Jensen On The Ropes, Sam Altman’s Conflicts, Allbirds’ GPU Pivot
TL;DR
Jensen Huang looked least convincing on Nvidia’s two hardest questions: competition and China — Alex Kantrowitz argues Huang never cleanly explained why Nvidia’s moat holds if Gemini and Claude were trained on Google TPUs/Amazon Trainium, or how selling more compute into China doesn’t raise cyber risk for the U.S.
The strongest Nvidia defense was there — but Alex had to articulate it better than Jensen did — Ranjan Roy lands on the clearest version of the moat: Nvidia’s advantage is CUDA, supplier relationships like TSMC, and sheer capacity, while Anthropic’s biggest criticism is still a lack of capacity.
Anthropic’s Claude “Mythos” is shifting from PR stunt to government priority — the hosts note an AI Security Institute eval claiming Mythos was the first model to complete a simulated 32-step corporate network attack end-to-end, and Axios reports Dario Amodei meeting White House chief of staff Susie Wiles.
A thaw between Anthropic and the Pentagon now looks likely — after Alex interviewed Pentagon official Emil Michael, he came away thinking the Defense Department is effectively “Claude-pilled” and increasingly open to reconciling with Anthropic after earlier tensions.
The real Wall Street Journal scoop on Sam Altman wasn’t his side deals — it was succession chatter — the headline focused on conflicts around bets like Helion and Stoke Space, but Alex zeroes in on investors privately floating OpenAI board chair Bret Taylor as a possible replacement ahead of a future IPO.
Allbirds’ absurd GPU pivot says more about markets than infrastructure — after collapsing from a roughly $4 billion public-market high, Allbirds was acquired for about $39 million, then repurposed as an AI compute story that briefly sent the stock up 582%, which Ranjan calls a perfect “dead beloved brand becomes GPU company” meme trade.
The Breakdown
Jensen’s awkward Dwarkesh moment
The show opens on Jensen Huang’s unusually prickly appearance with Dwarkesh Patel, where he dropped lines like “we are not a car” and went off on a “loser premise” that Dwarkesh never actually used. Alex’s take is that the heat wasn’t the story — the bigger issue was Huang struggling to give a coherent answer on Nvidia’s future against custom chips and export controls.
Nvidia’s moat: chips, stack, or just capacity?
Dwarkesh’s core challenge was simple: if Claude and Gemini were trained on TPUs/Trainium, what exactly keeps Nvidia on top? Jensen kept returning to CUDA, the full stack, and decades-long supplier relationships, while dismissing Anthropic as “a unique instance, not a trend,” but Alex wasn’t buying the clarity of that defense. Ranjan, playing translator, says the cleanest version of Jensen’s argument is just that Nvidia can reliably deliver capacity at scale — and Anthropic’s own weak spot is that it often can’t.
The China answer that never quite landed
On export controls, the hosts think Jensen looked even shakier. Dwarkesh pressed him to admit that giving China more compute could speed up models with real cyber-offense potential, especially as U.S. institutions are currently evaluating Claude Mythos before broad release. Instead, Jensen bounced between “selling American technology is beneficial” and broad cooperation rhetoric, which left Alex saying he was trying to be both public-company CEO and national-security minister at the same time.
The better case for Nvidia in China
Alex then lays out the stronger version of the argument Jensen could have made: AI models are also vehicles for cultural soft power. If the U.S. shuts China out completely, it may create a separate open-source ecosystem where Chinese values spread globally through models used in places like India and Africa; if Nvidia stays involved, American tooling and norms keep shaping the stack. It’s more nuanced than “sell everything” or “ban everything,” and the hosts agree that nuance was exactly what was missing from Huang’s performance.
Mythos goes from marketing spectacle to Washington leverage
Last week they joked about Anthropic’s rollout theater; this week they concede the capability claims are getting harder to dismiss. Alex cites an AI Security Institute result saying Claude Mythos Preview was the first model to complete a 32-step simulated corporate network attack that would take a human expert about 20 hours, and says that “spooky” performance is changing how both companies and governments treat Anthropic.
Anthropic and the Pentagon may be making peace
That shift is now visible in Washington. Alex points to Axios reporting that Dario Amodei met with White House chief of staff Susie Wiles, while a source framed access to Mythos as something the U.S. government would be “grossly irresponsible” to ignore — a line Alex jokes sounds straight out of Anthropic comms. He also says Pentagon official Emil Michael opened the door to reconciliation, and his own read is that the Pentagon likes Anthropic’s tech more than the politics around it.
Alex explains his interview style
Before the break, Alex pauses to address a listener accusing him of being too deferential to powerful guests. His answer: the goal is to get real decision-makers in the room and ask the hardest possible questions without turning the interview into journalist theater or a “did you order the code red?” performance. Ranjan backs him up, saying the value is in pointed but non-combative questioning, especially in an era when too many podcast interviews are softballs.
Sam Altman’s conflicts — and the succession subtext
On the Wall Street Journal’s Sam Altman piece, Alex says the headline about side hustles buried the real news: some OpenAI investors are already floating board chair and former Salesforce co-CEO Bret Taylor as a possible successor. The hosts see it as a sign that OpenAI still lacks the mature operating structure investors want before an IPO, even as Altman keeps doing the part no one else can — raising eye-watering sums, including a recent $122 billion raise at an $852 billion valuation.
Allbirds’ GPU pivot is the most 2026 story imaginable
They close on pure chaos: Allbirds, once a DTC darling beloved by VCs and bankers in the 2021 era, has now pivoted into AI compute infrastructure. Alex reads Matt Levine’s killer joke that tech buyers might actually trust a former sneaker company more if it’s called Allbirds, because everyone in tech used to wear them. Ranjan, who watched Allbirds from the inside of the DTC boom, says the brand nostalgia itself is the asset — a perfect shell for a hilariously unserious but very marketable GPU story.