All-In PodcastSpaceX’s $2T Case, Nvidia’s Shock Selloff, America Turns on AI, Trump Pulls AI Order, Bond Crisis?
CHAPTERS
Gavin Baker returns; setting the agenda in a hectic AI/markets week
Jason welcomes Gavin Baker as a guest co-host with Sacks out, teeing up a packed episode spanning Anthropic, SpaceX, Nvidia, macro risk, and US–China geopolitics. The group frames the show as a "big week" for both AI progress and investor sentiment.
Karpathy joins Anthropic: why elite talent and recursive improvement matter
The hosts discuss Andrej Karpathy’s move to Anthropic to lead a new pre-training team, using it to examine how top technical talent can create step-function progress. They focus on recursive self-improvement and continual learning as potential “frontiers” that could accelerate model gains.
Anthropic’s hypergrowth meets profitability: ROI narrative shifts
Gavin argues Anthropic’s reported profitability and high-margin inference changes the AI investment debate from speculative to return-driven. The group extrapolates the possibility of hundreds of billions in LLM ARR across labs and products, even before counting broader ad/recommender AI wins at hyperscalers.
“Covert” on-device AI and the emerging PR problem: framing vs. usefulness
A debate breaks out after Jason claims Chrome quietly shipped Gemini Nano; Friedberg pushes back on language that feeds a boogeyman narrative. They argue that breathless coverage of models fuels fear, and the better story is user outcomes (science, medicine, productivity).
Why Americans are turning on AI: inequality, psychology, and foreign influence
The panel explores reasons for rising backlash: fear of power concentration, unclear personal benefit timelines, and AI’s “anti-human” vibe. Friedberg also posits that foreign state actors historically exploit media narratives to slow rivals’ progress, while Gavin urges Americans in tech to advocate for AI’s upside.
Should we slow AI down? Regulation, US–China “KYC,” and the pulled Trump AI order
They discuss whether frontier models should be tested or governed, prompted by reports that a Trump executive order was drafted then pulled. Chamath favors limited guardrails like KYC-style controls to reduce catastrophic misuse, while Gavin warns about one-way regulatory ratchets and prefers liability/courts plus international alignment.
Jobs, robots, and safety: the politics of displacement vs. preventing deaths
The group contrasts warehouse/robot automation debates with self-driving, where safety improvements could create legal pressure to adopt autonomy. Chamath argues critics often fail to ask workers what they want, while Gavin predicts cities that ban autonomy will feel unsafe and inconvenient over time.
AI-enabled policing and privacy tradeoffs: Flock Safety, Cambridge vs. Vegas
A tangent on surveillance tech becomes a case study in local governance: Cambridge votes down gunshot detection while Vegas invests heavily in AI-enabled policing and rapid response. They argue crime reduction is increasingly a “choice,” and privacy can be protected via retention limits, audit trails, and constrained use.
SpaceX S-1 teardown and the $2T valuation case: Starlink + “Elon Web Services”
The episode shifts to SpaceX financials and valuation logic: Starlink as a cash engine, space launch as smaller and less profitable, and a major AI/compute business emerging via data center buildouts and large offtake deals. Gavin highlights SpaceX’s speed building gigawatt-scale data centers as a key competitive advantage.
xAI’s product stack: Cursor/Composer gains, Grok Build, and “harness” as moat
Gavin argues Cursor’s data and reinforcement learning cycles can drive outsized coding performance gains, and that agent “harnesses” (runtime, memory, integrations) are as important as raw models. With Grok Build, xAI gains a comparable harness to Claude Code/Codex, enabling faster iteration and broader utility.
Orbital compute and Starship: what must happen and when it hits the P&L
They debate feasibility and timelines for space-based data centers, noting an H100 has already been used in space and that SpaceX’s payload advantage expands with Starship. Gavin emphasizes rapid reusability as the true breakthrough, and estimates orbital compute becomes meaningful in the 2028–2030 window.
Nvidia earnings “shock selloff” context: share-loss narratives, CPU surprise, and financing GPUs
Despite extraordinary growth, Nvidia’s stock reaction is framed as skepticism, narrative battles about ASIC competition, and mega-cap “stairstep” reratings. Gavin argues comparisons often ignore segmentation and China effects, notes frustration that competitors avoid benchmarks, and highlights Nvidia’s unexpected $20B CPU business as evidence of deep co-design leverage.
Macro and “bond crisis?”: inflation, yields, Japan carry risk, and AI as countervailing force
Jason lists warning signals: rising oil, inflation expectations, and global bond yields, prompting Friedberg’s “credit crisis” concerns tied to Japan’s yields and carry trades. Gavin argues multiple truths coexist—rates are a risk, but AI fundamentals are strengthening and the US remains relatively advantaged via energy and tech leadership.
US–China trip outcomes: symbolism vs. substance, chips policy, and avoiding the Thucydides Trap
They assess whether the CEO-heavy China trip produced a “grand deal” or mostly optics, with Friedberg skeptical of any decisive de-escalation. Chamath believes closed-door alignment matters most, while Gavin argues continued dialogue is inherently beneficial and discusses strategic logic around selling older GPUs to China to shape ecosystems and reduce destabilizing paths.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome