Skip to content

AX Says · Nov 15, 2026Nvidia Leadership Vacuum

Huang Resigns as Musk Rallies 110,000 to Demand His Return

Jensen Huang really has left Nvidia. Led by Elon Musk, 115,000 people have signed a petition asking the board to reinstate him. Asian governments are lining up to confirm that 260,000 GPUs and an overseas headquarters project are still on track. Nathan Lambert worries Nvidia’s funding for open-source models could dry up. First, look at how many threads Huang’s departure pulls.

Huang Resigns as Musk Rallies 110,000 to Demand His Return

Huang’s departure is no ordinary retirement. Nvidia’s board has not released the findings of its independent internal investigation. CFO Colette Kress is acting CEO, while management faces responsibility for signing off on internal controls in the November 17 third-quarter report. In the meantime, Musk, Replit CEO Amjad Masad, and Epic Games CEO Tim Sweeney signed a petition demanding Huang’s reinstatement. By Saturday afternoon, it had more than 115,000 signatures, including 5,965 current employees, 1,108 former employees, and 63,063 shareholders.

What stands out is that this is not just ordinary users venting. Executives and employees have both stepped in. Musk’s computing needs are deeply tied to Nvidia, while Replit and Epic also depend on GPU deliveries. But their demands focus on reinstatement during the independent investigation and more disclosure from the board. That suggests the real worry is not Huang’s personal fate. It is whether Nvidia has someone who can carry its strategy forward while the investigation remains unresolved.

The board’s choice already suggests this is complicated. It picked the CFO as acting CEO, not a product or engineering leader. That looks more like an effort to steady the books and customer relationships than to continue Huang’s personal, long-term bet on CUDA and the AI platform. So watch the language in the November 17 third-quarter report. If management qualifies its statements on internal controls, the reinstatement campaign and downstream customers’ compute contracts could shift again.

Asian Governments Seek Confirmation on 260,000 GPUs and an Overseas Headquarters

As soon as Huang left, several Asian governments offered different kinds of reassurance. South Korea’s science ministry said Nvidia reaffirmed plans to supply 260,000 advanced GPUs and continue joint AI projects. Taiwan’s economy ministry said the overseas headquarters project remained on schedule. Japan’s industry ministry said it would monitor developments closely and ensure a stable semiconductor supply. All three statements seem meant to reassure local customers, but their wording differs. South Korea confirmed a supply figure. Taiwan said the project was unchanged. Japan only promised to “monitor” the situation.

Those differences expose Nvidia’s hardest problem right now. Management has changed, but customers still need confidence that signed chip deliveries and project commitments will be honored. A board statement that “the strategy has not changed” cannot settle that by itself. Upstream reports point to uncertainty around China-related computing projects in particular. If an Asian customer starts renegotiating delivery schedules, that will be the first domino. Huang’s resignation will have turned from a governance issue into a supply-chain issue.

The Money Behind the Resignation: $530.5 Billion in Off-Balance-Sheet Commitments and Circular Financing

Look deeper into Huang’s departure, and Nvidia is carrying a vast pile of financing it helped set in motion. One article puts its off-balance-sheet and investment commitments at $530.5 billion. A management change could affect downstream customers’ credit, lease guarantees, and commitments to purchase compute capacity. That figure exceeds the assets of many sovereign wealth funds. The limits of that credit depend heavily on management’s personal credibility and continuity of decisions.

Meanwhile, reports say Nvidia began slowing, scaling back, or pausing some financing deals in early summer. An earlier deal with OpenAI included a $250 billion backstop for data-center funding. TechSpot relayed a claim that Nvidia finances customers to absorb excess TSMC wafer supply and keep GPU prices high. GeForce prices now often exceed 160% of their original MSRP.

Here is the loop: financing helps customers buy GPUs. Those purchases consume wafers. GPU shortages sustain high prices. High prices then support still larger financing structures. Huang was not just the CEO in this arrangement. He was more like the guarantor of the whole loop. His absence will not break it overnight, but each new commitment to purchase compute capacity could face another review. Tighter compute credit is still only a market-pricing scenario. If banks and customers act on it, downstream compute providers will feel the financial pressure before it appears in Nvidia’s earnings.

Nathan Lambert: Huang’s Departure Could Cut Off Funding for US Open-Source Models

Nathan Lambert approaches this from another angle. He is not worried about Nvidia’s share price. He worries that Nvidia’s funding for open-source models could dry up after Huang leaves. His view is that the company’s first priority is now to steady itself and deliver to customers. Spending on an open-source ecosystem that “does not directly generate revenue” is easiest to sacrifice.

This is not a baseless worry. Under Huang, Nvidia strongly supported open-source models, research communities, and the CUDA ecosystem. That support was both a technical bet and an extension of how he made decisions. An article traced his long-term bets from Riva 128 to CUDA to the AI computing platform. His bets and his personal management style were intertwined. Without a clear successor, the gap in leadership and strategic vision could leave long-term investments with no immediate payoff unprotected first.

Analysts interviewed said product roadmaps and manufacturing plans are locked in through 2027. Engineering teams and supply chains remain in place, so announced chips will not disappear. But open-source models need continuing access to compute, collaboration before release, and framework support. None of that is locked in by contract. If Nvidia cuts external support to protect customer deliveries, major companies will not feel the chill first. Open-source teams relying on Nvidia compute and CUDA will. Huang’s departure could leave a vacuum in the US open-source model ecosystem that no one quickly fills.

Ahead of Anthropic’s IPO, Dario Calls Compute Its Deepest Business Challenge

The week Huang left, Anthropic, which is preparing to go public, also laid out its compute problem for the market. Before the IPO, CEO Dario Amodei said the business faces huge uncertainty next year. Compute is the company’s deepest business challenge. This is not pessimism. It puts the mismatch in compute purchasing out in the open. Anthropic must buy compute one to two years ahead. Buy too little, and customer service and revenue suffer. Buy too much, and the company risks overextending itself financially.

Anthropic plans to cushion that risk with cash reserves, fewer advance purchases, and extra capacity from the spot market. There is a practical calculation behind this. If Nvidia tightens compute credit after its management change, major customers like Anthropic will face changes in GPU costs and delivery timing. Anthropic must prove revenue growth in its IPO while betting on compute prices for the next two years. That dilemma is harder to answer than “Will there be a new model next year?” Dario’s candor suggests pre-IPO market discussions have reached the stage of laying out the worst case first.

Small Signals Today

  • Cursor security incident winds down: Attackers used an exposed Fireworks access key to read coding agent traces. API credentials were exposed for fewer than 0.01% of customers. The incident was smaller than first feared. But the Fireworks link shows that agent tools need tighter key management.
  • The Federal Reserve names AI: A New York Fed survey found 87% of market participants ranked AI as the top financial risk, up sharply from 50% in spring. As of October, AI infrastructure debt exceeded $680 billion. The largest banks’ exposure to related financing vehicles was close to $1 trillion. This is the first time regulators have linked AI enthusiasm so closely to credit risk.
  • Lovable apps exposed at scale: Researchers analyzed 159,041 Lovable projects. They found multiple apps that let anyone read passport information, more than 200,000 chat messages, and 80,000 user profiles without logging in. Repeated Supabase integration patterns let one permissions error spread across many apps. People are starting to count the cost of low-code-generated apps.
  • Anthropic prompt advice: Claude Opus 5.5 users are advised to remove urging phrases such as “think hard.” Starting at Medium reasoning effort makes responses start faster, with no measurable drop in quality.
  • Runway releases a robotics model: Praxis-1’s weights and code are available under Apache 2.0. It achieved a 46.2% success rate on RoboLab-120 in Runway’s own tests, but that result has not been independently verified.

Translated from the Chinese original. Read the Chinese original

Some sources via AI HOT

The day's picks