Late on Monday night, July 14, 2025, the ninth richest man in the world broke some momentous news: The US government would allow him, Nvidia CEO Jensen Huang, to sell H20 processors to Chinese customers again.

To people following the Trump administration and its seemingly unending announcements and reversals of trade restrictions, this might not sound like much of a blockbuster. So President Donald Trump wanted to ban some trade and a CEO got him to change his mind (perhaps with some help from Chinese government promises on rare earth metals). What could be more classically Trump?

But there’s more going on here than just that. Huang was effectively announcing that a massive American effort, going on for nearly three years, to deprive China of the chips needed to build advanced AI systems, is over, or at best on pause. The high-tech chip export control regime built up and expanded by the Biden administration, and enforced largely intact by Trump’s team, now has a loophole in it large enough to drive a self-driving truck through.

The H20 chip only exists at all as part of Nvidia’s efforts to get around those export controls by designing something weaker than its flagship H100 chip, which has become an indispensable tool for training cutting-edge AI models and which the US still definitely doesn’t want Chinese firms like DeepSeek or Tencent to access.

That might seem like a reasonable compromise — let China have the weaker chip while holding back on the powerful one. The problem is that while the H20 is definitely worse than the H100 for some important tasks, for others it’s actually more powerful than its big brother, and comes at half the price. In April, enforcers at the Bureau of Industry and Standards in DC effectively figured out what Nvidia was trying to do with the H20 and informed the company that they couldn’t export the chip without a license, meaning in effect they couldn’t send the chips to China at all.

But Nvidia now says that Trump, having met personally with Huang, is promising to issue those licenses, which would enable Chinese AI companies to greatly accelerate model development and infrastructure buildout. And even though the H100 chips are officially still supposed to be off-limits, China may be able to get its hands on them as well.

About a week before Huang’s big news, reporters at Bloomberg broke the story of a massive data center construction project in Yiwu County in the Gobi Desert, encompassing multiple firms requiring over 115,000 high-end chips for training AIs from Nvidia. The project documents Bloomberg reviewed gave no indication of how the project would get those chips legally, the implication being that they wouldn’t — they’d illegally smuggle them instead. Now, thanks to Trump, they will likely be able to get the chips they need legally.

The Yiwu project and the H20 flipflop are both graphic illustrations of just how important advanced chips — often called simply “compute” by AI experts — have become, not just to AI development but to geopolitics writ large. They are the kind of things that the president of the United States takes meetings with private industry about.

That’s because compute is a very odd commodity. Advanced chips are overwhelmingly designed by one company (Nvidia), and manufactured by one company (Taiwan’s TSMC), using machines built by one company (the Netherlands’ ASML). The world can only produce so many, building out the fabrication plants (“fabs”) that produce them takes years, and past efforts to break those three companies’ dominance have almost always ended in failure.

The result is a scarce commodity that is unbelievably valuable in training the kind of advanced AI models that the US, China, and other leading powers view as indispensable for their economic and military futures.

If you’re not an AI nerd or a US-China watcher, this may all seem rather technical. But, to paraphrase Trotsky, even if you’re not interested in compute, compute is interested in you.

The chip war to date

In many historical eras, there is a commodity collected and traded in vast quantities that recent technological progress has made invaluable, which comes to........

© Vox