers in China just use Chinese technology, they’re going to get used to it, they’re going to get better at it, and they’re going to go faster.”
The key insight, Lutnick explained, came from Huang himself. “So let’s break them apart, let them use the American technology stack, and therefore you split their developers. That was Jensen Huang’s argument. The President decided that he was willing to go with that for certain parts of this technology, and that’s the business deal.”
Lutnick characterized the arrangement as a meeting of minds between two business leaders. “So it’s the great American technologist talking to the great businessman president and they reached what they thought was the right [deal].”
The implications of this strategy are significant. By allowing controlled access to advanced US hardware like the H200—while keeping the most cutting-edge technology like Blackwell and Rubin chips restricted—the US aims to create a dependency that could slow China’s progress toward technological self-sufficiency. The approach forces Chinese AI developers to work within two incompatible ecosystems: American hardware with its associated software stack, and domestic Chinese alternatives. This fragmentation could potentially slow development cycles, create training inefficiencies, and prevent the emergence of a unified Chinese AI infrastructure.
However, the strategy carries risks. Chinese AI companies have already demonstrated remarkable capability despite hardware restrictions, with open-source models like DeepSeek gaining significant traction and, according to recent reports, surpassing US models in user adoption. The move could be seen as a delicate balancing act: providing just enough access to maintain American technological influence while withholding enough to preserve a strategic advantage. Whether this approach will successfully slow China’s AI ambitions or merely fund them with American technology remains to be seen. What’s clear is that the AI race between these superpowers is being fought not just with chips and models, but with carefully calculated strategic decisions about technology access and developer ecosystems.