All-In PodcastIPOs and SPACs are Back, Mag 7 Showdown, Zuck on Tilt, Apple's Fumble, GENIUS Act passes Senate
CHAPTERS
- 0:00 – 13:00
Cold Open, Banter, and East Meets West Conference Recap
The episode opens with the usual comedic banter about babies, steaks, and inside jokes among the besties, before shifting to a recap of Thomas Laffont’s East Meets West conference. They touch on LA’s sluggish post‑COVID restaurant recovery and the structural issues in LA’s economy versus tech‑driven San Francisco.
- •Lighthearted intro: jokes about Friedberg’s baby, Long Hill Wagyu, and the pod’s influence on niche steak supply.
- •Thomas describes East Meets West: AI dominated the agenda across SaaS transformations, infrastructure, and high‑profile moves (Zuck/Scale, Nat Friedman rumors).
- •Discussion of LA’s weak restaurant recovery: ~50% behind national per‑location recovery; factors include entertainment industry decline and filming moving to New York, UK, Georgia, Las Vegas, Toronto, Saudi Arabia.
- •Regulation and cost make California uncompetitive for filming and restaurants; MrBeast/Beast Games are cited as an example of productions actively avoiding California.
- 13:00 – 23:00
AI Productivity, Onshoring, and Real‑World Use Cases
The group explores whether AI‑driven productivity could materially boost U.S. GDP and counter debt concerns, with a focus on real use cases across medicine, law, coding, and local services. They highlight how AI can increase throughput for doctors and small businesses, driving both lower costs and higher overall economic activity.
- •Thomas outlines a bullish AI‑GDP thesis: AI‑driven productivity might justify lower‑than‑expected interest rates by expanding GDP faster.
- •Open Evidence example: AI diagnostic engine used by roughly a third of U.S. physicians, especially in oncology, with frequent daily use to assist diagnoses.
- •Friedberg’s preventative‑care argument: if AI boosts doctor throughput 10x, per‑visit costs can fall while total diagnostic revenue and access expand, increasing GDP.
- •LA dentist VO3 ad story: a cheap AI‑generated viral video floods a local implant practice with patients, illustrating AI as a marketing and leverage tool for small businesses.
- 23:00 – 35:00
Zuck ‘On Tilt’: Meta’s AI Talent War and Scale AI Deal
Jason frames Zuck as ‘tilted’ and scrambling to catch up in AI, citing rumors of $100M signing bonuses and Meta’s $14B+ stake in Scale AI. The panel dissects these moves as a rational response to an existential AI threat and likens the Scale deal to Facebook’s old Onavo acquisition that cut competitors off from key data.
- •Sam Altman clip: claims Meta offered top OpenAI staff ~$100M signing bonuses plus ~$100M per year in comp; he says none of their best people have left so far.
- •Meta invests >$14B in Scale AI for ~49% stake; described as quasi–acquihire to secure training data pipelines and talent while avoiding direct acquisition scrutiny.
- •Scale’s business: high‑end data labeling, including sophisticated reasoning datasets—not just simple ‘this is a dog’ labels.
- •Comparison to previous ‘shadow acquisitions’: Microsoft–Inflection, Google–Character.ai, Amazon–Adept, and Meta’s earlier Onavo buy, which removed a critical analytics tool from market.
- •Thomas calls Meta’s moves rational: if 50% of Meta’s $1.7T market cap is at AI‑risk, spending a few percent to improve odds of winning is justified.
- 35:00 – 45:00
The Stack of Secrets: Why Hardware–Model Coupling Matters
Chamath lays out a framework for AI advantage as a ‘stack of secrets’—labeling/training, model/app layer insights, and deep integration with compute hardware—and argues Meta is only now starting to acquire the first two. He emphasizes that winners like OpenAI, Anthropic, DeepSeek, and Google all tightly couple models to custom or optimized hardware, something Meta still lacks.
- •Recap of Facebook’s historic HTML5 vs native app mistake; Chamath calls it a politicized, ‘religious’ error against an obvious technical choice, analogous to today’s AI decisions.
- •He breaks down the AI secrets stack: (1) training + labeling methods (Scale), (2) application/agent tricks (Nat Friedman + Daniel Gross portfolio), (3) infrastructure & compute optimization (missing piece).
- •Examples of tight coupling: OpenAI with Azure, Anthropic with TPUs, DeepSeek’s hardware‑aware training, Google’s Gemini stack on TPUs.
- •Chamath’s earlier bet on CUDA transpilation failed because transformer attention mechanisms must be hand‑tuned for each silicon target; generic ‘just run on NVIDIA’ produces inferior models.
- •Conclusion: unless Meta develops or acquires deep hardware/compute secrets, LLaMA will lag frontier models, and Zuck must now buy that last layer to complete the strategy.
- 45:00 – 1:10:00
Mag 7 Divergence and Who Wins the AI Prize
Using year‑to‑date stock performance of the Mag 7, Thomas and Chamath argue the market is beginning to differentiate between winners and losers in AI. They each pick their top two AI ‘winners’ over the next five years, centering the debate on vertical integration, control over destiny, and the role of custom silicon.
- •Mag 7 performance snapshot: Meta +18%, Google –8%, NVIDIA +8%, Tesla –20%, Apple –21%, Amazon –3%, Microsoft +13% (approximate figures cited).
- •Chamath finds it odd that NVIDIA is up while others building their own silicon (Google, Tesla) are down, and singles Apple out as the only one clearly in harvest mode.
- •Control‑of‑destiny lens: Google and Tesla praised for vertical stacks (chips, models, data, hardware); Amazon seen as lacking its own foundation model; Microsoft partially controls OpenAI.
- •Five‑year AI prize picks: Thomas picks NVIDIA #1, Tesla #2 (vertical integration across silicon, models, and embodied hardware like Optimus). Chamath picks Tesla #1, Google #2 (best vision models + xAI for LLMs; Google’s Gemini + TPU + massive user funnel).
- •Jason picks ‘Elon’ plus Google, arguing Tesla and xAI should merge to concentrate talent and unify data (Tesla FSD/Optimus + X social data + xAI models).
- •Friedberg emphasizes option value: Tesla as a massive humanoid robotics call option, NVIDIA as durable but exposed to Chinese chip competition, and Google as a diversified AI portfolio with Waymo, quantum, and bio (Isomorphic).
- 1:10:00 – 1:42:00
Apple in Decline? Ambient Assistants, Robots, and Strategic Complacency
The conversation turns to Apple’s apparent lack of AI urgency and what, if anything, could revive its innovation engine. Friedberg defends Apple’s potential to win an ambient AI assistant race using its device ecosystem, while Chamath argues the company is locked into short‑term revenue optimization and culturally incapable of disruptive bets.
- •Jason laments Siri’s decades‑long stagnation and Apple’s failure to release breakthrough AI products despite massive cash; suggests humanoid robots as an obvious missed opportunity.
- •Friedberg’s bullish take: Apple is uniquely suited to build an ‘ambient AI assistant’ across iPhones, Macs, AirPods, Watches, CarPlay—an ethereal agent that follows you across devices and contexts.
- •He imagines Star Trek–style ‘computer’ interactions: identity‑aware, context‑persistent across home, car, office, and shared environments.
- •Thomas notes Apple successfully shifted its gross profit mix from one‑time iPhone sales to recurring services, showing it can pivot economics, though this AI pivot is tougher.
- •Clip of Craig Federighi downplaying new AI devices beyond phones/wearables is played; Chamath calls him ‘very competent’ at sustaining a cash cow, but not at reinventing the company.
- •Chamath’s indictment: Apple’s revenue per product chart shows iPhone stalled and growth driven by accessories and cables. He calls that ‘a this and that strategy’, optimizing for lost AirPods and connector changes, not breakthrough innovation. He views Apple’s creative destruction as both normal and acceptable, likening it to HP, Lotus, GE, AOL’s declines.
- 1:42:00 – 2:04:00
IPOs, SPACs, and the Coming Dispersion in Public Markets
The besties analyze a burst of new IPOs (CoreWeave, Circle, Chime) and a wave of AI‑ and crypto‑driven M&A as evidence that capital markets are reopening for future‑themed companies. They also debate the health of SaaS, the merits of various listing structures, and whether this is the right time for Chamath to relaunch a SPAC.
- •CoreWeave IPO: up ~4x, ~$81B market cap; Circle: ~25x oversubscribed, up ~6x to ~$48B; Chime: initially +40% then –20%, settles near $12B.
- •M&A blitz: multi‑billion‑dollar deals across AI infrastructure, devtools, and data platforms (e.g., Google–Waze, OpenAI acquisitions, Salesforce, Databricks–Neon, DoorDash, Uber).
- •Friedberg’s thesis: institutional crossover funds were overexposed to private markets post‑2021 crash, went quiet for three years, and now have pent‑up demand for high‑growth public tech issues.
- •Thomas’ SaaS data: median growth down from 17% (2021) to 9% now; share of >25% growers falls from ~25% to ~5%. Investors can no longer just ‘own the SaaS index’ and must seek new compounding stories in AI and crypto.
- •Chamath predicts wholesale rebuild of enterprise software with AI‑driven SDLCs, enabling small teams to replace bloated license stacks. He’s skeptical of consumption‑based pricing (e.g., Snowflake) as data volumes explode.
- •SPAC talk: Chamath’s Twitter poll about launching a new SPAC gets ~58k votes. He says sophisticated Wall Street and crypto players are urging him to do it, but emphatically warns retail not to participate, framing it as an advanced‑capital vehicle.
- •Thomas: from his perspective, what matters is business quality, float size, proportion of shares public (~20%+ ideal), and lockup structure—not whether it’s an IPO, SPAC, or direct listing.
- 2:04:00 – 2:25:00
AI vs the S&P 493: Roll‑ups, PE, and Who Survives
Zooming out from IPOs, the crew discusses how AI will reshape the broader S&P 493, including the role of private equity in buying laggard companies and modernizing them with AI. They debate whether PE can hire top AI talent, whether roll‑ups have terminal buyers, and how to pick incumbents likely to survive the AI transition.
- •Friedberg posits a future with sharp dispersion among the S&P 493: major alpha opportunities from picking incumbents that aggressively rebuild workflows with AI versus laggards that get obsoleted.
- •Chamath recounts a CIO with an $18B annual IT budget as emblematic of bloated, opaque spend; he argues IT and boards ‘speak different languages,’ enabling waste.
- •He’s working with a mega‑PE firm via 80/90 to rip out hundreds of millions in software licenses and replace them with tens of millions in bespoke AI‑enabled systems, massively lifting OpEx and business quality.
- •On PE roll‑ups (accounting, legal, IT services): Chamath worries there may be no terminal public buyer if agents eventually one‑click automate much of that work.
- •He prefers filtering the S&P 493 by defensible offline assets and post‑AI relevance; suggests specialty chemicals or tangible industrials over pure services as more resilient.
- •Thomas emphasizes Anthropic’s role: Q1 numbers suggest it may have added ~70% of all net‑new ARR generated by the public SaaS sector, underscoring how code‑gen vendors themselves are now the growth engines disrupting SaaS.
- 2:25:00 – 2:48:00
Microsoft, Clouds, and the Future Workforce in an AI World
The hosts consider how AI will affect big cloud providers (AWS, Azure, GCP), code generation, and tech workforce size, particularly at Microsoft. A quick poll splits the group on whether Microsoft’s headcount will be higher or lower in five years, highlighting disagreement on how fast AI can replace complex enterprise coding and how much cloud will still grow.
- •Microsoft shows vanity metrics like ‘% of code generated by AI’; Chamath argues this overstates impact, because most AI‑generated code is low quality when used in complex, long‑running enterprise contexts.
- •He cites Yann LeCun’s critique: compound error rates in multi‑step agents make most large code‑bases generated by current AI unreliable; today’s layoffs are more about cover to trim fat than direct AI substitution.
- •Poll: Will Microsoft have more employees in five years? Chamath says more (larger Azure + more bundled offerings); Jason says roughly the same; Friedberg says fewer (cloud commoditization, revenue pressure); Thomas sides with ‘more’.
- •Multi‑cloud reality: CIOs of Fortune 50 companies increasingly run workloads across AWS, Azure, and GCP for diversification, not just cost or best‑of‑breed.
- •Group consensus: if you could synthetically own only the cloud businesses of AWS, Azure, GCP, that index would be the only thing you’d need to own for the next five years.
- •Jason muses that if Elon replicated Colossus as an AWS competitor, he could be formidable, but Chamath notes xAI’s private fundraising environment is currently more advantageous than burdening Tesla with that capex.
- 2:48:00 – 3:12:00
The GENIUS Act: Stablecoins, Crypto Policy Pivot, and Onshoring
David Sacks joins late to explain the GENIUS Act—bipartisan stablecoin legislation that just cleared the Senate—and how it represents a dramatic turnaround from the previous administration’s anti‑crypto posture. The discussion covers Gensler’s ‘honeypot’ enforcement style, the impetus for bringing stablecoins onshore, and the political calculus that finally moved Democrats to support the bill.
- •GENIUS Act passes Senate 68–something, surpassing the 60‑vote filibuster threshold and signaling strong bipartisan backing.
- •Sacks recounts prior regime under Gary Gensler: startups were invited to ‘come in and talk’ then faced enforcement actions—regulation by prosecution that drove activity offshore.
- •Trump administration’s shift: early executive order signaling support for crypto and desire to make the U.S. ‘crypto capital of the planet.’
- •Political turning point: Sherrod Brown’s loss in Ohio after being an Elizabeth Warren‑aligned crypto blocker made Democrats ask, ‘Why die on this hill?’
- •Key GENIUS Act provisions: all USD stablecoin issuers must move onshore within three years; quarterly audits; full 1:1 reserves in U.S. bank accounts/T‑bills/money markets; creates a clear, predictable regime favored by issuers like Circle.
- •Compromise on interest: stablecoin issuers cannot pass interest income to token holders, a concession to community banks concerned about losing deposits. Sacks hopes this will be revisited once banks themselves issue stablecoins.
- •Enforcement: offshore issuers that fail to comply will find their tokens unlistable on U.S. exchanges and themselves in violation of U.S. law, pushing major players like Tether to adopt onshore, audited structures.
- 3:12:00
Wrap‑Up: Shifts in Tech Power, AI, and a Crypto Reset
The episode closes with shout‑outs, jokes, and references to upcoming All‑In Summit plans and Vinny Lingham’s carnivore documentary, but the underlying themes are clear: we are at a pivotal point where AI, cloud, and crypto policy are reshaping the hierarchy of tech giants and the broader market.
- •Quick banter as Sacks departs; Jason praises the speed and magnitude of the GENIUS Act achievement five months into the new administration.
- •Consensus that GENIUS plus market moves (CoreWeave, Circle, etc.) signal that crypto and AI are now central, not fringe, to U.S. economic strategy.
- •Light close: mention of All‑In Summit, tequila launch party, and Vinny Lingham’s pro‑meat documentary ‘Animal’ as they joke about slaughterhouses and caloric intake.
- •Implicit through‑line: tech, markets, and policy are aligning around AI, cloud, and regulated crypto as the dominant themes of the next decade.
