The Seller Also Lobbied the Referee
Jensen Huang argued for years that chipmakers should serve the world and let the rules decide. Deferring to the rules is honest only if you didn't help write them.
Chips with the computing power of roughly 1.6 million Nvidia H100s reached China in violation of US export restrictions by the end of 2025, according to an estimate by the research group Epoch AI cited by Al Jazeera this week. Much of the smuggling, the same report says, ran through Southeast Asia. Nobody at Nvidia has to have wanted that for it to have happened, and that is precisely what makes the number interesting.
On Wednesday Nvidia's board accepted Jensen Huang's resignation as chief executive, according to the company's filing, and on Thursday the company announced it, with the chief financial officer, Colette Kress, as interim CEO. The surrounding reports concern his private life and a former executive at an Nvidia customer whom US officials reportedly suspect of working for Chinese intelligence, which China's foreign ministry calls a smear. That is for investigators. What interests me is an argument Huang made in public for years, one this week has put on trial without settling.
The argument is about who decides where a technology goes. At one end sits Dario Amodei, who has compared selling advanced chips to China to "selling nuclear weapons to North Korea." Huang's reply, on Dwarkesh Patel's podcast in April, was that the comparison was "lunacy." "We're not enriched uranium," he said. "It's a chip, and it's a chip that they can make themselves." In a CBS interview in September he put the principle plainly: "Every single chip company should go and serve the world, compete for the world." Semafor, in an analysis of 42 of his podcast interviews published this week, counted the phrase "American tech stack" 34 times in eight of them. He wanted the world to build on what he made.
His position was always better than "sell to anyone who pays." He never claimed makers should ignore the law; he said they should compete within it. And there is a real danger in the opposite view, which is that the maker gets to decide. Nvidia reportedly controls an estimated 80 to 90 percent of the market for the chips that run frontier AI. A company with that share deciding on its own which countries deserve computing power would be running a private foreign policy that no voter anywhere chose. Huang often pointed out that about half the world's AI researchers are Chinese or work in China. A graduate student in Hangzhou didn't vote for anything either, and a CEO's distaste for her government is a poor reason to decide what she gets to work with. From anywhere other than Washington, the idea that a few founders in California should sort the world into the trusted and the untrusted looks less like responsibility and more like reach.
A maker owns some share of every use of what he makes: that was the sentence this essay was supposed to defend, and the graduate student in Hangzhou is enough to break it. The maker should not be the judge, and that a world where the people who build a technology also ration it by their own politics is worse than the alternative. Somebody accountable to the public should draw the lines, and the maker should live inside them.
But that is a description of a referee, and Huang was not standing on the sideline. According to CNBC, one analyst this week called him "the industry's most effective lobbyist against tighter rules." He told Bloomberg in October that restricting sales to China had been "a mistake," and that he had tried to persuade policymakers otherwise. That is entirely legal, and arguably his job. It also changes the moral shape of "let the rules decide." If you hand the decision to someone else, the outcome is theirs. If you spend years working to make that someone else decide your way, a good part of the outcome is yours again. Deferring to the rules is honest only if you didn't help write them.
The same logic reaches the 1.6 million chips. Those went around the rules, not through them, and no maker can inspect every shipment that passes through a distributor in Southeast Asia. Still, a maker who spent years arguing that the restrictions don't work has little standing to be surprised when they leak. How much of a misuse belongs to the maker depends partly on how hard the maker chose to look, and an argument that the looking is futile is a way of choosing not to.
Neither of these points requires siding with Amodei. Notice that his position, for all its alarm, has the same structure as Huang's in reverse: one maker asking the referee to close a market, another asking it to stay open, both claiming to speak for the national interest, both running companies with something riding on the answer. That makes neither of them wrong. It does mean "the market should decide" was never a neutral description. Markets decide inside rules, and in this one the people who make the most important product were among the loudest voices shaping the rules.
The question reaches well beyond chips. An engineer who argued that a review step was bureaucratic, and then watched her code ship without it, can say accurately that she didn't make the final call. So can the consultant who helped draft the regulation his clients now follow. Both leaned on the scale. A share of what happens next is theirs, roughly in proportion to how hard they leaned.
On Tuesday, Senate Democratic leader Chuck Schumer called on Congress to pass chip bills. Other lawmakers are pushing their own versions, and the Commerce Department is reportedly finalizing a rule on remote access to chips through overseas data centers. The man an analyst called the industry's most effective lobbyist against them will not be in those rooms. The next few months should show how much of what the market used to decide was really him.
— The Andes