Breaking News: Jensen Huang

Sex, Spies, and AI Hardware

I like Jensen Huang. He was good for NVIDIA, and I formed that opinion the way most of us do, from the products and the company's performance. This week showed me what that method leaves out. Marital infidelity is a personal matter, and you already know where you stand on it. The business question is a different one: can a relationship give somebody else leverage over the person running a company you depend on?

// Huang resigned Thursday morning. NVIDIA's board announced that he had resigned as president, CEO, and director, effective immediately. It gave no reason, and the announcement carried no comment from Huang. CFO Colette Kress is interim CEO and keeps the CFO job. Harvey Jones, the former Synopsys CEO, becomes the board's first-ever chair and will run the search for a permanent CEO, with inside and outside candidates.

// The first affair is admitted, and the national security claim is not proven. Huang told employees on November 8 that he was unfaithful to his wife. NVIDIA's board opened an independent investigation into reporting that he was the target of a foreign intelligence effort, and said it had no reason to believe national security or company secrets were compromised.

// The woman at the center worked inside the Singapore chip route. U.S. officials believe Megaspeed executive Alice Huang, no relation, was working for Chinese intelligence, which Beijing denies. Megaspeed reportedly bought $4.8 billion in NVIDIA chips over two years and has been investigated in the U.S. and Singapore over possible diversion to China.

// A second allegation turned a security story into a governance story. The New York Times reported Wednesday that Huang also had a relationship with a vice president who reports directly to him. It is unconfirmed. Hours later, The Wall Street Journal reported that the board was preparing to remove him.

// One person was carrying too much of the company. Until Thursday, NVIDIA's board had no chair. There is still no named successor, and Huang had about 60 direct reports, according to Fortune. Analyst Dan Ives had said NVIDIA would be worth 25% to 30% less without him. The stock opened down 12% on the news.

The leaders who come through this well will have stopped filing their AI suppliers under vendor management and started treating them as supply chain and security exposure, people included.

I had a different Deep Dive ready for this morning. It was about how gaming built NVIDIA's AI advantage, and it opened in a Denny's booth in 1993 with three founders and a lot of coffee. It was a warm piece. I meant every word of it. I'm not sending it to your inbox. It's on the web if you want the history, written before this week, and I've left it that way.

Earlier this week in AI Tangle, I asked whether Jensen Huang was a supply risk. My answer was no, he's a concentration risk. That was written on Sunday's news. By Wednesday night there was a second allegation and a report that the board was preparing to oust the man who has run the company for 33 years. By Thursday morning he was gone.

So let me say where I stand. I like Jensen. He was good for NVIDIA, and it's a shame it ended like this. I also have to admit where that opinion came from. I based it on the chips, the software, the revenue, and the keynotes. I've spent my career evaluating technology, and the habit is to read the benchmark, the roadmap, and the income statement. NVIDIA passes all three. None of them would have told you a thing about this week.

I spent a good part of my career in open source, where the old line was that you trust the code because you can read it. Then came xz. In 2024, a contributor calling himself Jia Tan was caught planting a backdoor in a compression library that sits underneath a huge share of Linux systems. He didn't break in. He spent about two years being helpful, earned the trust of one worn-out volunteer maintainer, and got himself made co-maintainer. The code was fine right up until the person was compromised.

That's the lesson I keep having to relearn. I'm increasingly looking at the people involved, and I think you should be too.

This story moved fast, and a lot of it rests on anonymous sourcing. Here's how I'd sort it.

Two things deserve care. First, nobody has shown that Jensen Huang knowingly gave anything to anyone. His note linked his infidelity to the national security concerns in the news without naming the other person, and the board's own statement says it has no reason to believe secrets were compromised.

Second, people keep asking whether there were earlier allegations. There isn't a long public record. What exists is this week's reporting: CNBC cited a former employee who said rumors about the second relationship had circulated inside the company for years, and reported that Huang had retained a crisis PR firm as early as last year. That's a rumor and a retainer. It isn't a finding, and I'm not going to dress it up as one.

The market didn't wait for findings. NVIDIA shares fell 9.5% across Monday and Tuesday, then dropped again after hours Wednesday to about $160.95, 13.8% below the November 6 close. After Thursday's announcement the stock opened down 12%, putting NVIDIA on track to lose more than $700 billion in value since last week's close.

Because without it, this is a tabloid story. With it, this is an export control story.

For two years, people who follow chip policy have asked how restricted NVIDIA GPUs keep turning up in China. Singapore sits in the middle of most of the answers. In February 2025, Singapore charged three men with fraud in a case tied to servers allegedly shipped onward to China. NVIDIA's explanation for its large Singapore revenue line has been consistent: customers book invoices there, and actual shipments to Singapore were under 2% of fiscal 2025 revenue.

Megaspeed is the biggest name in that conversation. A New York Times investigation in October 2025 described a little-known Singapore data center company, spun out of a Chinese gaming firm, that spent roughly $2 billion on NVIDIA's most advanced hardware and was suspected of making it available to Chinese customers. A U.S. Commerce official visiting its Malaysian site reportedly found NVIDIA equipment still sitting in boxes. NVIDIA has said it found no evidence its chips were diverted.

Alice Huang was a Megaspeed executive. That's the connection that changes the category of this story. The person U.S. officials believe was working for China's Ministry of State Security wasn't a stranger at a party. She was tied to one of NVIDIA's largest customers in Southeast Asia, a customer whose entire business depended on a decision NVIDIA makes every quarter: who gets the chips.

Now add the policy side. Huang has been the most effective lobbyist in the industry on China. After the administration restricted NVIDIA's H20 chip in April 2025, a move the company said would cost it $5.5 billion, Huang lobbied the President and the restriction was reversed that July. The Times has reported that Trump was briefed on the relationship in late 2025 and did not act, and that a classified memo naming Alice Huang as a Chinese intelligence asset circulated this spring.

I am not saying any allocation or any lobbying position was influenced by that relationship. Nobody has shown that. I'm saying the board now has to prove a negative across years of decisions, and so does everybody who relied on those decisions.

This is the comparison I keep hearing, and it's a fair one. Two of the most famous founders in tech had affairs that became public. Neither one ended up in front of the FBI.

Jeff Bezos's relationship became public in January 2019. The other person didn't work for Amazon, wasn't a customer, and had no tie to a foreign government. Here's the part people forget, though. Even that one produced a leverage attempt. Bezos accused the National Enquirer's publisher of extortion and blackmail over private photos, and he ended it by publishing the threat himself. A secret is only leverage while it stays a secret.

Bill Gates is a different category. Microsoft's board brought in an outside law firm in late 2019 after it received a concern about his involvement with an employee back in 2000. He left the board in March 2020, before the review finished, and his spokesperson said the two were unrelated. Gates hadn't been CEO for two decades. That was a workplace conduct question, and a board handled it.

So I'd sort executive affairs into three kinds, by who the other person is.

A private party. It's a family matter and a reputational one. Bezos.

Someone inside the company. It's a governance matter, because of power and policy. Gates. If Wednesday's report is confirmed, the second Huang allegation lands here.

Someone with business in front of the company, or a government behind them. It's a control matter. Somebody other than the board and the shareholders may have had a say. That's the first Huang affair, as U.S. officials describe it.

The first two kinds have happened plenty of times. The third is the one that gets a president briefed.

The federal government worked this out a long time ago, and its answer is useful because it isn't moralistic. The standard used to grant security clearances treats sexual behavior as a concern when it creates "vulnerability to coercion, exploitation, or duress." Adjudicators generally don't pull a clearance for adultery alone. What they care about is whether the conduct is hidden, because the hiding is where the leverage lives. The usual fix is disclosure.

The government doesn't grade morality here. It grades leverage. I think that's the right test for a CEO too.

And blackmail is the crude version. The more likely version never involves a threat at all. It's access: who gets the dinner, the text reply, the benefit of the doubt. It's information: what gets said casually about allocation, roadmaps, or what Washington is thinking. It's judgment: a person you care about asks for something that's almost defensible on the merits, and you find the merits.

Fortune notes that Huang personally texts and calls customers and sits at the center of most decisions. That's been described for years as a strength, and I described it that way myself in the piece I pulled. It's also what makes one relationship matter so much. When allocation, pricing, and policy all run through one person, getting close to that person is worth a great deal to somebody.

It's a real case, and I don't want to skip it.

The bull case. The results are extraordinary: $215.9 billion in fiscal 2026 revenue, up 65%. Nobody has produced evidence that he knew who he was dealing with or that he gave anything away. A Blind survey found 88% of 232 NVIDIA employees want him to stay. And removing a CEO without a successor has a track record. When Intel pushed out Brian Krzanich in 2018 over a past relationship with an employee, it had nobody ready, and a year later the stock was down 8.7% while the S&P 500 was up 8.5%.

The bear case. If the second allegation is confirmed, it's the kind boards have fired for regardless of performance. Krzanich went. McDonald's fired Steve Easterbrook in 2019 for the same reason, named his successor in the same announcement, and the stock beat the market over the following year. There's also the apology itself. If his November 8 note left out a second relationship, the board has a candor problem on top of a conduct problem. And Congress is talking about a subpoena.

My read: the first affair made him a risk the board had to investigate. The second, if it holds up, is a policy violation with precedent. Neither one tells you he was manipulated. Both tell you this board let one person become the entire control system and never wrote down who comes next.

The board settled the first question Thursday. Look at how it did it, though. It made the CFO interim CEO, left her in the CFO job, and named a chair to run a search with no timeline. Intel did the same thing when Krzanich resigned in 2018: CFO Bob Swan became interim CEO and stayed CFO. It's a sensible stopgap. It also means the person who runs the money is now running everything else too, five days before earnings.

No. Earlier this week I called it a concentration risk, and that undersold it.

Vendor risk is a supplier missing a ship date or raising a price. This is a supply chain risk, because a handful of companies supply the compute that every model you use runs on. It's also a security risk, because the people at the top of those companies are now targets. That isn't my word. NVIDIA's own board said it was investigating whether its CEO was the target of a foreign intelligence effort.

The people supplying the market with this critical infrastructure matter more than they ever have. We spent two years asking where the chips go. We should have spent some of that time asking who decides.

That's the question in my inbox this morning, and the timing makes it fair. Anthropic's IPO roadshow is underway, with a debut targeted for the week before Thanksgiving. NVIDIA reports earnings Tuesday. The two biggest AI stories on Wall Street are about to land in the same week.

I get the appeal. If you want AI in your portfolio and you just watched a CEO scandal help knock more than $700 billion off a company in a week, a lab whose control is spread across seven co-founders looks like less key-person risk. I'd slow down on that for two reasons.

It isn't a way out of NVIDIA. NVIDIA invested $10 billion in Anthropic in November 2025, and it's one of the chipmakers putting money into the IPO itself, part of a $26 billion private placement with Microsoft, Google, TSMC, Broadcom, AMD, and Micron. And Thursday's market didn't treat Huang's exit as a reason to buy more AI. AI stocks fell around the world, and the money that moved went to NVIDIA's chip rivals, AMD and Broadcom.

Anthropic concentrates control too, and it writes it down. Its S-1 gives the seven co-founders 50.1% of the vote through a single Class F share, while its Long-Term Benefit Trust elects a majority of the board. That's more checks than NVIDIA had, it's spread across seven people instead of one, and it sunsets if fewer than three founders remain. It's still the question this week taught everybody to ask.

Here's the part that matters if you buy AI rather than trade it. Once Anthropic is public, it has to disclose who controls it, who sits on its board, and what it depends on, and keep those disclosures current. That's most of the key-person assessment I'm about to ask you to write. Whether you buy the stock is your call. Either way, the filings are about to get a lot more useful.

Misstep 1: Treating this as a morality play. Where you stand on infidelity is yours to decide. It won't help you assess a vendor. Ask whether the conduct created leverage and whether anyone with oversight knew.

Misstep 2: Treating performance as character evidence. I did this. Great products and great quarters tell you a company executes. They don't tell you who has influence over the person making the calls.

Misstep 3: Confusing an allegation with a finding. The first affair is admitted. The espionage claim comes from unnamed officials and is denied by China. The second relationship is unconfirmed. Hold those three at three different confidence levels.

Misstep 4: Filing this under vendor management. Your next GPU order isn't in danger, so procurement will call it handled. The exposure sits upstream of procurement: export rules, allocation conditions, and whether the people at the top of your supply chain can be worked on by somebody else. That belongs with your security and risk leaders.

// Key Takeaways

Treat AI compute as critical infrastructure in your supply chain. A handful of companies supply the chips every model you use depends on. Give them the same review you give your cloud, your power, and your network, with a named owner.

Put the people in your threat model. Whoever controls allocation, pricing, and policy at a critical supplier is a target, and this week a board said so in writing. Review the board's independence, the succession plan, and how much runs through one person.

Write down your second source this quarter. I said it earlier this week and it's truer now. A plan B counts when it names a provider, a workload, and a date you tested it.

Apply the leverage test inside your own company. Your executives hold model access, customer data, and pricing authority that outsiders want. A disclosure policy for relationships with employees, customers, and counterparties protects them as much as it protects you.

Here's what I'd take into the next planning meeting. First, move your top three AI suppliers out of the vendor file and onto the supply chain and security risk register, with an owner who isn't in procurement. Second, a one-page key-person assessment for each: chair or lead director, named successor, direct reports to the CEO, who decides allocation, and any regulatory exposure tied to one individual. Third, a tested fallback for one production workload, with a real number for what the move costs.

I still think NVIDIA's chips are great. I still think Jensen built something most of us never will. But the people supplying this infrastructure matter more than they ever have, and I've stopped believing the products and the numbers are the whole diligence.

Your AI Sherpa,