Trump’s China Reversal Gives Nvidia a Hollow Victory

The chip has sailed

Too much has been made of the Trump Administration’s tortuous reversal that allows Nvidia to sell a lesser species of its GPUs to China. Fundamentally, this particular turnabout doesn’t change the salient narrative involving the U.S., China, and Nvidia.

While Nvidia will squeeze some relatively modest incremental revenue from sales of its H200 GPU in China, there’s not much else to see here, folks. Some critics of Trump’s latest volte-face are saying the move portends disaster, that “China could harness the technology to supercharge its military.” That’s no need to worry about that, for reasons we’ll explore shortly.

More noteworthy is that Trump intends to take a 25% cut from sales of the H200. When did extortion become a modest operandi of the U.S. government’s stewardship of trade and its relationships with homegrown technology companies? Imagine a throaty, baleful rumble: ‘It would be a real shame if something happened to your export markets. But maybe we can make a deal.”

How embarrassing for everyone involved. Okay, perhaps not for Trump, who seems beyond mortification or shame, but for everybody else. By all accounts, Jensen Huang is an extremely intelligent man. He must know that all this genuflection and sycophancy toward the White House capo demeans everybody involved. To make matters worse, the questionable dividends that ensue from the compromise are not worth the melted candle and burned fingers that accompany them.

I want to state the following as unambiguously as possible: China is not about to buy the H200 in volumes large enough to deploy in pursuit of military dominance. They will buy a few, and everybody will pretend it’s business as usual. Nonetheless, the smart money knows that China isn’t about to become dependent on any U.S. technology. That chip has sailed.

Besides, the H200 is based on the Hopper architecture, which means it is one generation behind the current Blackwell architecture, which will soon be succeeded by the Rubin architecture, scheduled to reach market in the first half of 2026. China is not getting first run at the GPU buffet.

It’s Not the Best, But It’s What You Get

Even before Howard Lutnick said the quiet Machiavellian part out loud, that the U.S. goal was to leave China behind in the AI race by fobbing it off with fourth-rate AI silicon, the game was over. Admittedly, the H200 is not as inferior as the fare Lutnick might have preferred to ship to China, but it’s not the latest and greatest silicon Nvidia offers.

China knows the U.S. will never sell it the same grade of foundational technology that U.S. companies and organizations are permitted to buy and use. For that reason, China will develop its own silicon and essential technology infrastructure.

China won’t be able to catch up overnight, and the effort will require prodigious investments and assiduous effort, but if it buys castoff and end-of-life products from American companies, it will never catch up. It would be like starting a race 50 meters behind your competitor and then being forced to run on a treadmill.

The Chinese are smart enough to figure this out, folks. If you’re in dire competition in any discipline or realm, you can’t allow your opponent to establish ground rules that put you at a permanent, insurmountable disadvantage. There is no possibility of your ever gaining ground. You’ll always be behind, always at the mercy of your foreign technology supplier and its government.

I’m not gifting you with pearls of wisdom, by the way. This is basic reasoning. But I’m still grappling with the disconcerting realization that we’ve entered a period of willful irrationality and stupefying unreason. In the land of the dolt, the middling mind is the disheveled king.

Jensen Huang and the Nvidia braintrust know what time it is. Huang knows that Chinese will buy a limited number of H200s, not because the Chinese are keen to become permanently dependent on them, but for two reasons: First, the H200s might serve as useful purpose as a stopgap contingency, to get from point A to point B in a sequence that goes all the way to point Z; second, that buying the H200s represents a tactical head fake, a furtive diversion that gives China time to refine and improve its homegrown AI chips while lulling the Trump Administration into complacency.

A benefit for Huang and Nvidia is that the reprieve from China trade buys time for Nvidia to devise its own post-China revenue plan. Everybody benefits, but not for long. The trust is gone, presuming it ever existed, and an adversarial realpolitik rules.

Subscribe to Crepuscular Circus

Don’t miss out on the latest issues. Sign up now to get access to the library of members-only issues.
jamie@example.com
Subscribe