All-In PodcastNew SEC Chair, Bitcoin, xAI Supercomputer, UnitedHealth CEO murder, with Gavin Baker & Joe Lonsdale
CHAPTERS
- 0:00 – 4:14
Sacks’ New Role and Guest Lineup Introduction
Jason and Friedberg open by announcing that David Sacks has been appointed the White House AI and Crypto ‘Czar’ under Trump, explaining why he and Chamath are absent. They set expectations that the episode was recorded before that news, then introduce guest co‑hosts Gavin Baker and Joe Lonsdale.
- •David Sacks appointed to guide U.S. AI and crypto policy and chair PCAST (Presidential Council of Advisors on Science and Technology).
- •Show recorded before the appointment news, so discussion may not reference it.
- •Guest hosts: Gavin Baker (Atreides, public and private markets) and Joe Lonsdale (8VC, co‑founder of Palantir, Addepar, OpenGov).
- •Light banter about Jason staying at Sacks’ house and the pod’s continuity despite changes in besties’ schedules.
- 4:14 – 18:50
The ‘Trump Bump’: Deregulation, Debt Spiral, and America’s Advantages
Baker, Lonsdale, and Friedberg analyze market optimism after Trump’s election, focusing on deregulation, government efficiency, and the national debt. They argue that simplifying rules and cutting bureaucratic bloat could unlock enormous growth, comparing America’s situation to a mismanaged monopoly finally getting competent leadership.
- •Baker likens Trump’s opportunity to Satya Nadella taking over Microsoft: stop ‘really dumb’ policies to unlock latent strength.
- •U.S. structural advantages: geography, resources, self‑sufficiency in food and energy, but policy‑driven underperformance (California as prime example).
- •Lonsdale describes federal bureaucracies as dysfunctional ‘zombie companies’ kept alive with ever more money and staff.
- •Historical note: 100+ years of meritocratic civil‑service exams eroded in the 1970s, then further degraded by over‑protection and identity‑based hiring.
- •Proposal: sunset clauses plus data‑driven, difficult renewal processes for regulations to reverse regulatory ‘cancer’.
- •Friedberg ties government inefficiency and overspending to the looming debt death spiral and outlines a strategy: deregulate to get GDP growth to ~4–5% while cutting deficits.
- 18:50 – 20:07
Energy, Electricity Capacity, and Beating China
The conversation shifts to energy as the fundamental driver of prosperity, security, and AI capacity. Friedberg, Lonsdale, and Baker argue for an aggressive U.S. nuclear build‑out to counter China’s rapid electricity expansion and to support domestic manufacturing and AI scaling.
- •Friedberg highlights China’s projected move from ~2 to ~8 TW of capacity vs. U.S. ~1 to ~2 TW, calling this the most overlooked strategic metric.
- •Lonsdale links cheap power to middle‑class prosperity, industrial competitiveness, and AI deployment; cites Vaclav Smil’s work on energy and civilization.
- •U.S. regulatory structure effectively blocks new nuclear and makes large‑scale power build‑out uncompetitive vs. China.
- •Baker calls nuclear the most environmentally friendly baseload source available now and expects solar + batteries to dominate in ~50 years but not soon enough.
- •Examples: Texas as top U.S. solar state purely due to ease of building, not environmental ideology; Austin housing as evidence of what pro‑build policies can do.
- •Consensus that ‘unlimited energy’ is technically within reach but politically and bureaucratically blocked by ‘midwits’ and over‑cautious regulators.
- 20:07 – 33:00
New SEC Chair Paul Atkins, Crypto, and Capital Market Rules
With Trump nominating Paul Atkins to replace Gary Gensler at the SEC, the hosts debate what a more deregulatory, pro‑crypto Chair means for markets. They contrast Atkins’ investor‑choice rhetoric with Gensler’s enforcement‑heavy style, and wrestle with how far to go in loosening restrictions on private markets.
- •Lonsdale vouches for Atkins as smart, fair, pro‑innovation, and opposed to ‘gotcha’ enforcement tactics that characterized Gensler’s crypto approach.
- •Criticism of Gensler’s refusal to clearly define crypto rules and his selective engagement (e.g., access for SBF vs. Coinbase’s Brian Armstrong).
- •Baker emphasizes that U.S. capital markets are the world’s best; any reforms must preserve trust (e.g., accurate financials, limits on insider trading).
- •Libra/Facebook case: bipartisan political pressure (including mafia‑like threat letters) effectively killed the project despite its potential for cheaper remittances and payments.
- •Underlying state concern: loss of control over money supply and monetary policy if crypto gains too much ground.
- •Friedberg defends the SEC as one of the strongest global regulators, warning that fully opening private markets to unsophisticated investors could turbo‑charge fraud.
- •Debate around accreditation: Atkins’ 2007 speech criticizing wealth tests as paternalistic and exclusionary; suggestions for sophistication exams instead of net‑worth thresholds.
- 33:00 – 39:10
Bitcoin, BRICS, and the Future of Dollar Dominance
The panel parses Bitcoin’s explosive rally, Trump’s rhetoric, and whether crypto or BRICS pose a bigger threat to U.S. monetary power. They distinguish Bitcoin’s role as ‘digital gold’ from speculative alt‑coins and consider Bitcoin as both a check on U.S. fiscal mismanagement and a potential long‑term challenger to the dollar.
- •Friedberg calls Bitcoin a fundamental threat to the dollar in the long run, contrasting it with Trump’s pledge to punish BRICS currency experiments while embracing BTC.
- •Consensus that Bitcoin is trending toward a ‘digital gold’ store‑of‑value safe haven, not a near‑term transactional currency.
- •Lonsdale: BRICS currency coordination would be an authoritarian, anti‑U.S. development and rightly opposed; Bitcoin by contrast is a pro‑liberty check on excesses of the state.
- •Baker doubts BRICS can displace the dollar due to U.S. rule of law, but believes Bitcoin could eventually be a ‘serious threat’ to the dollar’s dominance.
- •Bitcoin is framed as a market‑driven discipline mechanism: if deficits and debt remain out of control, capital can flee to BTC, imposing an external check on policymakers.
- 39:10 – 48:00
Access to Private Markets: Who Should Be Allowed to Invest?
Using Atkins’ 2007 speech on accreditation as a springboard, the group debates whether restricting private fund access to the wealthy is paternalistic or necessary consumer protection. They explore private equity vs. venture risk profiles, fraud risks, and potential test‑based alternatives to wealth thresholds.
- •Atkins’ argument: extra $2.5M ‘investments’ hurdle for hedge/PE funds presumes non‑rich investors are unsophisticated and denies them upside participation.
- •Baker notes that post‑2007, private equity delivered huge returns; if ordinary investors had access, they might have materially benefited.
- •Friedberg counters that $20T+ in public equities already provide wealth‑building opportunities; opening opaque private markets broadly would invite predatory marketing and fraud.
- •Examples: FTX as a case where even sophisticated investors failed on diligence; fraud and non‑GAAP reporting in venture make it risky for unsophisticated investors.
- •Lonsdale points out the irony that ‘the left’ equates wealth with sophistication; suggests this is insulting to capable but less‑wealthy individuals.
- •Floating solution: rigorous but accessible sophistication tests or education modules (akin to a driver’s license) as an alternative to pure wealth tests.
- 48:00 – 49:07
Saylor, Leverage, and Defense‑Tech’s New ‘Primes’
The discussion pivots to Michael Saylor’s MicroStrategy Bitcoin strategy and then to the state of U.S. defense technology. Baker warns of structural risks in Saylor’s leverage model, while Lonsdale outlines a nascent ecosystem of ‘new primes’ reshaping warfare through drones, autonomy, and electronic warfare.
- •Baker explains Saylor’s play: issuing convertible debt to buy Bitcoin on the assumption BTC keeps rising; questions sustainability given MicroStrategy’s ~$400M revenue and ~$75M interest cost.
- •Concern that, at scale, debt investors may not fully trust BTC collateral, limiting leverage and breaking the ‘magic money machine’ narrative.
- •Lonsdale, bullish on Bitcoin overall, urges investors to understand MicroStrategy’s leverage risk instead of blindly chasing the trade.
- •Defense tech: Lonsdale describes a shift toward swarms (air, land, sea), AI‑driven coordination, and electronic warfare (e.g., Epirus) as the new center of gravity.
- •He expects 7–10 ‘new primes’ (e.g., Anduril, Siren/Sironic, Epirus) to emerge rather than hundreds of winners, predicting many zeros for late‑stage or undisciplined capital.
- •China’s posture has already triggered a strategic shift in the U.S.; still, ~95% of defense spending remains tied to legacy, cost‑plus programs like carriers and F‑35s.
- •Lonsdale wants incremental dollars flowing to autonomous systems and smart munitions rather than more legacy platforms.
- 49:07 – 57:00
xAI’s Colossus: GPU Coherence, Scaling Laws, and the Grok‑3 Bet
Baker gives a technical deep dive into xAI’s Memphis supercomputer and why making 100k+ GPUs coherent is such a big deal. He frames Grok‑3’s training run as the first true test of whether training‑time scaling laws still hold at unprecedented compute levels, with major implications for OpenAI, Microsoft, and AI strategy broadly.
- •Coherence concept: GPUs must ‘know’ what others are doing to act like one giant accelerator; historically ~25–32k H100s was viewed as a hard ceiling.
- •xAI’s achievement: a radically redesigned data‑center architecture enabling coherence across ~100k H100s, reportedly scaling to 200k and targeting 1M GPUs long‑term.
- •Networking stack: combination of NVIDIA NVLink/NVSwitch and InfiniBand/Ethernet; Baker notes Llama 3.1’s paper renewed interest in Ethernet‑based clusters.
- •Grok‑3 is now training on this cluster in Memphis (converted Electrolux factory), making it the first real data point beyond GPT‑4 on whether increasing training compute still yields major capability gains.
- •Baker discloses Atreides is an investor in xAI and expects Grok‑3 to be ‘significantly’ more capable if scaling laws remain valid.
- •Implication: xAI’s breakthrough helped NVIDIA by prompting buyers to pull forward H100 purchases while Blackwell slipped, reducing any near‑term demand air pocket.
- 57:00 – 57:40
New Scaling Dimensions: Test‑Time Compute, Context Windows, and ROI
Beyond brute‑force training, the group explores new axes of AI progress like test‑time compute, longer context windows, and smarter system architecture. They also confront the question of AI’s economic ROI and whether massive GPU investments are justified or a hype‑driven bubble.
- •Test‑time compute (‘inference scaling’): letting models ‘think longer’ on hard problems (more passes, more tokens) measurably boosts apparent IQ, akin to giving a person more time on a difficult exam.
- •Models‑of‑models: many application startups already chain multiple models, using cheap ones first and more powerful ones for verification or complex tasks.
- •Context windows are rapidly expanding, allowing whole books or corpora to be loaded and queried; performance and latency at large context are now major research areas.
- •Hardware gains: H200s deliver roughly 50% more effective compute per watt than H100s due to more memory and efficiency; Blackwell will bring another architectural leap.
- •ROI evidence: public AI spenders’ ROIC has sharply improved, especially Meta and Google via better ads and internal automation; startups are building profitable thin‑wrapper tools that directly replace labor.
- •Baker frames big‑tech AI capex as a prisoner’s dilemma: even if marginal ROI falls, fear of losing the ASI race and existential risk to laggards ensures continued massive spending.
- 57:40 – 1:06:00
China, Chips, and Strategic Export Controls
After Lonsdale departs, Baker and Jason analyze U.S. efforts to kneecap China’s AI progress via chip and networking export controls. Baker likens U.S. policy to a ‘Three‑Body Problem’ style SOFA field over China, while warning that necessity could spur indigenous innovation if Washington miscalculates.
- •Baker observes that despite restrictions, recent Chinese models remain close behind U.S. leaders—evidence of strong domestic talent and improvisation.
- •As NVIDIA’s Blackwell and other advanced chips arrive, Baker expects China will be unable to keep pace without access, making the U.S. export regime a very aggressive but potentially effective strategy.
- •Discussion of rare earths and materials: the U.S. has ample reserves but offshored messy refining; regulatory constraints and cost disadvantages are the real bottlenecks.
- •Example: gallium nitride is vital for advanced systems (including defense tech), but refining constraints limit domestic production; again, regulation more than geology is the problem.
- 1:06:00 – 1:07:56
AI in the Trenches: Developer Tools, Copilots, and Human‑Language Programming
In a lighter segment, Friedberg and Baker share hands‑on experiences with AI tooling like Zoom’s AI summaries, Notion AI, and the Cursor coding environment. They argue we’re quickly approaching a world where natural language becomes the dominant programming language and non‑developers can reliably build production apps.
- •Friedberg praises Zoom’s AI meeting summaries and Notion’s AI features as real productivity wins, not hype.
- •At an internal hackathon using Cursor, many non‑engineers built working internal tools end‑to‑end, then deployed them into production.
- •Today AI coding is ~70–80% there: humans still debug and handle deployment, but the trajectory suggests near‑full automation of boilerplate, testing, and UX iterations.
- •Baker notes his own experiments (e.g., trying to build a multiplayer Skyrim mod with AI coding help) and predicts that by next year, ‘human language will be the dominant programming language.’
- •They foresee AI handling QA, UX design, iteration, and even sales/support, allowing users to specify apps in plain English and receive fully functioning products hours later.
- •Macro point: startups adopt such technologies first due to resource constraints, reinforcing AI’s role as a lever for doing more with fewer people.
- 1:07:56 – 1:24:19
UnitedHealth CEO Murder, Healthcare Rage, and Moral Hazard of ‘Oppressor’ Narratives
The show ends on the shocking assassination of UnitedHealthcare’s CEO Brian Thompson in Manhattan. The hosts condemn social‑media celebrations of his death, unpack the anger toward insurers, and warn about a cultural shift that justifies violence against perceived ‘oppressors’ while absolving the ‘oppressed’ of moral responsibility.
- •Facts: CEO of a major U.S. health insurer shot outside a NYC hotel; shell casings reportedly inscribed with ‘deny, defend, depose,’ echoing critical literature about insurer tactics.
- •Speculation that the shooter or hirer might be connected to a denied claim that led to a loved one’s death, but no confirmed motive.
- •Friedberg: crucial distinction between holding CEOs legally liable for fraud/negligence vs. personal retaliation for corporate outcomes; if CEOs are directly liable for every bad outcome, no one will run essential but risky businesses.
- •Baker notes UnitedHealth’s ~85% medical loss ratio and warns that viral charts about denial rates may be misleading (e.g., initial vs. final denials).
- •They connect celebratory reactions to the same mindset that justified Hamas atrocities: a rigid oppressor/oppressed frame where oppressors ‘deserve’ unlimited harm and oppressed actors can ‘do no wrong.’
- •Concern that this moral framework, amplified by social media, is corroding basic norms about individual dignity, due process, and rule of law.
