Gavin Baker: Jensen Huang is irreplaceable, open source AI is essential for cyber defense
Nov 17, 2026 with Gavin Baker
Key Points
- Jensen Huang's public willingness to return as Nvidia CEO signals irreplaceability rooted in supply-chain relationships and buyer dynamics that cannot transfer to successors like CFO Colette Kress.
- Open-weight AI models are national security infrastructure for cybersecurity; closed models have failed to solve major hacks while ensemble deployments of open models enable defense-dominant posture against novel attacks.
- AI is positioned to accelerate genomics and personalized medicine where human cognition previously hit a ceiling, with cyber defense budgets rising sharply as institutions enter peak AI hacking phase.
Summary
Read full transcript →Gavin Baker: Jensen is irreplaceable, open source wins the cyber war
Jensen Huang's return is the clearest signal the market was missing. Until now, no one knew whether Jensen wanted his job back. Baker, who co-wrote a letter with Brad Gerstner calling for Jensen's return, says every Nvidia shareholder, employee, and corporate partner who has reached out since has been supportive. Jensen saying publicly he would be "honored to return" changes the calculus — it's the difference between a vacancy and a succession.
Baker thinks Jensen is irreplaceable, and the Apple-after-Jobs comparison only goes so far. Apple post-Jobs faced limited competition and a clear strategic path — lighter, faster, thinner, cheaper. Nvidia faces a fundamentally different environment: open-weight models eroding switching costs, a concentrated monopsonistic buyer base that could reassert pricing power if the model layer consolidates, and supply chain relationships that run on handshakes. Jensen has never had a contract with TSMC and yet holds its best relationship. Those aren't transferable. Colette Kress, Nvidia's CFO, is capable enough to serve as a steady interim CEO, which removes pressure on the board to move fast — but steady is not the same as irreplaceable.
“Since Brad and I wrote a letter, I have yet to speak to a single NVIDIA shareholder, employee, corporate partner who is not supportive of Jensen's return. Jensen is irreplaceable. Those supply chain relationships are very unique. He said they've never had a contract with Taiwan Semi, and yet they have the best relationship with Taiwan Semi.”
Open source as national security infrastructure
Baker's sharpest argument concerns cybersecurity. Every major hack to date, he says, has originated from closed models and been solved with open ones. The reason open-weight models matter for defense is structural: individual AI models have jagged capability profiles, excelling in one or two narrow areas. Defending against a novel AI attack requires an ensemble of open-weight models, each spiking on a different defensive task, deployed as a swarm. Closing off access to open-weight models would strip American institutions of that ensemble and, Baker argues, produce far more successful attacks.
He also frames inference speed as a military asset. Atreides is an investor in Cerebras, currently the fastest token-generation platform available, and Baker sees that speed advantage becoming decisive in cyber — if you can generate tokens faster than your attacker, you hold an edge. Dario Amodei at Anthropic has suggested cyber will eventually become defense-dominant; Baker agrees, and thinks the ensemble-plus-speed model is the mechanism that gets there. Cognition, another Atreides portfolio company, just launched what Baker calls a "security swarm" product — its fastest-growing offering — built on the same ensemble logic. Cerebras has a partnership with CrowdStrike.
Baker's read on timing: AI cyber budgets at companies and institutions rose every day last week, and he thinks the industry is currently in "the teeth of peak AI hacking." Defense dominance is coming, but not yet.
Biology as the next frontier
Baker ran a biopharmaceutical fund for three years in a prior life. He draws a direct line from the year-2000 genome-sequencing optimism — which followed Moore's Law on cost but didn't meaningfully accelerate medical progress — to today. The difference, in his view, is that genomics and personalized medicine were too large a problem space for human minds or human-written software. AI is not similarly constrained. He expects material acceleration in human lifespan extension as a direct result, though he stops short of naming specific investments or timelines.
Every deal, every interview. 5 minutes.
TBPN Digest delivers summaries of the latest fundraises, interviews and tech news from TBPN, every weekday.