All-In PodcastWhy Anthropic could be the most powerful monopoly ever made
Sacks compares Anthropic's current trajectory to Standard Oil; the SpaceX compute deal eases supply constraints and turns Elon into a hyperscaler.
CHAPTERS
- 0:00 – 4:36
Besties banter + LA mayor race as a signal of voter mood
The crew opens with quick mic checks, jokes, and a discussion of viral political ads and debate clips in the Los Angeles mayoral race. They use it to broaden into homelessness, public safety, and how media/PR is shaping politics.
- •Viral Spencer Pratt ads and debate performance as “next-gen” campaign strategy
- •Homelessness framed as addiction/mental health vs housing supply
- •Safety concerns and politicization of wealth/CEOs being targeted
- •California vs Texas/Florida relocation sentiment among business leaders
- 4:36 – 6:01
Elon–Anthropic compute deal: Colossus capacity, power constraints, and “EWS” emerges
They break down the reported SpaceX/xAI data center leasing deal with Anthropic, focusing on GPU scale, megawatts, and immediate product impact (rate limits lifting). The group frames it as Elon becoming a hyperscaler—an “Elon Web Services” competitor to AWS/Azure/GCP.
- •Anthropic demand vs supply constraints: compute and power as the real bottleneck
- •Colossus I lease details: GPUs/energy scale and why Claude saw rate-limit issues
- •EWS positioning: Elon as a new hyperscaler player
- •Elon’s speed in building capacity becomes strategic leverage in the AI race
- 6:01 – 11:08
SpaceX valuation logic: terrestrial compute as a bridge to orbital data centers
Chamath and Brad argue that monetizing terrestrial data center capacity reduces the bear case around delayed orbital data centers. They present compute-as-revenue as a way to subsidize xAI/Grok development while strengthening the SpaceX IPO story.
- •Orbital data centers are a key long-term narrative; terrestrial capacity derisks timing
- •Leasing inference-heavy clusters (e.g., H100s) as near-term cash generator
- •Subsidizing frontier-model CapEx while xAI ramps revenue
- •How hyperscaler-like revenue can change IPO multiples and investor perception
- 11:08 – 20:25
Power, protests, and activist organizing: the data-center permitting battlefield
The conversation shifts to energy infrastructure and the backlash against new data centers. Brad claims many protests are coordinated activist efforts, drawing parallels to anti-nuclear movements that constrained U.S. energy supply.
- •Power availability as the ultimate AI constraint; local resistance threatens new capacity
- •Claims of organized, traveling activist campaigns vs “organic” local concerns
- •Debunking narratives about water/electricity bills; Texas vs NY/CA comparisons
- •Geopolitical angle: U.S. energy buildout vs China’s nuclear expansion
- 20:25 – 26:48
Distributed compute and “home + grid” strategies: Powerwalls, cars, and builders
Jason sketches a future where compute spreads from mega data centers to homes and devices, leveraging Starlink connectivity and Tesla’s installed base. They cite a homebuilder partnership model that colocates GPU clusters near housing developments.
- •Data centers as factories; Tesla/SpaceX competence in manufacturing + energy
- •Speculation: distributed compute via cars, home batteries, and Starlink networking
- •Example: homebuilder + Nvidia-style local cluster deployments
- •Implications for where compute lives (Earth vs space) and who controls it
- 26:48 – 27:19
Is Anthropic becoming a monopoly? Explosive ARR debate and competitive response
Sacks claims Anthropic’s growth trajectory could create an unprecedented monopoly; others argue it’s far too early and competition remains intense. They debate TAM, compute limits, and how rivals (OpenAI/Google/xAI) will respond by focusing on coding.
- •Anthropic’s reported growth rates and what “10x/year” implies at scale
- •Coding as the revenue center of AI; “misfocused” product bets by competitors
- •Competitive reaction: OpenAI’s coding push, Google’s strengths, xAI/Cursor deal
- •Leading vs catching-up dynamics: inertia favors the current leader
- 27:19 – 35:17
‘Safe Oil’ analogy: safety rhetoric, regulatory capture, and anti-competitive tactics
Sacks introduces a Rockefeller thought experiment to argue that ‘safety’ can be used to justify regulation that entrenches incumbents. The group spars over whether safety framing is sincere or a path to regulatory capture and market lock-in.
- •Safety debates can distract from consolidation and monopoly formation
- •Regulatory capture risk: approvals/licensing regimes favor incumbents
- •Example concern: banning competitors’ products from using models (anti-competitive behavior)
- •Washington tends to react late to monopolies; question is whether to act early
- 35:17 – 45:10
White House ‘FDA for AI’ panic: what’s real vs media spin
They dissect reports that the administration is considering an FDA-like review for AI models, with Brad and Sacks arguing the ‘approval regime’ framing is overstated. They position the likely reality as coordination and preparedness rather than pre-approval gatekeeping.
- •NYT report vs administration intent: skepticism about true ‘FDA for models’ plan
- •Difference between coordination (preparedness) and mandatory pre-approval
- •Sacks references existing framework: targeted laws vs blanket power grab
- •Political dynamics: fear of blame after AI-enabled cyber incidents
- 45:10 – 52:01
Cyber-capable frontier models: coordination, KYC, logging, and getting tools to defenders
They focus on the real near-term risk: AI rapidly boosting cyber offense/defense capabilities. The proposed response is accelerating deployment of tools to cybersecurity firms, using KYC during preview access, and coordinating with government on suspicious usage.
- •Cyber capability will diffuse across labs (and open source) in months
- •Defense strategy: empower CrowdStrike/Palo Alto + startups with frontier tools
- •KYC for preview access; debate over logging and privacy in general release
- •Argument that ‘pre-release approval’ solves the wrong problem; labs already self-regulate
- 52:01 – 1:00:01
Fixing AI’s public perception: giving, community benefits, healthcare and education upside
The group argues AI has a messaging and legitimacy problem, and proposes ways to share gains more broadly. Jason advocates structured giving (e.g., IPO allocations to citizens) and delivering visible benefits in health, education, and local communities.
- •‘Vibe shift’ against tech/AI oligarchs; public focuses on downside narratives
- •Proposals: structured giving (e.g., IPO slices into citizen investment accounts)
- •Reframing AI as deflationary and beneficial to cost of living and services
- •Local benefit ideas: e.g., cheaper/free electricity where data centers are built
- 1:00:01 – 1:16:17
Markets and the AI trade: hyperscaler growth, valuations, and the ROI question
They move into investing and macro, citing hyperscaler growth rates, Big Tech valuations, and semiconductor/memory multiples. The debate becomes whether AI’s revenue is translating into broad corporate margin expansion and durable productivity gains.
- •Cloud growth as AI demand proxy: AWS/Azure/GCP acceleration
- •Valuation argument: not a classic bubble if earnings multiples stay moderate
- •Portfolio positioning toward compute/memory and why revenue proof mattered
- •Fork-in-the-road thesis: token spend must produce measurable enterprise ROI
- 1:16:17 – 1:22:01
Jobs, productivity, and wrap-up: conflicting signals and the ‘AI boom’ narrative
They close by debating labor market indicators, participation rates, and whether AI is already boosting productivity without raising unemployment. The episode ends with playful jabs about monopolies, media narratives, and America’s competitive position in AI.
- •Unemployment remains low; debate on participation rate vs headline unemployment
- •College grad job market improvement as counterpoint to ‘AI destroys entry-level’ story
- •Argument that AI-driven construction/energy buildout is a broad economic tailwind
- •Closing banter: monopoly jokes, media ‘fake news’ framing, and optimism about U.S. innovation
