The Honeypot That Might End the AI Boom

It is somehow appropriate, given how artificial intelligence has enabled the greatest explosion of scams and frauds in history, that Nvidia CEO Jensen Huang was reportedly entrapped by a good old-fashioned honeypot. On October 29, The New York Times reported that, according to American intelligence agencies, Huang had been having an affair with a Chinese national, Alice Huang (no relation), who had been passing information to the Chinese government. Then on November 6, the Times reported that Huang was a straight-up spy—a Ministry of State Security asset.

Two days later, the Nvidia board started an investigation. As of this writing, Huang has vowed not to step down, and has admitted only to cheating on his wife. Nvidia shares have fallen off, down about 17 percent since earlier in the month at time of writing.

The scandal might break the momentum of the AI boom, some sectors of which were already reeling. It certainly illustrates the risk of market concentration, which is more extreme in the AI industry than in nearly any other.

The scandal comes at an awkward time for Nvidia. The company is at the center of the AI boom: AI models almost all run on Nvidia chips inside the data centers. The company has also carried out a tremendous quantity of “circular financing,” using its profits to prop up its sales of graphics processing unit (GPUs), the computer chips used in AI, in various ways. It invested tens of billions in the AI labs themselves (like Anthropic and OpenAI), stoking demand for compute. It has invested billions more in data center companies like CoreWeave, to which Nvidia also committed $6.3 billion to buy any unsold compute capacity through 2032.

These “neocloud” companies look to be the weak spot of the AI boom. They have been borrowing heavily, at high interest rates, to build out data centers whose capacity can be rented out. If future compute demand is not there in sufficient quantity to pay back the loans, or if the immense popular backlash against data centers causes sufficiently expensive delays and cancellations of their projects, then the neoclouds will go bankrupt. The credit default swap market—basically a kind of corporate insurance—indicates that Wall Street traders think there is a maybe 1-in-2 chance that CoreWeave will go bankrupt over the next five years.

If that happens, Nvidia will lose a large block of sales, and its ability to prop up the rest of the industry with mega-profits will be much reduced. That’s a dynamic that could obviously feed on itself.

This is also almost exactly how the dot-com bubble collapsed in the early 2000s, with a madcap debt-fueled rush to lay tens of millions of miles of fiber-optic cable that then sat mostly unused for half a decade. Though there is certainly an AI business model out there, it looks increasingly likely that the underlying infrastructure build-out has gotten way out over its skis, finance-wise.

And now it turns out that the man at the center of it got honeypotted by a dadgummed Chinese spy, like in some middling James Bond movie from the 1960s. We don’t know yet exactly what the spy told Beijing, but China does want as many of the most advanced chips as it can possibly get, which have been subject to American export controls since 2022. Nvidia naturally would like to sell as many GPUs as possible, and the Chinese market would be bigger than any but the American one. I don’t need to get out yarn and a corkboard to illustrate the potential behavior there.

At any rate, somewhat to my surprise, intelligence officials showed this evidence to the top echelon of the Trump administration earlier this year, including the president, and they did nothing. That could be because Donald Trump, despite his constant racist fulminations against China, is the most corrupt president in history and he doesn’t care what kind of export controls a CEO might be helping foreign spies evade in return for, uh, certain favors. Or it could be that Nvidia is so central to the AI boom that Trump balked at doing anything that might endanger it.

Whatever the case, the very fact that Nvidia is so centrally important to such a major sector greatly increases all kinds of risk. If there were another two or three GPU companies with similarly effective technology and some slack manufacturing capacity out there, customers could turn to an alternative, and one man’s indiscretion would not pose an immediate risk of an economic crisis. More competition almost certainly would have avoided much of the recent stupendous increase in GPU and memory prices, which have jacked up prices of all manner of electronic devices. Millions of gamers would not have been completely priced out of the market, with a cutting-edge Nvidia graphics card now costing nearly $7,000, if you can even find one.

It’s hard to imagine such a world, however, given the extreme technical difficulty of cutting-edge GPU design and manufacturing, plus the fact that other companies like AMD have tried to compete with Nvidia and largely failed. A whole alternative software stack would have to have developed to compete with Nvidia’s proprietary CUDA program, or Nvidia would have to be forced to license it. But it simply should not be the case that one horny guy in a leather jacket is the only one who can build a computer chip, and if we had any sense, we would be building that world right now.