Latest / Super Intelligence News and Report / Jensen Huang's Affair, Intelligence Probe, and Nvidia's Key-Man Risk
Transcript
- Super Intelligence News and Report — Jensen Huang's Affair, Intelligence Probe, and Nvidia's Key-Man Risk Welcome to Super Intelligence News and Report. Today is Monday, November 9th, here are the headlines. Maya: This weekend, trust in the people and institutions that oversee AI got tested from several directions at once, and markets kept repricing the build-out anyway. We start in Asia, where chip stocks bounced hard on Monday after Friday's rout. Taiwan's main index rose as much as six point seven percent during the day, and TSMC closed up almost five percent. That's about the same size as Nvidia's drop in US premarket trading. Samsung and SK Hynix gave back most of their early gains, but the takeaway for now is that investors see the Huang scandal as an Nvidia problem, not a problem for AI demand. Next, Anthropic and power. The Financial Times, via CNBC, says Anthropic committed about three point four billion dollars in late September for heavy-duty gas generators from OnePWR Solutions in San Antonio. That's reported, not confirmed, and neither company commented. The figure to remember is one point seven gigawatts, roughly the output of one or two large power plants. Theo: Do we know where they're going? Maya: No, the sites aren't known. The odd detail is that OnePWR ordered these from Caterpillar two years ago and couldn't get financing. Why it matters: labs are building their own power to skip the grid queues, just as voters push back on data centers. That leads into Anthropic's other big week, the IPO. Reports say the roadshow starts this week, but as of Sunday no price-range amendment had shown up on EDGAR, so it's reported, not confirmed. Bloomberg's Monday newsletter says OpenAI and Anthropic are both trying to close large financings this week at lower targets after the midterm sell-off. And The Information says AI companies have sold or priced more than three hundred billion dollars of stock this year. So the biggest AI listing yet is heading into a market already absorbing a flood of AI shares. In Washington, Trump told 60 Minutes he doesn't mind guardrails on AI. Reuters reports he said the risks can be managed and he's open to working with Democrats, as long as the US doesn't end up in what he called a deep second position behind China. In September he posted that a strong president was the only guardrail AI needs. The tone is softer, and it comes just as a Democratic Congress gets its safety bills ready, but no policy came with it. The agency that's supposed to do the actual testing is in rough shape. The AP, citing two former officials, reports that staff at the Commerce Department's frontier AI testing center were recently asked to report informally to a White House office. That's reported, not confirmed. The center is now called the Center for Advancing Innovation and Standards for Super Intelligence. The director hired in April was fired within days, the replacement resigned after a few months, and the center stopped publishing its model-safety reports, which Congress began asking about in August. It runs on about ten million dollars a year. The concern is that the government's only independent frontier-model tester may be losing its independence, and the Brennan Center's Amos Toh warns that treating AI safety as a national-security issue could push aside consumer and civil-rights protections. One more on policy, following up our Google coverage from last week. On Monday, Senator Ed Markey wrote to Sundar Pichai asking Google to let K-12 schools switch off AI Overviews and AI Mode while keeping Search, and whether an AI-free Search is even technically possible. His letter cites a lawsuit alleging a thirteen-year-old confided suicidal thoughts to Google's AI on a school Chromebook, and he wants Google's research on student safety and learning by the end of November. Putting AI inside Search is now a political issue, and it comes with a deadline. Over to you. Theo: My sleeper pick for the weekend is also Nvidia, but it has nothing to do with the scandal. On November 8th, Nvidia, the Korean chip startup Rebellions and SK Telecom announced that Rebellions' R200 inference chips will plug into Nvidia systems through NVLink Fusion, which is Nvidia's program for connecting other companies' chips to its networking, built on Nvidia's MGX rack design. They split the work by phase: Nvidia GPUs do prefill, reading the prompt, and Rebellions does decode, writing the answer. Maya: So Nvidia's happy to share the rack with a rival. Theo: As long as it's Nvidia's rack. SK Telecom will deploy it with backing from Korea's science ministry, and systems are expected in 2028. Nvidia would rather sit inside rival chips' systems than get replaced by them, and that could become the template for sovereign AI deals. Next, xAI, and this one comes from Musk's own posts on X. Paying subscribers to Grok Bot, xAI's agent product, complained that their usage meters hit a hundred percent by midweek and that the agent had gotten worse. On November 8th Musk replied that he'd dig into it that day and promised better token efficiency. A Hong Kong outlet reports he then posted about a review meeting with engineers and fixes overnight. There's no higher limit, no reset and no price change, and nobody knows what the fix actually is. Back on November 1st, Musk said Grok Bot cost more to serve than it brought in. It's a clear example of agent products running into what compute really costs. Last one is from the bookshelf, so it's a week old. Bloomberg reported on November 2nd that Goldman Sachs restricted Claude Code and GitHub Copilot to its roughly twelve thousand engineers and quant specialists, which cuts off most bankers and traders. That's reported, not confirmed. Goldman cited productivity and risk management under guidance it updated in October. Maya: Wasn't their CIO cheering this exact thing over the summer? Theo: In July, yes. He was publicly praising non-engineers who built their own tools with AI. Now a major bank is pulling back on vibe-coding across the whole firm. Maya: Next, we go deep on Nvidia's board investigating Jensen Huang, and what it means for Washington's China chip policy. Now for the Main Episode Maya: Here's the question for today. Nvidia, and Washington's whole China chip policy, have treated Jensen Huang as irreplaceable. Can they keep doing that now that he's admitted an affair with a former customer executive whom US officials reportedly flagged as a Chinese intelligence recruit? Theo: A week ago nobody would have asked that. Start with the people, because the names are confusing. Maya: Two Huangs, no relation. Jensen, Nvidia's founder and CEO. And Alice Huang, Chinese name Huang Le, a Chinese citizen and former executive at Megaspeed. That's a Singapore-based data-center and cloud company that buys Nvidia chips, and the US suspects it of helping China get around the limits on AI chip sales. Theo: Where does Megaspeed come from? Maya: It was created in 2023, when a Chinese gaming and cloud company called 7Road, which has state-linked backers, split off its overseas operation in Singapore and renamed it. Alice Huang was listed as managing director for its first eight months, and she presented herself as CEO. Singapore authorities have confirmed to the Times that Megaspeed is under investigation. Megaspeed says it complies with all applicable laws. Theo: So not some random contact at a conference. Maya: No. Megaspeed was reportedly set to buy about two billion dollars of Nvidia AI hardware, with more than three billion more planned. Those orders stopped after official investigations began and she left. We don't know when or why she left, or what she does now. Theo: And the size of that matters, doesn't it? The bigger the customer, the more the access starts to look like the point. Maya: That's my read. Being a big customer doesn't make the relationship innocent. It makes it more worth asking about. Theo: So how does the intelligence piece come in? Maya: Through The New York Times. On Friday it reported that US policymakers got a memo last spring identifying Alice Huang as recruited by China's Ministry of State Security, its main civilian intelligence agency, and describing her relationship with Jensen. On Sunday the Times added, citing people briefed on the matter, that Secretary of State Marco Rubio oversaw the memo's creation. That intelligence thread is reported, not confirmed. Theo: And there was earlier reporting about the president. Maya: Around the end of October, the Times reported that Trump had been briefed about a year earlier and took no action. We have that through secondary coverage, and the White House didn't respond to CNBC on it. If it's right, Washington knew about this long before the public did. Theo: Then on Sunday, Huang goes to his staff. Maya: He emails Nvidia employees and admits the affair. He says he's deeply sorry, apologizes to his wife and children, says he has fallen short of his responsibilities, and says he's proud and grateful to be American. Theo: That last sentence isn't from a normal apology. That's written for Washington. Maya: Exactly. The same email discloses that the board has opened an independent investigation. A committee of independent directors, working with outside counsel, will look into allegations that Huang was the target of a, quote, foreign intelligence effort. The board says it has no reason to believe national security or Nvidia secrets were compromised, and that Huang supports the review. Theo: What jumped out at me is what's missing from that statement. Maya: The affair. The board framed it entirely as a security question, and a spokesperson said Huang is committed to continuing as CEO without interruption. Theo: That's a deliberate choice. An investigation can plausibly close the security question. It's much harder to close the judgment question. Maya: And some of Nvidia's earlier lines look awkward now. It had described the contacts between the two Huangs as ordinary customer interactions, not unusual, and said it doesn't share classified information with customers. Theo: Ordinary customer interaction is a hard line to hold once your CEO has admitted an affair with the customer. So how did Washington react? Maya: Fast, and from both parties. Senator Chris Coons, who's set to chair Senate Foreign Relations in January, posted that Congress will investigate. Representative Seth Magaziner on House Homeland Security wants Commerce Secretary Howard Lutnick to testify alongside Huang. And Republican Senator John Cornyn called a related Times report outrageous and asked for a congressional investigation. Theo: And there was already a House request out before any of this. Maya: Three House Republican chairs, John Moolenaar of the China Select Committee, Brian Mast of Foreign Affairs and Rick Crawford of Intelligence, had asked Huang last week to attend a closed-door briefing on November 18th. That's the day after earnings. Theo: Earnings on the 17th, Capitol Hill on the 18th. Rough week. Maya: And at the end of the month, Nvidia's big conference is in Washington, D.C., of all places. Theo: What did the White House say? Maya: Trump praised him. He said he hadn't spoken with Huang about the investigation, called him an honorable guy, and said he should stay on as CEO. Theo: Which is interesting, given that reported briefing a year ago. Maya: And Democrats are asking exactly that: what did the administration know, and did it act on the spring memo? We don't know. Beijing also weighed in on Monday. Foreign Ministry spokesperson Guo Jiakun called the so-called Chinese spy narrative unsubstantiated. Theo: Okay, the stock. People are going to look at it and draw conclusions. Maya: Nvidia was down nearly five percent in premarket Monday, after already sliding about four percent in overnight trading on Sunday. And that comes on top of six straight weekly declines. Theo: So how much of that is the scandal, and how much is the trend? Maya: Hard to separate. Nvidia is about twenty-one percent below its early-October peak. And before this weekend it traded at roughly twenty-four times trailing earnings, against a five-year median around fifty-five. So it was already cheap by its own history. Theo: Less than half its usual multiple. Maya: You can read that two ways. The market had already priced in a lot of bad news, and this is one more push. Or, as Mohamed El-Erian put it in secondary coverage, key-man risk looms large and probably isn't priced in. Theo: And the rest of the chip world shrugged. Maya: Asian chip stocks rose on Monday, and TSMC closed up almost five percent. So investors are treating this as a Nvidia problem, not an AI demand problem. Joachim Klement at Panmure Liberum agrees the impact is contained for now. But if Huang's position is really in danger, he says, Nvidia might pull back its support for other companies, and the pressure would spread across the sector. Theo: That's the part I care about on the build side, because Nvidia is a hub other companies plug into. Literally, this weekend. It announced that the Korean startup Rebellions will connect its inference chips into Nvidia systems through NVLink Fusion, with SK Telecom deploying them. Deals like that depend on stable, engaged leadership. Maya: And they take years. That one isn't expected to be ready until 2028. Theo: Right. If you're a partner betting on a multi-year roadmap, a distracted CEO costs you something even if he keeps the title. Maya: So, the people arguing about Huang himself. The survive camp is loud. Dan Ives of Yorkville Ives called it a black eye moment and a game changer, but said talk of replacing Huang is, his word, comedy, and that Nvidia is still two to three years ahead of China in AI chips. Shaun Maguire at Sequoia called Huang irreplaceable for US AI leadership, and Elon Musk agreed. Bill Ackman said Huang is human and has taken full responsibility. Theo: I find the irreplaceable argument partly persuasive, and not because of the hype. Nvidia has no clear number two, and nobody is publicly answering the succession question. Remove a founder with nothing behind him and you haven't removed risk. You've swapped it for a bigger one. Maya: But irreplaceable is the risk. That's what key-man risk means. Nobody being able to name a successor isn't a defense of Huang. It's an indictment of the board's planning. Theo: Fair. Both can be true. He's the best person to run it today, and it's a governance failure that he's the only option. Maya: I'll take that. On the other side, Gary Black at The Future Fund doubts Huang can stay as CEO and expects portfolio managers to cut their holdings. But he doesn't see long-term damage to Nvidia's value. Theo: The CEO goes, but the company's fine? Maya: It holds together. The business doesn't hang on one man's title, but institutions don't want headline risk in their books. Bernstein, in a note via Bloomberg, is more measured: Huang's stewardship and judgment are now under debate, and the overhang will last. Theo: That feels like the honest middle. He's not out, and it's not a joke either. It just won't resolve in one news cycle. Maya: And here's where I push back on Ives. Being two to three years ahead of China is an argument about Nvidia's technology. It isn't an argument about whether its CEO should be handling its China relationships, and the survive camp keeps answering the first question. Theo: But who else handles them? Huang has been the face of the push for broader chip exports to China, and he's been closely aligned with Trump on AI policy. That's a personal relationship, not a box on an org chart. Maya: Which is exactly why it's compromised. Can he credibly walk into a room on China policy while Congress investigates whether China targeted him? I'd lean no. Theo: So this is where it stops being about one man's judgment. Maya: Right. That's the related Times story from Saturday. The Commerce Department approved dozens of Chinese companies to handle Nvidia's H200 chips. One is New Matrix, a Beijing firm cleared to assemble H200s into AI servers. Reporters went to its listed address and found an empty building with wires hanging from the ceiling. Theo: Oof. Maya: And its owners are reportedly tied to Wise Road, a Chinese private-equity firm that Commerce itself put on its Entity List, its export blacklist, in 2024. Commerce says every application got a national security review. Theo: We should be careful, though. These were approvals to handle chips, not confirmed deliveries. Maya: Correct. But the approval is the vetting. If the vetting clears an empty building owned by people linked to a blacklisted firm, the question isn't only whether Jensen Huang was careless. It's whether the export system works. That's why Magaziner wants Lutnick sitting next to Huang. Theo: One piece of context cuts the other way. The FT reported earlier this month that Nvidia told its cloud partners to keep Chinese customers below forty-nine percent of their business worldwide. Maya: So Nvidia knew the exposure was a problem. You can read that as diligence, or as drawing the line exactly where it thinks it can get away with it. Theo: I'd call it diligence. It's a crude rule, but it is a rule, which is more than that empty-building approval suggests Washington had. Maya: That's where we differ. To me, Nvidia's guardrails and Commerce's both look thin right now. Patrick Moorhead at Moor Insights makes the point I think matters most: probes from the government, the board and Congress, but the real debate moves to China policy, and he thinks export-control legislation will probably pass eventually. Theo: And that hits the people I cover harder than anything about Huang. If you build on Nvidia hardware outside the US, or you run a cloud selling into Asia, a law lasts much longer than a CEO does. Maya: Especially with a Democratic Congress arriving in January and Coons holding the Foreign Relations gavel. Nobody knows yet whether it becomes legislation, but the motive just got a lot stronger. Theo: The board says it has no reason to believe secrets were compromised. How much weight does that deserve? Maya: Some, not much. It's a statement on the first day of the board's own investigation. We don't know what the committee will find, how long it'll take, or whether the findings will be made public. We don't know which agency, if any, was watching the relationship, or since when. And we don't know whether anything reached Beijing. Theo: So that one sentence is holding up a lot. Maya: Mostly the stock price. Theo: So where do you land? Can they keep treating him as irreplaceable? Maya: Nvidia probably can, for now. The board's framing, the president's backing and the lack of a successor all point to Huang staying. But Washington can't keep treating him as the irreplaceable go-between on China. That's the part that's actually broken. And I suspect the bigger scandal ends up being the administration's: the spring memo, the reported briefing a year ago, and the New Matrix approvals. Theo: I'm more comfortable with him keeping the whole job, because right now the alternative is a vacuum. But irreplaceable is now a liability Nvidia has to fix in public. If the November 17th earnings call brings a real board update, and someone other than Huang credibly answers questions, that's the beginning of a succession story. If it's all Jensen and all deflection, the key-man discount El-Erian describes starts to look real. Maya: And the next morning is the one I'm watching: the closed-door briefing on the 18th. Does Huang show up, and does Lutnick get called too? If Lutnick ends up in that room, this has become a story about Washington's vetting, not one man's affair. Theo: It started as a question of whether Nvidia can survive without him. It might end as a question of whether US chip policy should ever have depended on him. Maya: And within two weeks we'll hear the board's first answer on an earnings call, and Congress's first questions behind closed doors. That's when we find out how much one person was really holding up. Thanks for listening to Super Intelligence News and Report. Please subscribe and share it with your friends.