Nvidia’s China Problem: Tariff Mayhem Aggravates Matters

We can anticipate some events, and make contingency plans before their arrival, but surprises sometimes come to claim us. To what extent can you anticipate and plan for unpleasant surprises?

I suppose the answer turns on our powers of discernment and the nature of the surprises. We know, for example, that natural disasters — earthquakes, wildfires, and floods — can strike, but we don’t know when they’ll hit or how severe they will be.

Still, we known there are certain conditions and geographies that are prone to natural disasters, and people and organizations can plan accordingly to maintain a semblance of continuity in their personal and professional lives. You can’t completely escape nature’s wrath, nor can you accurately predict the timing of its disasters, but, with forethought and planning, you can mitigate the damage they leave behind.

What about the ravages of politics and geopolitics? In a democracy, the electorate ostensibly chooses its leaders at the ballot box. We know it’s not that simple. Campaign financing and feral lobbying are qualifiers that condition the quality and quantity of candidates that are placed on the ballots. The vagaries of different democratic systems complicate matters further. Even so, one would think that, even though the electorate might not necessarily get the government it wants or necessarily needs, it should be able to act with enough collective force to ensure that elected representatives don’t act entirely against the interests of those whom they ostensibly serve.

Again, I know it’s not that simple. We have more information at our disposal than at any time in human history, and yet many of us remain blissfully ignorant or oddly misinformed. Or perhaps, some of the time, we are disinformed. The dark arts of propaganda, like so many other deceptive practices, have also embraced the mantra of digital transformation.

Is this profusion of information and misinformation, which we consume readily and continually, a blessing or a curse? It’s probably a bit of both, depending on how well you are able to navigate the data torrent and cut through it with analytic discipline. Sometimes, however, you are left with a Hobson’s choice, which is not much of a choice at all. That’s when you realize that reform of the system is imperative.

Technology vendors, like people in their private lives, face their own unpleasant dilemmas, their own Hobson’s choices.

Between a Rock and a Hard Place

Just look at the conundrum confronting Nvidia, which is caught between the unforgiving rock of the Trump Administration’s impenetrable tariff-fueled trade war and the hard place of China’s increasing resolve to develop technological sovereignty in silicon, software, cloud services, and, of course, artificial intelligence.

Life in the fast lane is hard enough for any company in the technology industry, but maintaining exceptional prosperity becomes far more difficult when governments conspire against the best-laid strategic plans of corporate braintrusts. Even in the best of all possible worlds, Nvidia would have faced mounting competition, if only for the simple reason that surging market growth inevitably attracts competitors who want a piece of the action. But what’s happening now — amid increasingly inflamed economic, industrial, and political tensions and a veritable trade war between the U.S. and China — makes the best of all possible worlds seem as remote as far-off galaxies

Even before the pugnacious Trumpian recrudescence, Nvidia had reason to believe that its glory days, if not over, were nearly in the rearview mirror. The cloud giants, never inclined to supplication or reliance on a single supplier, were bridling at Nvidia’s costly GPUs and market dominance, so much so that have chosen to develop their own AI accelerators. Meanwhile, in China, where U.S. technology embargoes were already in effect before Trump’s return to the White House, the Chinese government was directing and encouraging domestic chip production, sensing that the infrastructure spigot from the U.S. might be turned off or slowed to a trickle at any moment. If knowledge is power, then having sovereign command of your own supply of advanced IT infrastructure is a prerequisite to digital knowledge.

So, the portents of trouble — a market squeeze on multiple fronts — were already apparent to Nvidia before Trump, as is his wont, took a combustible set of circumstances and ensured that they became a raging conflagration. Nvidia is now running from the fire, tryign to salvage whatever it can before the flames consume the best furniture.

Recent events are summarized in this report from the Wall Street Journal:

Nvidia said that the U.S. government on Monday informed it that it will implement a license requirement to export its AI H20 chips to China and other countries, that will be in effect for the indefinite future.
“The USG [U.S. government] indicated that the license requirement addresses the risk that the covered products may be used in, or diverted to, a supercomputer in China,” Nvidia said Tuesday in a Securities and Exchange Commission filing.
As a result, Nvidia announced that it will take anticipating a charge of up to $5.5 billion tied to exporting its artificial-intelligence chips to China, according to a regulatory filing from the semiconductor company.
The graphic chip maker said Tuesday its first-quarter results for the three months ending April 27 are expected to include the charge associated with its AI H20 chips for inventory, purchase commitments and related reserves.

Damage Control and Reconnaissance

With that news echoing through an anxious market already traumatized by Trump’s capricious tariff blitzkrieg,, Nvidia CEO Jensen Huang flew (on an aircraft, of course) overseas for separate meetings with China’s vice-premier He Lifeng and DeepSeek founder Liang Wenfeng.

At least partly, Huang’s China mission can be classified as damage control (or damage limitation if you live in the UK). Although Nvidia’s revenue stream from sales of GPUs to China will inevitably dry up (I will refrain, as a humanitarian gesture, from invoking an off-color simile here), Huang has a an abiding interest in managing the decline so that is graceful rather than precipitous. The other part of his Chinese visit, especially the meeting with DeepSeek’s founder, is likely focused on knowledge gathering and intelligence, motivated by a desire to gain a deeper understanding of DeepSeek’s models and processing requirements.

The events of this week, however, is just the latest installment of narrative that was already unfolding. Even before the tariff mayhem, China was firmly committed to gradually reducing its reliance on and exposure to U.S. technology.

Some in China contend that advances in homegrown AI accelerators mean that chips from Huawei Technologies (Ascend series), Hygon Information Technologies, and Cambricon Technologies are capable of performance that is comparable to that of Nvidia’s market-leading GPUs.

Where China still needs to do remedial work, according to Li Guojie, of the Chinese Academy of Sciences, is in developing software that rivals Nvidia’s CUDA platform. Work is presumably occurring on that front. If Nvidia latest and greatest chips are permanently off limits as options for Chinese buyers, then necessity will again serve as the mother of invention, just as in the case of DeepSeek, whose algorithm optimizations are credited with enabling it to cost-effectively improve AI model efficiency and performance.

Questioning the Scaling Law

In an article first published in the South China Morning Post (SCMP), Li Guojie, citing the breakthroughs made by DeepSeek, questioned whether the “scaling law,” viewed as an axiom in Silicon Valley, is more supposition than reality. From the SCMP article:

"In AI, the scaling law is viewed by some as an axiom ... and companies like OpenAI and the US AI investment community have treated it like a winning formula," Li said. "But the scaling law is not a scientifically verified principle like Newton's laws; it is a generalisation based on the recent experiences of OpenAI and others in developing large language models.
"From a scientific research perspective, it is an educated guess; from an investment standpoint, it is a gamble on a specific technological pathway," Li said, noting the delay in the launch of the GPT-5 model could signal a decline in scaling.
Li added that DeepSeek's success illustrated an alternative approach to improving model performance, which was through algorithm optimisation.
"The emergence of DeepSeek has forced the AI community to seriously reconsider whether to keep burning money and gamble, or to seek a new way to optimise algorithms," he said. "DeepSeek's achievements suggest that algorithmic and model architecture optimisation can also lead to miracles."

I should note that Li isn’t dismissing the scaling law, but he is suggesting that the verdict on whether it qualifies as an ironclad law remains in question.

What is not in dispute is that China seeks to incrementally and then completely go its own way, blazing its own trail in GPUs and AI accelerators. This is not a new development, but the next stage in the process. Perhaps Trump didn’t start the fire, but he definitely threw some accelerant on the blaze.