The Jensen Huang Affair
Summary
- NVIDIA’s dependence on Huang turns the episode’s allegations into a governance risk for the whole company. Stephen Witt describes 60 direct reports and no intermediate management tier: Huang is effectively “king of NVIDIA.” Witt doubts the board will fire him, but sees congressional testimony as a route to serious legal exposure.
- The alleged chip diversion exploits the gap between delivery and actual use. Witt describes real Malaysian facilities whose customers and equipment movements remain difficult to verify. Substituting dummy servers is something he says “seems to have possibly happened”; delivery to a legitimate address would not resolve where the chips ultimately went.
- Huang’s commercial argument for China access could lose influence even if it predates the alleged relationship. Witt sees both another enormous infrastructure market and a reason to prevent Chinese competitors from building their own hardware stack. He thinks Huang would advocate this anyway, but says “his credibility is zero” and anticipates NVIDIA losing China sales entirely.
- Witt interprets the disclosure as a possible struggle over access to Trump. He recounts a reported intelligence memo that Trump did not act on, then suspects China hawks wanted to “reduce Jensen’s influence over the president.” His explanation of who leaked remains an inference.
- The muted market response could reflect existing concerns about infrastructure demand and benefits for competitors. Witt describes “very cheap valuations” as pricing a possible sales decline over the next couple of years. He also argues that damage to NVIDIA could benefit AMD, Broadcom and Google.
- A diminished Huang could weaken an influential argument against AI regulation. Witt contrasts Huang’s confidence in fixing engineering failures with developers’ fear of losing control. Yet Witt considers “If we stop doing this, China wins” the strongest counterargument to slowing development.
Deep dive
1. Huang’s concentrated power makes the allegations a succession problem
Read opens with allegations he attributes to The Times: Huang had an affair with an alleged Chinese spy whose company, Megaspeed, was suspected of routing restricted NVIDIA chips to China. He also describes possible interference with Commerce Department investigations and says Trump had known about the allegations for a year. NVIDIA is launching an internal investigation.
Witt describes a founder who has run NVIDIA for 30-plus years, with 60 direct reports and no conventional second management tier. “NVIDIA is Jensen,” he says, making disruption unusually consequential for both the company and, to some extent, the American economy. He would nevertheless be surprised if directors fired Huang: he thinks that would hurt shareholders by driving the stock down.
Congressional testimony is Witt’s bigger concern: Republicans seek closed proceedings, while Democrats could pursue public hearings after taking control. He argues the alleged relationship makes ignorance of Megaspeed’s activities harder to defend, potentially exposing Huang personally and NVIDIA to a “massive fine.” His explanation for Huang’s judgment is “pure hubris”: “He got played.”
Asked about his successor, Huang told Witt, “Nobody.” The board said in 2024 that it had considered candidates, including Jay Puri, now retired and in his 70s. Witt doubts Huang’s children can run NVIDIA because they lack technical backgrounds; he thinks the CEO must be a “world-class computer scientist.” Madison studied culinary arts and Spencer photography; son-in-law Nico Caprez has risen quickly within the company, though Witt leaves his possible role open. Despite these vulnerabilities, he thinks Huang “will probably survive this.”
2. Real facilities leave the alleged diversion route difficult to police
Witt says the export concerns were already familiar: NVIDIA revenue from Malaysia at “10 or 15%” looked to him like Chinese demand routed elsewhere. Singapore had been busting smuggling rings since 2024. He nevertheless stresses that Malaysia has a legitimate data center industry: “I don’t wanna smear them all as some kind of, like, criminal conspiracy.”
Read describes NVIDIA selling to the US arm of a Chinese company, which then sells to Megaspeed. Witt calls this the structure of a smuggling operation. His sharpest example remains explicitly uncertain: dummy servers are something that “may have happened.” Valuable equipment could arrive, substitute nonfunctional equipment could occupy the facility, and NVIDIA hardware could leave through the back door for China.
Megaspeed appears to have formed in 2023, Witt says, becoming an NVIDIA preferred cloud partner that year. It has facilities, data centers that seem to use a lot of power, and paying computing customers. The questions concern who pays and uses the equipment. Witt says Alibaba—“the Chinese eBay”—appears to be its main customer. He thinks this is a violation of export restrictions, though says it “depends on how you read the law.”
Commerce investigators have limited authority outside the United States, Witt explains: confirming delivery to an actual Malaysian facility does not reveal subsequent movements. He has also heard NVIDIA has tightened checks this year, requiring data centers to establish that their customers are legitimate and outside China. Physical security around the facilities makes independent reporting difficult.
3. China policy becomes a question of credibility and presidential access
Witt’s commercial explanation is straightforward: a Chinese data center build-out comparable to America’s would create another enormous equipment market, which “presumably would double the value of his company.” Huang also fears that refusing sales will encourage a competing Chinese hardware stack. Witt traces that anxiety to Huang’s longstanding fear of cheaper Chinese clones undermining his business.
Witt does not think the alleged romance explains this position: “I think Jensen would say this anyway.” But he believes the allegations destroy its persuasive force: “His credibility is zero.” His prediction is sweeping, while explicitly presented as his expectation: “I imagine Nvidia will in short order not be able to sell anything to China ever again.”
Witt recalls a politically disengaged Huang in 2024 who had never donated or endorsed a candidate and skipped Trump’s inauguration. He explains their later closeness as engineering applied to flattery: adjust inputs until the system produces the desired outcome. Huang understands “how to program him,” Witt says. Read quotes Trump still calling Huang “an honorable guy” and planning to take him to China.
Witt recounts a wiretap on Alice, the woman, in which Huang appeared; Secretary of State Marco Rubio then prepared a memo identifying him and shared it with Trump, who took no action. A May short-seller report and internal whispers had not established the affair. Witt suspects someone in the White House wanted the story in the press to weaken Huang’s influence, amid China hawks’ effort to retake the narrative and power. He does not identify a confirmed leaker; he names Peter Thiel, Palmer Luckey and Rubio as China hawks opposed to Huang’s softer stance on China.
4. Suspected influence damage has not produced a wider market shock
Asked about intelligence targeting, Witt thinks Anthropic and OpenAI are more valuable targets for secrets. NVIDIA’s designs alone would not let China manufacture equivalent chips without the necessary equipment. He therefore suspects an influence campaign, while acknowledging the original motive is unclear: the relationship reportedly began before Huang became close to Trump.
Read observes NVIDIA shares falling while the wider market remains relatively calm. Witt says AI infrastructure stocks trade at “very cheap valuations,” interpreting that as anticipation of a sales drawdown “in the next couple of years.” Some weakness may already be priced in, he suggests. NVIDIA’s difficulties could also benefit AMD, Broadcom and Google, which he identifies as rival chipmakers; Google in particular has become one of NVIDIA’s top rivals.
5. A weaker Huang could change the balance of the AI safety debate
Witt reconstructs Huang’s view “from the microchip up”: control over individual components makes a loss of control an engineering failure, and “we know how to fix engineering failures.” More worried voices, including Dario Amodei and Sam Altman, reason from the opposite direction: “I don’t know how this thing works. It’s extremely powerful, and I might lose control of it.”
Huang’s confidence predates the alleged affair; Witt traces it to 2015. David Sacks and Marc Andreessen share it, but Witt considers Huang its most powerful advocate. “I think that Jensen has lost this one,” he says, pointing to Trump’s tentative interest in regulation and voter concern about AI.
Witt nevertheless calls the China competition argument the strongest response to AI pessimism: slowing America does not slow its rivals. He reports little appetite for catastrophic-risk arguments in China, Japan, Korea or Singapore. AI risk is his most common question in the United States; in Asia, he says, “I never get AI risk questions out here.”