...
By Shane Croucher
Senior Editor
0
Share
Newsweek is a Trust Project member
See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.
There is still much we don’t know about the China drama engulfing Nvidia right now.
But it points to a conclusion about the bigger picture story of the U.S.-China race on artificial intelligence, even as the doomers cry catastrophe: it’s time for America to accelerate.
This week, the world-leading chipmaker’s charismatic founder and CEO Jensen Huang, the heart and brain of its incredible three-decade rise, resigned amid reports of at least three extramarital affairs, two of them allegedly with Nvidia staff.
Read More on Analysis
Alice Huang, the former chief executive of the Singapore-based data-center operator Megaspeed, no relation to Jensen, is said to be the third. Jensen Huang admitted in general terms to being unfaithful, though he has not confirmed any specific details or affairs.
A U.S. intelligence memo, in circulation for some time and reported by The New York Times, has suggested that Alice Huang may have been recruited as a Chinese spy. The Trump administration, which is investigating, has not yet reached that conclusion.
Beijing is calling the claims a smear. Nvidia's board, which commissioned an independent investigation, said it has "no reason to believe that either national security or NVIDIA secrets were compromised."
Whatever the truth turned up by the various investigations, one of which is being conducted by the FBI, the story has an export-control spine running right through it.
Related Story
The AI Revolution Meets America’s Measles Regression
Megaspeed rents Nvidia-powered computing to Chinese customers and has been investigated over whether its chips reached China, which it denies.
Nvidia's general counsel called the Commerce official overseeing export controls last year to ask about that probe and whether Washington had a role in Singapore's detention of Alice Huang, Bloomberg reported this week. Nvidia says he was unaware of the affair.
On Thursday, Singapore said its own inquiry, into suspected false declarations rather than espionage, found insufficient evidence against her, but barred her from returning to the city-state for misleading police about her role.
Much of Washington will read this as a story about a leaky fence intended to protect the American innovation—like Nvidia’s most advanced chips—that gives it the edge in AI technology. But perhaps it’s better read as a broader warning.
If AI is beginning to help build its successors, America's computing lead could compound—but only if the U.S. sprints in three directions at once.
It means building even faster at home, aggressively guarding the chips that feed the loop, and using a wider margin to buy time for safety.
Leaky Fence, Wide Gap
There is no escaping leaks of chips. China has consistently found ways to get around American export controls and will continue to do so, in what is a tacit admission by Beijing that it is lagging the U.S. in the AI race.
China's Northwestern Polytechnical University, a military-linked institution blacklisted since 2001, bought two Nvidia H200 chips in October and rented access to more, Reuters reported.
Since January, Commerce has allowed some H200 sales to China case by case, under security conditions and not to blacklisted buyers.
Epoch AI, a research nonprofit, estimates that between 290,000 and 1.6 million H100-equivalents were smuggled into China through 2025, with a median of 660,000—roughly a third of China's AI compute.
Related Story
Nvidia’s Jensen Huang Scandal: Chips, Spies and Strategic Overreaction
Public allegations alone total nearly 300,000. Most are unproven in court, but Epoch judges the true figure almost certainly higher because so much smuggling goes undetected.
Yet the leaks have not closed the gap. Epoch's customs-based tracker finds that since 2022, about 72 percent of the world's AI compute has ended up in the U.S., against roughly 10 percent in China, including chips China makes itself.
Export controls and thin capital leave Chinese labs short of compute, Kyle Chan of the Brookings Institution told Congress in April. Hence they look to the black market for the best of American tech.
The Machine That Builds the Machine
That gap matters more if AI is starting to do AI research, widening the distance between those who are ahead and those in second place. The destination is recursive self-improvement (RSI), where AI is capable of autonomously improving itself.
AI is getting better at getting better. On November 5, the independent evaluator METR estimated that Anthropic's models had been improving on its internal capability index at about 1.9 times their pre-October 2025 rate. Most of the speedup likely came from the company's own use of AI.
At OpenAI, agents logged about three eight-hour workdays of runtime for every human workday by mid-August—a measure of activity, not output.
Treat this information with some care. OpenAI's figure is self-reported, and Anthropic, which is pursuing a hotly-anticipated IPO, redacted much of the evidence that METR found most informative.
Public benchmarks have not yet clearly accelerated, the Forecasting Research Institute (FRI) noted on November 10, and humans still supply most of the judgment in the research.
Related Story
What Kind of ‘Asset’ Was Alice Huang—and What Did China Want?
This trend upends a comfortable assumption: that if the race shifts from hardware to intelligence, chips will matter less as models become much more efficient. But the reverse is more likely, at least in the short to medium term.
Automated researchers run on chips. The more of the work AI does, the more compute becomes the input that compounds—so every diverted processor is worth more, and selling H200s to China, even case by case, becomes harder to justify.
Nor does compounding automatically favor America. China's top models lag U.S. rivals by several months or more, Chan testified. But he said China is more focused on "making progress in model efficiency, AI adoption, and the integration of AI into the physical world."
Georgetown's Center for Security and Emerging Technology highlights Chinese strengths in robotics, manufacturing, and electricity.
A lead measured in months compounds only while the inputs stay lopsided.
The Case for Brakes
Of course, the strongest argument against acceleration of frontier AI is one shouted by the labs themselves: the race itself is the real and existential danger.
On September 12, Anthropic CEO Dario Amodei urged the industry to slow capability gains so urgent safety work can catch up. OpenAI's Sam Altman agreed the same day.
Evan Hubinger, who leads alignment science at Anthropic, put the chance of AI killing all humans within a decade above 10 percent. A University of Chicago working paper models how competition pushes developers to trade safety for speed.
Related Story
Trump Wants Vance to Drain the Wrong Swamp
Nobel Prize-winning Professor Geoffrey Hinton, a computer scientist known as the "godfather of AI," is joining the chorus, warning that disaster may be just a handful of years away as humans lose control of the superintelligence they have created.
President Donald Trump's reply to such worries was blunt: "whoever wins AI wins."
Each side holds half the answer. A country a few months ahead cannot pause without fearing it will be overtaken. Neither China, nor the U.S., has the incentive to stop, or the trust that the other side will stick to an agreed pause under the guise of safety.
But a country years ahead of its competitors has bought itself time to reflect. A margin is what makes caution viable. The U.S. sprinting on AI, leveraging its many advantages, is what makes it possible to eventually slow down, for safety, at the frontier.
What a Real Sprint Looks Like
A sprint would require a rapid build-out of more infrastructure. And that means a lot more data centers. Yet on November 3, voters backed limits on data-center development in 40 of 41 local ballot measures across 10 states, some of them advisory.
Power, permits, and local consent are becoming constraints as real as chips. Public opinion on data centers is an urgent political problem that needs resolving before it becomes a grave national security issue.
Sprinting also requires firmly protecting technological advantages. So insulate export enforcement from pressure applied by the companies it polices. And pause licensed H200 sales to China until the compounding question is settled decisively.
This may well complicate diplomatic relations with China at a sensitive time, which the Trump administration is reluctant to do. But it would help to win an AI race that Trump says is essential for the U.S.
Then there is pace. Make independent evaluation a standard, not a courtesy. Anthropic began giving METR's team employee-like access in October, and OpenAI has pledged to follow. Let’s see how far ahead American frontier AI truly is.
Huang's fall is a lesson about the AI race, not one man. The factory, the fence, and the brakes are all part of a single strategy. Washington should bring the three into harmony.
View 0 Comments
Submit Correction
View Editorial & AI Guidelines