The race for artificial intelligence is usually told through models, applications and the latest promises of automation. But beneath all that there is a less attractive layer and much more difficult to replace: the hardware. Without advanced chips, training massive models, deploying them at scale, and competing on the front lines becomes much more difficult. That is why NVIDIA continues to occupy such an important place for the United States and China. The question is no longer just who develops AI better, but who can access the most powerful chips first.
Huang’s message. That debate landed this week in Los Angeles, during the Milken Institute Global Conferencea forum that brought together bankers, investors, policymakers and executives in Beverly Hills. Over there, according to Nikkei AsiaJensen Huang was asked directly about a particularly sensitive question: whether China should have access to NVIDIA’s “latest and greatest” chips. His response was as brief as it was forceful: “No.” The CEO later added that the company supports the United States having “the first, most and best” chips, a phrase that pretty much sums up the balance it is trying to defend.
Sell yes, but not the last. Huang’s position is not about removing China from the trade equation, at least not according to what he proposed in that same forum. The CEO defended that American semiconductor companies continue to compete in global markets, including China, because that also strengthens the North American country. Huang said that increasing exports helps raise tax revenue, improve economic security and contribute to national security. The message, therefore, has two layers: technological leadership first, and commercial presence under control later.
The border is in the generation. Not all NVIDIA chips occupy the same place in this discussion. The H200, remember, is a high-end AI processor and places it above of the H20the chip the company designed for China after US export restrictions. But the agreement announced by Donald Trump in December did not include neither Blackwell nor the next-generation Rubin products, two families that represent a more advanced layer of NVIDIA’s roadmap.
The regulatory framework still has several moving parts. Donald Trump said in December that he would allow NVIDIA’s H200s to be sold to “approved” customers in China, as long as the US Government received 25% of those income. The company obtained official export authorization this year, and Huang said in March that NVIDIA had already received orders from “many Chinese customers.” But that does not mean that everything is resolved: the final shipment will also depend on whether Beijing allows such sales and in what quantities.
The bottleneck is not just political.Tom’s Hardware suggests that there may also be an industrial explanation behind the absence of recent shipments. According to the media, Hopper and Blackwell are manufactured in the same factories and production lines compatible with TSMC’s N4/N5, a capacity that is not infinite. If that reading is correct, NVIDIA would have reason to reserve more production for Blackwell, a more advanced and expensive family, especially for US customers, instead of using part of that capacity in H200 for China with a 25% commission to the US Government. According to that reading, Rubin’s arrival at N3 could free up margin later.
A pending trip. Trump said he will visit Beijing this month and that trade issues could be on the table in his meeting with Xi Jinping. In that context, Huang’s words do not sound like a simple corporate reflection, but rather a way of marking a position in a debate that goes far beyond NVIDIA. The dispute not only revolves around which chip is more powerful, but also about who accesses it first, under what conditions and with what margin so as not to be left behind.
Images | NVIDIA
GIPHY App Key not set. Please check settings