Table of Contents
Nvidia’s $530 billion in guarantees and export fights land on Kress
In sum – what to know:
- A thin succession bench – Colette Kress is interim CEO with no timeline for a permanent pick, and Nvidia had no formal succession plan for Huang’s roughly 60 direct reports.
- A financial backstop at risk – Nvidia carries about $530 billion in gross guarantees, and Kress, who resisted a $250 billion OpenAI backstop as CFO, could pull back.
- Export rules and rivals – Huang’s exit has improved the odds for three chip export-control bills while investors shift toward AMD, Broadcom and Google’s TPU.
Jensen Huang has resigned as Nvidia’s president, CEO and board member, the company announced on Nov. 12. He co-founded Nvidia in 1993 and ran it for 33 years, long enough to see it grow into one of the most important companies in AI. CFO Colette Kress will serve as interim CEO while Nvidia searches for a permanent replacement. Harvey C. Jones was named the board’s first-ever chair, and Nvidia gave no timeline for filling the role.
The resignation followed reports of extramarital relationships, and some US intelligence officials suspect China’s Ministry of State Security recruited one partner, Alice Huang, a former Megaspeed executive, to gather information on Huang. Attorney General Todd Blanche said on Nov. 12 that the Justice Department has not yet determined whether she was a Chinese spy, and officials disagree on how strong the evidence is. Nvidia’s board said on Nov. 8 it had no reason to believe company secrets were compromised.
Nvidia has lost roughly $1.7 trillion in market value since the slide began in early October, a slide that predates the scandal, and the executive who drove its technology bets, supply deals, financing arrangements and political strategy is gone. How much of the AI boom depended on one person?
Nvidia’s leadership gap
Near-term execution probably isn’t the issue, since Counterpoint Research argues that Nvidia’s roadmap from Blackwell AI processors to Vera Rubin is largely locked in, keeping its engineering bench busy for 18 to 24 months. Beyond that, the decisions get considerably harder. Bernstein analyst Stacy Rasgon has noted that Huang built Nvidia through choices made a decade ago, and no obvious candidate stands out to make the next decade’s choices.
Huang’s management style left a real vacuum, one that’s hard to fill quickly. He reportedly had around 60 direct reports, no named successor and no formal succession plan, which ValueEdge’s Nell Minow said surprised her. Kress is among analysts’ leading internal contenders, though her finance background raises questions about technical vision. Janus Henderson also names Ian Buck, creator of Nvidia’s CUDA software for developing GPU applications, as a frontrunner, and GPU engineering chief Jonah Alben is another possibility. Arm CEO René Haas and AMD CEO Lisa Su are among the external names floated.
Janus Henderson remains bullish, citing Microsoft, Google, Apple and Amazon as precedents for successful founder-to-manager transitions. The comparison is reasonable, but Huang’s role was unusually broad, touching nearly every part of the business. Nvidia’s Nov. 17 earnings report, with consensus revenue around $108.7 billion to $109.2 billion, will be Kress’s first chance to demonstrate that the company can function without him.
The AI boom’s financial backstop
The harder replacement may be Huang’s checkbook, since SemiAnalysis notes that Nvidia has roughly $530 billion in gross guarantees kept off its balance sheet, up from $184 billion in its latest 10-Q, including $36 billion in take-or-pay capacity floors for neoclouds, specialized cloud providers that rent out AI computing. Those floors require customers to pay for reserved capacity whether or not they use it, and Huang’s backing let companies with weaker credit ratings borrow at rates close to investment-grade levels. Axios puts the resulting funding hole at $500 billion and expects a more cautious next chapter.
Kress could accelerate that caution. The Information reports that she resisted a proposed $250 billion OpenAI backstop as CFO, although no outlet has reported that she canceled an OpenAI deal since becoming interim CEO. The stakes extend well beyond chip sales. Latitude Media says Nvidia’s guarantees underpin the PORTS-Pike project in Ohio, a campus of about 8 IT gigawatts with 4.25 gigawatts covered by those guarantees, and without them SB Energy likely couldn’t raise project debt. Rob Day’s most likely outcome is slow deflation rather than collapse, with construction easing and neoclouds feeling pressure first.
There are reasons not to panic just yet. Anthropic offers one hedge, since it runs models on Google’s Tensor Processing Unit chips, Amazon’s Trainium processors, AMD GPUs and some Nvidia GPUs, which reduces its dependence on Nvidia ahead of expected Nov. 18 IPO pricing.
Rivals and export rules gain ground
The political shift may be more immediate. Eurasia Group’s David Meale called Huang the industry’s most effective lobbyist against tighter rules, and his exit has improved the odds for three export-control proposals. The AI Overwatch Act would give Congress more power over certain chip licenses, the Chip Security Act would require location verification to combat smuggling, and the MATCH Act would push allies to restrict chipmaking-equipment sales to China. All three could be attached to the National Defense Authorization Act, Congress’s annual defense bill. Senate Majority Leader John Thune now says passage is “becoming a high probability.”
The response isn’t unanimous, and Democratic Sen. Maria Cantwell wants Congress to weigh what it has already done before adding more rules, while Republican Sen. Todd Young has cautioned against treating the response as formulaic. Chinese analysts have warned that Beijing could retaliate by tightening rare-earth and magnet exports.
Investors are already hedging, with AMD rising about 30% and Broadcom 24% for the week through Thursday’s close. Analyst Ming-Chi Kuo sees Google’s TPU as the biggest beneficiary, with Broadcom and MediaTek also in focus. SemiAnalysis estimates its 2027 HBM4, the high-bandwidth memory used alongside AI processors, will cost about $3.00 to $3.30 per gigabit from SK Hynix, versus $3.70 to $4.10 for AMD, Broadcom, Amazon and Google. IDC puts Nvidia at roughly 34% of China’s AI-accelerator shipments in the first half of 2026.
The immediate calendar offers few answers, though House Republicans want to interview Huang behind closed doors on Dec. 1, and Senate Banking has requested a public hearing on Dec. 3, while Commerce Secretary Howard Lutnick has been summoned to testify on Dec. 10. Those appearances may clarify the national-security questions, but they won’t settle whether Nvidia can outlast the relationships and judgment that built it.