The Information

Inside Nvidia's Succession Crisis

Key points

Key takeaways from The Information interview with former Amazon board member Bing Gordon and SemiAnalysis president Doug O'Laughlin on Nvidia's succession crisis (November 2026):

A uniquely managed company. O'Laughlin argues Nvidia's flat structure — Jensen Huang and 35 direct reports — makes a fast succession terrifying: Huang was the deals man who rallied TSMC, Taiwan's suppliers and every ODM behind his roadmap, and no one inside or outside can replicate that pull.

Boards move too fast at their peril. Gordon says the biggest board mistakes come from moving too quickly ("the first question is always: how much time do we have?"), that duty of care means a botched CEO investigation haunts directors forever, and that the board must already have immediate and one-year successors named.

The internal bench. Interim CEO and CFO Colette Kress, CUDA inventor Ian Buck (the most visible public face), and VP of GPU engineering Jonah Alben are the credible internal candidates — though all are Huang's peers who would suddenly report to each other.

Outside hires fail. Big-company experience says external successors lack moral authority and waste 90 days "getting to know everybody"; if a hire must be made, a known returning insider like Arm's Rene Haas is the most credible external option.

Jensen's big swings. Huang committed ~$40B to supply during the 2020 chip shortage when peers committed ~$5B — the aggressiveness that won Nvidia its supply-chain dominance is unlikely to survive a caretaker.

The AI sector lost its 'central bank'. Nvidia's backstopping of neocloud financings will mostly continue under Kress, but less-established neoclouds may lose support, and weeks of transition uncertainty compound into quarters of lost focus.

A decapitation for US tech. Gordon frames it as an enemy force decapitating a foreign leader: with "the number one industry for AI" being war, losing Jensen — one of the oldest and most singular founders, who bet on CUDA parallel processing since the 2010s — risks US AI leadership, though O'Laughlin notes the CUDA moat has already flattened as LLMs simplify programmability and Google's TPU ecosystem closes in on inference.

Read more: The Information

More top stories