Nvidia’s Chinese Misadventure Takes Another Sharp Turn

At the risk of dislocating my shoulder, I will attempt to pat myself on the back. Maybe it isn’t much of an accomplishment, but I correctly perceived that Nvidia’s retrograde H20 AI chip, designed exclusively for unfortunate Chinese buyers, was living on borrowed time.

What I got wrong was just how little time remained in the H20’s commercial lifespan. When the book is finally closed on Nvidia’s intentionally underpowered AI chip for Chinese customers, its longevity will be more like that of a bald-faced hornet (nasty, and actually a wasp) than that of a dog or a cat. The H20 is one of those artificial monstrosities that only bureaucrats could conceive: a torpid compromise between Nvidia’s wish to sell full-featured GPUs to Chinese buyers and the U.S. government’s determination not to allow China to have access to Nvidia’s best, highest-performance AI silicon. In these situations, nobody wins and acrimony is the only sure outcome.

As discussed here previously, everybody knows that the resulting H20 was an unsatisfactory kludge. Nvidia recognizes the H20 as a hollow compromise, China knows it’s getting shortchanged, and the U.S. government should know that the law of unintended consequences never sleeps. What will happen — it’s happening even as I type this message — is that China has decided that it doesn’t want a slow ride to a dead end on the H20.

Put another way, China doesn’t want a second-rate GPU that leaves it at a permanent performance disadvantage to customers in the U.S. and other Western nations. China is also fully cognizant that the U.S. government might decide to further downgrade the performance of technology that Nvidia and other U.S.-based customers supply to Chinese customers. You may have noticed, but the Trump Administration isn’t exactly a model of equanimity and consistency. What is here today might be gone tomorrow, replaced with something even worse than what you had before.

If you were the Chinese, what would you do in these circumstances? You’d probably do exactly what China is doing, which involves taking control of its own technological destiny by developing and supporting domestic alternatives to technologies currently sourced from U.S. companies such as Nvidia. You might not like this turn of events, but you can objectively appreciate the logic behind it.

If the Trump Administration were more discreet in its machinations, perhaps the situation wouldn’t have escalated so quickly and conclusively. Nvidia’s cause isn’t helped by members of the Trump Administration, who seem to take great joy from gratuitously antagonizing the Chinese.

"We Don't Sell Them Our Best Stuff"

According to various reports, including this one at Endgadget, Chinese regulators reportedly discouraged domestic companies from purchasing Nvidia’s H20 chips because of “insulting” commentary from U.S. commerce secretary Howard Lutnick. What, you may ask, could Lutnick have said that caused the Chinese to take offense? The Endgadget article provides the following detail:

The US, if you'll recall, blocked NVIDIA from selling its H20 chips to China back in April out of concern that the Chinese military would use them to develop AI technology. When the US government reversed its decision in July and allowed the company to start shipping its chips to China, Lutnick told CNBC : "We don't sell them our best stuff, not our second best stuff, not even our third best. The fourth one down, we want to keep China using it... The idea is the Chinese are more than capable of building their own. You want to keep one step ahead of what they can build, so they keep buying our chips. You want to sell the Chinese enough that their developers get addicted to the American technology stack. That's the thinking." To note, a previous Times report stated that the government allowed NVIDIA to ship its products to China again after agreeing to hand over 15 percent of its profits.

Nobody can accuse Lutnick of keeping his cards close to his vest. He’s put everything on the table, and the Chinese, who already knew they were getting shortchanged in their procurement of U.S. technology, are now subjected to the public embarrassment of Lutnick’s boasting about the ruse. Henceforth, Lutnick’s quote should serve a textbook example for the phrase “rubbing salt in the wound.” One thing’s for sure: Nobody wants to be an addict, whether in thrall to hard drugs or AI chips.

Hope, however, apparently springs eternal. Nvidia CEO Jensen Huang is pitching a new chip for China, a successor to the H20. That chip, the B30A, is derived from Nvidia's Blackwell technology, and reportedly performs at about half the speed of Nvidia's main B300 chips. Huang concedes that the decision to make the B30A available to Chinese buyers is the U.S. government’s, not his, and you can sense some tacit frustration in that acknowledgment.

On the one hand, Huang feels compelled to offer conciliatory praise to the Trump Administration for nominally bolstering the U.S. technology industry. Conversely, he also knows that the government’s decisions are draining the lifeblood from some of his key export markets, with China at the front of the queue.

It’s a tough bind for Huang, and I don’t think the B30A extricates him from it. I suppose the purpose of the chip, now that China has firmly closed the door on the H20, Huang and Nvidia need to produce something more compelling. The B30A is better than the H20, but the geopolitical narrative will continue to work against Nvidia in China. China might get a slightly larger stack of pancakes, but U.S. customers will get a higher stack and better syrup.

Inevitable Parting of the Ways

The pursuit of the B30A comes on the heels of Nvidia’s decision to instruct its component suppliers to suspend production linked to the H20. As Chinese demand evaporates, Nvidia has no need to continue stoking the supply chain. The H20 is practically dead as a commercial offering, so Nvidia is moving aspirational toward the B30A. Still, regardless of whether the U.S. government gives Nokia a green, amber, or red light on the B30A, Huang and his executive team know it will amount to little more than a stopgap measure. China will continue its gradual, not-so-furtive move away from Nvidia and other U.S.-sourced technology.

Perhaps the breakup was inevitable, inescapable, but that doesn’t make it any less disagreeable for Nvidia and other vendors caught in the geopolitical crossfire. For the mutually reinforcing reasons of technological self-sufficiency and practical necessity, China will blaze its own trail in AI infrastructure, including essential GPUs and AI accelerators for training and inference. While Huawei’s Ascend chips get the most media attention, the AI chip ecosystem in China also includes Alibaba, Cambricon, Biren Technology, Moore Threads (which sounds more like a haberdashery), Enflame Technology, and Lisuan (Leon Tech). I’m sure more other vendors will join the party.

Are the chips these vendors produce currently on par with the best that Nvidia offers? By all accounts, or at least by consensus, the answer is no. Nonetheless, China’s AI-chip ecosystem is constantly iterating and improving, and the imperative of self-sufficiency means that the Chinese government will continue to push these companies and support their technological advances.

I’m not surprised in the least that the H20 is on its deathbed. I suspect that Nvidia’s mooted H30A will be congenitally compromised and doomed to a similarly short existence.

None of his is Nvidia’s fault. No matter how much you plan and how well you execute as a company, exogenous factors can occasionally leap from the shadows and give you a mugging.