All-In PodcastThe AI Cold War, Signalgate, CoreWeave IPO, Tariff Endgames, El Salvador Deportations
CHAPTERS
- 0:00 – 5:00
Intro, Banter, and Guest Setup: Nvidia, CoreWeave, and Gavin Baker
The episode opens with light-hearted banter about dogs and then introduces investor Gavin Baker from Atreides Management. The hosts set up the main topics: Nvidia’s dominance, the looming CoreWeave IPO, and broader AI and macro themes. Baker’s long relationship with Nvidia CEO Jensen Huang is highlighted to frame his credibility on semiconductor and AI infrastructure questions.
- •Humorous cold open about Friedberg’s dog and Chamath’s dogs sets a casual tone.
- •Gavin Baker (Atreides Management, ~$4B AUM) is introduced as a returning guest.
- •CoreWeave’s impending IPO and Baker’s longstanding relationship with Jensen Huang are teed up.
- •The table is set for a deep dive into Nvidia, AI infrastructure, and capital markets.
- 5:00 – 15:00
Nvidia’s Balance Sheet, Hopper-to-Blackwell Transition, and ‘Round Tripping’ Concerns
The group drills into investor worries about Nvidia’s rising accounts receivable and supposed ‘round tripping’ via stakes in CoreWeave and neo-clouds. Baker argues that Nvidia’s revenues would be similar without those investments and that the AR spike is tied to a once-in-a-generation GPU upgrade. They explore the physical and operational magnitude of moving from Hopper to Blackwell and why it understandably distorts working capital.
- •Baker dismisses the idea that Nvidia’s equity in CoreWeave materially props up demand; hyperscalers would have simply bought more GPUs.
- •Nvidia sought to reduce customer concentration by enabling neo-clouds and pushing reference server designs.
- •Hopper-to-Blackwell is described as the largest semiconductor product transition ever, with huge changes in weight, power, and cooling.
- •Accounts receivable ballooned because revenue was recognized ahead of customers fully deploying the new systems.
- •If AR normalizes after the July quarter, concerns fade; if not, then it’s a real issue.
- 15:00 – 40:00
CoreWeave IPO, GPU Clouds, and Why Running Clusters Isn’t a Commodity
The conversation shifts to CoreWeave’s IPO, its heavy leverage, customer concentration, and market skepticism that GPU cloud capacity is a commodity service. Baker argues that orchestrating giant GPU training clusters is operationally rare and difficult, likening it to scaling a national retail chain. The group compares market misperceptions around AWS in its early days and notes CoreWeave’s strategic acquisitions and long-term contracts.
- •CoreWeave is raising ~$1.5B at a reported ~$23B valuation, with ~$2B in revenue and ~$8B in debt.
- •Over 60% of revenue comes from Microsoft; Nvidia both supplies GPUs and owns ~6% of CoreWeave.
- •Baker asserts that building and running massive GPU clusters is a scarce, non-commodity capability.
- •Historical analogies to AWS and Azure underscore that Wall Street has mispriced such models before.
- •CoreWeave’s acquisition of Weights & Biases is framed as a differentiator that tightens its AI tooling stack.
- 40:00 – 52:00
AI Agents, MCP, and the Future of Work and Incumbents
The hosts explore the emerging ‘agentic’ paradigm via Manus and the Model Context Protocol (MCP) from Anthropic, which OpenAI has adopted. Baker sees agents as the main near-term driver of AI ROI and GPU demand, while Friedberg emphasizes their potential to unlock complex projects with tiny teams. Chamath zooms out to argue that agents could enable three-person teams or Chinese startups to outcompete massive American incumbents, especially in SaaS.
- •Manus is cited as an early ‘ChatGPT moment’ for functional agents performing real tasks.
- •MCP is emerging as a standard for LLMs to safely and easily call external tools like Stripe.
- •Baker notes agents will be constrained by compute, justifying high prices like $2,000–$20,000 per agent per month.
- •Friedberg argues agents will enable radically complex projects (biotech, infrastructure, engineering) with minimal human staff.
- •Chamath warns this will hyper-disrupt American incumbents, especially software firms, by collapsing labor costs for challengers.
- 52:00 – 1:01:00
AI Cold War: Export Controls, China’s Catch-Up Odds, and Global Competition
The discussion turns explicitly geopolitical: new U.S. export controls are targeting Nvidia’s ecosystem to slow Chinese AI progress. Baker likens GPU smuggling to drug smuggling, suggesting total enforcement is unrealistic and that friction may merely accelerate Chinese self-reliance and algorithmic innovation. He assigns low near-term odds to China matching top U.S. GPUs but flags serious longer-term risk given divergent time horizons between the U.S. and CCP.
- •New U.S. export controls aim to prevent next-gen Nvidia GPUs from reaching China directly or via third countries.
- •Baker states it’s practically impossible to fully stop high-value GPU smuggling, paralleling drug interdiction challenges.
- •Restrictions push China to build its own semiconductor stack and spur algorithmic breakthroughs like DeepSeek.
- •He sees almost zero probability of full Chinese hardware parity in five years, but meaningful uncertainty over ten.
- •The panel notes China’s long planning horizon (decades) versus the U.S.’s election-cycle mindset.
- 1:01:00 – 1:18:00
Tariffs, Reshoring, and the ‘Grand Economic Experiment’ to Fix the U.S. Economy
The besties dissect Trump’s tariff-centric strategy to rebalance trade, revive U.S. manufacturing, and lift working-class incomes, anchored by recent long-form interviews with Howard Lutnick and Scott Bessent. They discuss how tariffs, tax cuts for under-$150k earners, deregulation, and aggressive cuts to waste and military spending are meant to interlock. Baker views the approach as coherent but high-risk; the group stresses that missteps could trigger recession or inflation and blow out the deficit.
- •Chamath frames tariffs as a ‘level-setting’ tool to mirror other countries’ tariffs, not an arbitrary weapon.
- •Trump has been consistent on trade imbalances since at least the 1980s (e.g., a 1987 Larry King clip).
- •Friedberg articulates the need to pair higher tariffs (raising prices) with tax cuts (raising net income) and deregulation (lowering costs) to keep overall demand stable.
- •Baker cites a projected ~$1.9T deficit and notes ‘sharp money’ expects only ~$200–250B from tariffs, implying massive required spending cuts.
- •They highlight Elon's DoJ/“DOGE” team’s mandate to root out up to ~$1T in waste, fraud, and abuse, plus potential military spending reductions.
- 1:18:00 – 1:37:00
Execution, Social Security, and Media Spin on Deficit Reform
The group defends Lutnick’s comments on Social Security against what they see as bad-faith media clipping and misrepresentation. Baker clarifies Lutnick’s actual point—cutting fraud to avoid raising the retirement age—and shares a personal anecdote about unemployment fraud. They emphasize that both parties’ inspector generals have identified hundreds of billions in waste, arguing for systematic reform. Jason riffs on doing a ‘DOGE pod’ to bring more transparency to deficit-cutting efforts.
- •Lutnick’s actual stance was to preserve the current retirement age by eliminating waste and fraud, not cutting benefits.
- •Media outlets are criticized for selectively editing quotes to suggest the opposite policy intent.
- •Baker notes bipartisan inspector-general reports documenting massive waste and fraud in federal programs.
- •Personal stories of identity theft and fraudulent unemployment claims illustrate systemic weaknesses.
- •The group urges regular, audited ‘DOGE reports’ to build public trust around deficit-cutting measures.
- 1:37:00 – 1:57:00
Signalgate: Secure Comms, FOIA, and the Limits of Transparency
They unpack ‘Signalgate,’ where The Atlantic’s Jeffrey Goldberg was accidentally added to a high-level Signal group discussing strikes on Houthi targets. Chamath contextualizes the urgency: Houthi disruptions in the Red Sea were spiking shipping costs and threatening global trade, with the U.S. effectively the only power able to intervene. The ensuing debate covers whether officials should use apps like Signal, the role of desktop clients as attack vectors, FOIA obligations, and the value of candid, non-recorded deliberations in government.
- •Houthi attacks in the Red Sea drastically cut Suez Canal traffic, raising shipping costs and recession risk, especially for Europe.
- •The U.S. Navy was uniquely capable of securing the route; allied navies lacked sufficient capacity.
- •Chamath criticizes the use of Signal desktop clients, which multiply security attack surfaces.
- •Friedberg reviews the Federal Records Act and FOIA, noting ongoing litigation over private channels vs. archival obligations.
- •He argues that not all internal deliberations should be FOIA-able, or candid decision-making will be chilled.
- •Jason insists high-level war planning should never occur on personal devices and stresses the hypocrisy given past Republican attacks on Hillary’s email practices.
- •Baker notes the leaked content actually reveals continued U.S.–Europe alignment: the U.S. acts primarily for Europe’s benefit and still views them as ‘our side of the ledger.’
- 1:57:00 – 2:16:00
El Salvador Deportations, CECOT Prison, and the Ethics of Ends vs. Means
In the final major segment, the besties analyze the Trump administration’s deportation of 238 alleged gang members to El Salvador’s infamous CECOT prison under the 1798 Alien Enemies Act. They highlight documented cases of misidentification based on tattoos and social media posts and discuss the tension between drastically reducing crime and upholding American commitments to due process and human rights. The segment closes with consensus that poor execution and sadistic optics—like gloating prison videos—could severely damage the administration’s broader agenda.
- •238 alleged Venezuelan gang members were deported without traditional U.S. due process to a foreign supermax known for torture and harsh conditions.
- •Several high-profile cases suggest clear misidentification: a pro soccer player with a Real Madrid tattoo, a shoe salesman covering a scar, and a gay stylist fleeing persecution.
- •A U.S. district judge has paused further deportations, but the administration appears to be continuing and publicly celebrating them.
- •Friedberg opposes any regime of detention without due process, comparing it unfavorably even to shipping suspects back for trial in their home countries.
- •El Salvador’s crackdown reduced homicide from 103 to ~2 per 100,000, illustrating powerful but morally fraught ‘ends.’
- •Baker and Calacanis stress that innocent people in such prisons are morally unacceptable and politically self-defeating; acknowledging mistakes and adding legal safeguards would preserve legitimacy.
- •Chamath reframes this as a trolley-problem-like policy choice that voters effectively empowered the president to make, but agrees that mechanisms to rectify errors are essential.
- 2:16:00
Wrap-Up, Meta-Conversation, and Besties’ Dynamic
The episode ends with a mix of sharp disagreement and affectionate ribbing among the hosts, underscoring their willingness to air real policy differences in public. Jason leans into his self-described ‘virtue signaling’ on human rights, while Chamath and Gavin reiterate that execution and course correction will determine whether the administration’s bold experiments succeed. The closing banter reinforces the show’s blend of high-stakes policy analysis and irreverent friendship.
- •Jason reiterates his Amnesty International background and strong stance on due process and human rights.
- •Baker frames much of the policy debate as ends vs. means, insisting that human rights remain a non-negotiable American value.
- •The hosts acknowledge they’re only ~4.5% into the term and that early errors must be quickly fixed.
- •They close with typical All-In banter, teasing Jason as the ‘world’s greatest virtue signaler’ while affirming genuine camaraderie.
