All-In on Jensen Huang’s Nvidia Exit: The Leak, China and Anthropic’s IPO

The All-In Podcast hosts pick apart Jensen Huang’s sudden exit as Nvidia CEO: the leaked spy memo, the export control fight, the market fallout and Anthropic’s IPO. In this episode, recorded on November 12, 2026, David Sacks, Chamath Palihapitiya, David Friedberg and Jason Calacanis work through six days that ended a 33-year run, from the New York Times report that a Megaspeed executive named Alice Huang was suspected of gathering information for China, to Huang’s resignation and CFO Colette Kress stepping in as interim CEO. The allegations remain allegations: Beijing has denied the spying claims, Nvidia has said it had no reason to believe any secrets were compromised, and investigations are ongoing. The hosts say as much, then speculate anyway.

TLDW

Sacks argues there were two operations: whatever Alice Huang and Megaspeed were doing, and a criminal leak of a classified memo meant to take Huang down. Friedberg and Chamath think the board acted on code-of-conduct attestations and insurance covenants, not espionage. The hosts agree China does not benefit technologically, but that losing Huang slows Nvidia’s yearly chip cadence, freezes the “Bank of Nvidia” and could cost the US economy real growth. The episode closes on Anthropic’s IPO road show, where Sacks accuses the company of selling recursive self-improvement to investors while the AI safety movement cheers Huang’s fall.

Thoughts

China does not want Nvidia’s chips, and that one fact undercuts most of the theories about why Huang had to go. Sacks walks through how Beijing discouraged the H20, then kept the newly approved H200 out, because it is forcing its own market onto Huawei so Huawei can scale. That makes the China hawks and the Chinese Communist Party allies of convenience, both working to keep American GPUs out of China for opposite reasons. If that is right, then the policy question Huang was supposedly ousted over was already settled by Beijing, and his “sell lagging chips to take share from Huawei” strategy was dying on its own. Freeberg’s counterpoint deserves equal weight, though. Chinese companies plainly want the chips even if their government does not, and a founder who believes technology lifts everyone could rationalize selling to Singapore or Vietnam while knowing where the hardware ends up. That is the version of events the Justice Department investigation will actually test, and it is the one the hosts were least eager to dwell on.

The most consequential outcome may not be who runs Nvidia, but who holds the off switch. Chamath reads a congressman’s post demanding that Nvidia change its practices and appoint leadership that puts national security first, and calls it the terms for this to end. Calacanis points to the bipartisan Chip Security Act and argues location tracking and remote disabling are technically trivial, the same mechanism that bricks a stolen iPad. The real fight is in the detail he flags: whether the company enforces it, or a government agency holds the keys. Ten days ago, as Chamath says, a GPS chip and a government-run kill switch on every GPU would have been unthinkable on civil liberties grounds. A weakened Nvidia, led by someone who needs Congress’s goodwill, is far more likely to design those controls in voluntarily. Whatever you think of the hosts’ theory of who leaked the memo, that shift in leverage is real and worth watching more closely than the succession race.

Sacks also makes the best case for why one executive matters this much. America’s lead is not a particular chip; it is the cadence of Ampere to Hopper to Blackwell to Vera Rubin, a new design every year, each several times more powerful than the last. Huawei’s CloudMatrix strategy, networking 384 weaker Ascend chips with abundant cheap power to approach what Nvidia does with 72, only works as a catch-up plan if the frontier slows. Losing “a tempo or two” is exactly the opening it needs. Chamath’s framing complements this: hobbling Nvidia does not make anyone else’s racks, networking, HBM integration or CUDA replacement better, so the whole industry simply goes slower. The Tim Cook comparison raised by Calacanis is useful here. Apple could coast on Steve Jobs’s roadmap because phones mature; GPUs in 2026 do not.

The hosts’ strongest defense of Huang is also their weakest argument. Sacks keeps returning to the line that the only known crime is the leak itself, and he is right that leaking classified intelligence is a serious offense that deserves an investigation. But Friedberg and Chamath supply the more likely mechanism for the resignation, and it has nothing to do with whether Alice Huang was a spy. Section 16 officers attest every quarter that they have no undisclosed conflicts with customers, partners or subordinates; directors and officers insurance depends on those attestations being true. Chamath then names the damage that exoneration cannot undo: more than a trillion dollars of market value gone in days, conflicting statements from Nvidia’s own PR, and plaintiffs’ lawyers already lining up. Even in the best case, where the spy claim collapses, “those eggs cannot be unscrambled.” The calls on the panel to bring Huang back as CEO or chairman underrate how little room a board has once the disclosure question is live.

The closing segment turns the scandal into a power struggle over AI itself. Reading a Will Manidis post about “jubilee mode” in safety-minded group chats, Sacks argues that the effective altruist movement is celebrating the fall of open source’s biggest funder while Anthropic, the company it is closest to, markets recursive self-improvement and winner-take-all economics on its IPO road show. It is a pointed observation about incentives, and the contradiction he identifies (fearing rapid takeoff while cheering the lab racing hardest toward it) is worth taking seriously. It is also a partisan read from someone who has argued with Anthropic for years, and the hosts themselves admit they cannot connect any of this to the leak. Treat it as a map of the factions now fighting over chips and models, not as evidence of who pulled the pin.

Key Takeaways

Jensen Huang resigned as Nvidia CEO after 33 years, with CFO Colette Kress named interim CEO; the hosts say the stock fell about 15% in the week before Nvidia’s earnings report.
The New York Times reported on November 6 that Alice Huang, an executive at Singapore-based data center company Megaspeed, was identified in a leaked intelligence memo as suspected of working for China. The Times had first written about the relationship in October 2025.
Sacks frames the story as “two ops”: possible chip smuggling through Megaspeed, which should be investigated, and the leak of a classified memo, which he calls the only crime known to have been committed.
He contrasts Huang’s treatment with Eric Swalwell’s, who received a defensive briefing about a suspected Chinese spy and kept his seat on the House Intelligence Committee.
Rep. Jim Himes, the top Democrat on the House Intelligence Committee, said after a classified briefing that he was not sure whether Alice Huang was a spy or a “useful idiot.”
Friedberg and Chamath believe the board acted over code-of-conduct attestations and D&O insurance covenants rather than any espionage finding.
Chamath argues neither China nor Nvidia’s rivals gain technologically; the only clear winners are those who want government-held telemetry and kill switches on chips.
The “Bank of Nvidia,” the capital and guarantees Nvidia extends to startups and data centers, ran through Huang personally and may slow or stop without him.
Chamath warns of 50 to 150 basis points of lost US growth, and later predicts a recession by 2028.
Sacks says China has blocked the H20 and kept out the H200 to prop up Huawei, making the hawks’ policy “effectively the policy of the CCP.”
Candidates floated for CEO include Ian Buck internally, and Arm’s René Haas and Palo Alto Networks’ Nikesh Arora externally; Friedberg and Calacanis think Huang could return if cleared.
Anthropic’s IPO targets 825 million shares at $70 to $78, about a $1.4 trillion valuation, plus a $26 billion private placement; Calacanis says Q3 revenue of $17 billion passed OpenAI’s $10 billion.
Chamath says roughly 80% of Anthropic’s IPO price will depend on the broader AI economy, and therefore on Nvidia, rather than on Anthropic itself.

Chapters

1:20 Six Days That Ended Jensen Huang’s Tenure

Calacanis lays out the timeline: the New York Times spy report on November 6, Huang’s companywide email two days later admitting an affair and announcing a board investigation, and a further Times report of a relationship with a subordinate. He notes the story had been in plain sight for more than a year, including a short thesis from Culper Research that named Megaspeed months earlier, and that the White House is now facing questions about why Megaspeed was left off the export control list.

6:11 Sacks’s Two Ops and Four Questions

Sacks calls the situation tragic for Huang, Nvidia and the United States, then asks four questions: who leaked the classified memo and why, why Huang never got the defensive briefing Eric Swalwell received, what national security secrets a private CEO could even hold, and whether Alice Huang was a spy at all. He argues chip IP is stolen through design files, not pillow talk, and that buying a Blackwell chip gives no one its blueprints. Chamath adds a fifth question: why now.

15:56 Open Access Versus Concentrated Power, and the Board’s Real Trigger

Friedberg describes two futures, one of broadly accessible AI and one where chips become the point of control, and says Huang’s advocacy for open source and global access put him at odds with powerful interests. He then argues the board’s likely trigger was a misrepresentation in quarterly code-of-conduct attestations, before recounting Nvidia’s decade of a flat stock price and its 20x run to the most profitable quarter in business history.

22:21 Who Benefits?

Chamath explains how auditor questionnaires and D&O insurance covenants turn a personal disclosure failure into a board crisis. He tests the claim that China wins and finds it lacking in both scenarios, says AMD, Google and startups still face years of work, and identifies one clear beneficiary: anyone seeking government-owned telemetry and kill switches. Calacanis argues the shareholders, not a defensive board, should decide Huang’s fate.

31:36 The Bank of Nvidia and the Market Impact

The hosts ask whether Nvidia’s role as financier to the AI economy survives without Huang, who had the moral authority to make fast investments. Sacks says China benefits strategically because Huang was a tentpole of America’s AI ecosystem, while Chamath sees growth risk on top of expected data center slowdowns. Sacks argues the yearly cadence of new chips is America’s real advantage, and that a slowed Nvidia mostly helps Huawei’s CloudMatrix strategy close the gap.

44:21 Kill Switches and the Fear of a Surveillance State

Friedberg warns that the new Congress could require every AI chip and data center to connect back to a state system, which he calls an Orwellian turn that Huang opposed. Chamath reads a congressman’s call for Nvidia leadership that puts national security first and asks which agency would get the keys. Calacanis explains the bipartisan Chip Security Act and the difference between company-enforced and government-held tracking.

48:22 A Short History of Chip Export Controls

Sacks traces controls from the Biden administration’s 2022 split between leading-edge and lagging chips to the Trump administration’s move to limited, licensed sales, and accuses Democrats of rewriting that history. He raises activist opposition to data centers, then makes his central claim that China itself is blocking Nvidia chips to build up Huawei, citing the latest DeepSeek model trained on Huawei hardware. Friedberg asks whether Huang knew chips were reaching China through intermediaries.

58:24 Who Should Run Nvidia Next?

Chamath favors a “wartime” technical capital allocator such as Ian Buck, René Haas or Nikesh Arora over a steward. Friedberg says Huang should return if cleared of breaking the law, and Calacanis floats Huang running SpaceX’s chip fab effort or Nvidia buying Palo Alto Networks. Chamath counters that shareholder suits over more than a trillion dollars in lost value will follow regardless of the investigation’s outcome.

66:24 Anthropic’s IPO Road Show

Calacanis runs through the terms and a bull case of $4.2 trillion. Chamath says the raise is crucial but its price will be set by the next wave of Nvidia news, while Friedberg hears investors buying the idea of a runaway recursive self-improvement engine. Calacanis calls Anthropic the “Microsoft killer” for bundling meetings, spreadsheets and legal tools into one subscription, and Sacks says the company is now openly pitching a monopoly.

77:22 Curing Disease, a Recession Call and the Safety Paradox

A clip of Dario Amodei predicting the end of cancer, Alzheimer’s and heart disease within a decade draws cautious agreement from Friedberg, who says his own biology work runs on Anthropic, and a recession forecast from Chamath. Sacks closes by arguing that safety groups celebrating Huang’s fall while backing the lab racing hardest toward superintelligence reveals an agenda about centralized control, not safety.

Notable Quotes

“The only crime that we know was committed for sure in all of this was the leaking of that classified memo.”
David Sacks, on the investigation he wants opened

“So, it’s the momentum. It’s the dynamism. People get way too hung up on some chip that was state-of-the-art 4 years ago.”
David Sacks, on where America’s chip advantage really lies

“It’s just the entire industry gets slowed down. That doesn’t benefit anybody.”
Chamath Palihapitiya, on the idea that Nvidia’s rivals win

“The policy of the China hawks is effectively the policy of the CCP. Neither one wants American chips in China.”
David Sacks, on Beijing blocking Nvidia’s H20 and H200

“Today, we look at things like cancer, Alzheimer’s, heart disease, mental illness. These will be relics of the past in I think less than a decade.”
Dario Amodei, in a clip played during the Anthropic IPO segment

“It is not a joke. It is not made up. It is not some far-flung idea.”
David Friedberg, on AI breakthroughs in biology

“Recursive self-improvement is the most dangerous thing anybody can be working on.”
David Sacks, on Anthropic’s IPO pitch

Watch the full conversation here.