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
LA mayor race chatter: viral ads, homelessness, and political polarization
The besties open with banter and a discussion of Los Angeles politics, focusing on Spencer Pratt’s viral campaign ads and debate clips. They use the LA homelessness crisis as a springboard into broader themes about governance, public safety, and political messaging.
- •Spencer Pratt’s social media ads and debate performance praised as “next-gen” political campaigning
- •Homelessness framed as addiction/mental health vs. housing supply problem
- •Comparisons to national politics and concerns about ideological extremism
- •Side discussion of wealthy individuals being targeted and personal security implications
SpaceX–Anthropic compute deal: Colossus capacity, power constraints, and Claude rate limits
The panel breaks down the reported deal where Anthropic leases major capacity from Elon’s data center (Colossus). They emphasize that frontier model revenue is constrained more by compute and power availability than by demand.
- •Anthropic’s growth vs. capacity: demand is high, supply (power/compute) is the bottleneck
- •Colossus I scale cited: large GPU count, massive power footprint; Claude rate limits as symptom
- •Deal impact: improved Claude limits (paid tiers/API volumes) once compute expands
- •xAI also trains at Colossus; Elon’s early bet on compute begins to monetize
‘Elon Web Services’: SpaceX revenue narrative, hyperscaler competition, and IPO math
Brad and Chamath argue this is the clearest step yet toward Elon becoming a hyperscaler-like provider, not just a model builder. They connect the compute-leasing business to SpaceX’s valuation story, positioning infrastructure monetization as a subsidy for frontier model investment.
- •Elon’s “electrons to tokens” advantage and the case for EWS as a new cloud competitor
- •‘Layer cake’ framing: launch → connectivity → compute/hyperscaler → space data centers → apps/models
- •Compute monetization could add meaningful incremental revenue and offset xAI training costs
- •SpaceX IPO valuation debate: terrestrial compute blunts risks around delayed orbital data centers
Local opposition to data centers: activism, grid myths, and energy politics
The conversation shifts to the backlash against data center buildouts, with claims that protests are organized rather than organic. They dispute common narratives around water/electricity costs and argue that blocking capacity harms national competitiveness.
- •Claims of coordinated activism reminiscent of anti-nuclear movements decades ago
- •Counter-argument: bills rise where supply isn’t built; Texas cited as pro-build example
- •Power availability as the true limiter for frontier AI growth and national strategy
- •Call to investigate who funds and organizes anti-data-center efforts
Distributed compute ideas: homes, batteries, and ‘data centers beside every house’
Jason and Chamath explore how compute could move from centralized data centers to more distributed infrastructure—cars, homes, and batteries. They cite builder/utility-style partnerships that colocate GPU capacity near residential developments.
- •Analogy: data centers as ‘factories’—aligning with Tesla’s factory-building strength
- •Speculation: Powerwalls/cars as nodes in a distributed compute network; Starlink as connective tissue
- •Example discussed: homebuilder + Nvidia + electrical infrastructure partnerships
- •Broader implication: future compute footprint could extend from space to neighborhoods
Is Anthropic becoming a monopoly? Explosive ARR claims and competitive responses
Sacks makes a provocative case that Anthropic’s reported growth trajectory could lead to unprecedented concentration of power, while others argue it’s far too early to call. The group debates TAM, compute constraints, and whether rivals can re-focus on coding to catch up.
- •Sacks cites extreme growth rates and frames potential outcome as ‘biggest monopoly’ risk
- •Counterpoint: still early; competition includes OpenAI, Google, and cash-rich hyperscalers
- •Coding as the dominant revenue driver; rivals pivoting toward code/agents (e.g., Codex)
- •Constraints: compute/power limits and the speed of competitive reaction
Safety rhetoric vs regulatory capture: the ‘Safe Oil’ Rockefeller analogy
Sacks argues that safety narratives can be used to justify regulation that entrenches incumbents—comparing AI safety posturing to a hypothetical ‘Safe Oil’ Rockefeller. The panel tussles over whether safety initiatives are genuine guardrails or a path to capture and reduced competition.
- •Thought experiment: ‘Safe Oil’ + government safety agency as a distraction from monopoly-building
- •Claim: some safety proposals resemble moat-building and barrier-raising for newcomers
- •Debate over anti-competitive actions (e.g., restricting certain uses/access to models)
- •Shared principle stated: competition as ‘North Star,’ guardrails should be targeted
‘FDA for AI’ panic: White House messaging, what’s real vs ‘fake news’
The hosts react to reports that an AI-model approval regime is being considered, sparked by concerns over advanced cyber capabilities. Brad and Sacks argue the ‘FDA’ analogy is misleading and that an approval gate for model releases would be disastrous for innovation.
- •Reports: ‘AI working group’ and potential review procedures for new models
- •Brad’s view: coordination and preparedness, not a Washington pre-approval regime
- •Sacks: administration is pro-innovation; press spin amplified the ‘FDA’ framing
- •Acknowledgement: cyber capability escalation is real and requires faster defense readiness
Cybersecurity response: KYC for frontier access, logging, and public-private coordination
They workshop practical security measures for highly capable models during preview periods. The group supports tighter access controls (like KYC) and faster collaboration between labs, cybersecurity vendors, and government—without creating a permanent centralized approval bureaucracy.
- •KYC/identity checks proposed for early access to high-risk cyber-capable models
- •Labs already monitor API usage for abuse and anti-distillation; suspicious activity flagged
- •Policy stance: enable defenders quickly (CrowdStrike/PANW + long-tail startups) to harden systems
- •Critique of ‘never let a crisis go to waste’ attempts to build permanent regulatory structures
Fixing AI’s public image: giving back, healthcare/education upside, and ‘vibe shift’
Chamath and Jason argue AI backlash is rooted in fear of concentrated gains and poorly communicated benefits. They propose more visible, large-scale redistribution mechanisms and stronger storytelling about AI’s potential to reduce costs in healthcare and education.
- •‘Vibe shift’ against tech/AI oligarchy: perception of few winners, many losers
- •Proposal: IPO-linked giving (e.g., shares/funds to citizens via investment accounts)
- •Need to emphasize concrete benefits: healthcare outcomes, education cost reduction, broader uplift
- •Debate spillover into minimum wage/universal healthcare as part of addressing inequality fears
Trading the AI boom: hyperscaler growth, market multiples, and policy tailwinds
Brad lays out a bullish market view driven by accelerating cloud growth, AI revenue realization, and reasonable valuations relative to earnings. They contrast macro doom narratives with data showing strong GDP, controlled inflation, and large-cap margin expansion.
- •Cloud run-rate and growth: AWS/Azure/GCP growth cited as evidence ROI is showing up
- •Valuation argument: not a bubble—multiples described as moderate for mega-caps/memory names
- •AI revenue acceleration framed as key catalyst that prevented a market drawdown
- •Policy tailwinds emphasized: chips/models deregulation and energy buildout enabling data centers
The next fork in the road: proving enterprise ROI and real economy productivity gains
The episode closes with a debate about when AI spending must translate into measurable operating margin expansion and productivity improvements across the broader economy. Chamath argues token revenue isn’t enough without clear ‘X spent → Y earned’ results; others cite early margin and headcount data as evidence it’s already happening.
- •Chamath: AI must show durable profit/margin impact across the S&P and real economy within ~1–3 years
- •Counter: early signals—margin expansion, slowed headcount growth, faster top-line growth in leaders
- •Discussion of jobs narrative: low unemployment and improved outcomes for new grads vs participation-rate concerns
- •Consensus tension: near-term boom likely continues, but long-term legitimacy depends on broad, provable benefits
