Elon’s Anthropic Deal, The Next AI Monopoly?, “FDA for AI” Panic, Trading the AI Boom
In a Nutshell
Elon leased xAI's Colossus data center to Anthropic, fueling its explosive growth from $10B to potentially $100B+ ARR by year-end and positioning it as a potential AI monopoly rivaling MAG7 giants, while enabling Elon's "EWS" to challenge hyperscalers with $4-5B incremental revenue. Hosts debate monopoly risks, rejecting "FDA for AI" regulation as preemptive overreach that favors incumbents like Anthropic/OpenAI, advocating private-sector KYC, coordination, and competition among US labs to sustain the AI boom driving GDP growth, margin expansion, and blue-collar jobs. Trump policies unleashed this surge by scrapping Biden-era restrictions, proving AI's ROI through efficiency gains amid low unemployment, with calls for innovators to counter doomer narratives via giving and pro-growth policies.
These notes were generated by AI and may contain inaccuracies.
Chamath Palihapitiya, David Sacks, Brad Gerstner hosting All-In Podcast. David Friedberg absent due to flu, back next week with two incredible interviews.
Spencer Pratt ads are incredible, best political ads seen, social media team on fire. Low cost to produce, go viral. Pratt likely to win LA mayor race against Karen Bass (extremely left-wing mayor) and Indian Fidel Castro Ramon (city councilwoman further left, in charge of homeless programs). Pratt eviscerated her in debate: homelessness is addiction and mental illness issue, not lack of housing. Said she'd get stabbed on Skid Row. Viral clip reminded of Trump.
Retirement Protection and Savings Act ballot initiative to pass big, constitutional amendment protecting retirement savings and personal assets, banning wealth tax. Non-consensus optimistic view.
Mondami attack video on Ken Griffin's house, stood in front pointing at it, dangerous amid CEO shooting and Sam Altman house targeting. Griffin leaving for Florida, citing personal safety. Mondami mealy-mouthed response, no apology.
Elon leased all of Colossus 1 data center to Anthropic. Chamath predicted Elon-Dario deal last week. Anthropic added 220,000+ Nvidia GPUs, 300+ megawatts. Claude doubled rate limits, removed peak caps for paid users, increased API volumes for Opus models. xAI trading models at Colossus 2. Emergence of Elon Web Services (EWS) competing with Google Cloud, AWS, Azure.
Anthropic/OpenAI revenue tied to supply constraints in data centers/power, not demand. Infinite power would make revenues more parabolic. 5-year view robust, need more compute/power. Poor leadership at AI firms seen as untrustworthy/self-interested, turning political/community reaction negative. 9 gigawatts online this year, 50% protested, likely turned off. Elon king-making with terrestrial capacity via API/LIDT/Alice, subsidizes Grok training even if orbital data centers delayed.
Elon best at converting electrons to tokens. SpaceX 5-layer cake: launch, connectivity, compute, hyperscaler, space data centers, applications/models, other bets. EWS generates $4-5B incremental revenue this year on top of mid-20s analyst estimates. Subsidizes next-gen Grok. Facilities: Colossus, MacroHard (1.2GW Blackwell), MacroHarder. H100 great for inference, monetized to Anthropic. Takes pressure off xAI immediate revenue. Not last announcement. Detente between Anthropic/Elon crucial for American frontier models.
Activists protesting data centers not organic, highly organized like anti-fission 30 years ago (stopped US nuclear, China has 100). Misinformation on water/electricity bills (up in NY/CA without data centers, down in Texas with most).
Deal complementary. SpaceX profitable in space/telecom/Starlink. xAI huge losses from training clusters pre-revenue, revenue in enterprise/coding (xAI Cursor deal to catch up, no coding product yet). Leasing fixes balance sheet/capex.
Anthropic 10x growth/year last 3 years. Q1: $10B to $30B ARR. April: $30B to $44B ARR. Unprecedented at scale. Compute deal ensures 10x to ~$100B ARR year-end, trillion by 2027? Most valuable tech company ever, bigger than MAG7 combined. Hyperscalers: Google ~$100-120B/quarter, 20% YoY growth vs 1000%. Anthropic on track to most powerful monopoly ever (Dario calls AGI).
2025 revenues: Apple $420B, Microsoft $300B, Alphabet $390B, Amazon $700B, Nvidia $190B, Meta $185B, Tesla $110B. Total ~$2.3-2.35T. Anthropic $1T = MAG1. Dario Andorkish: combined AI leaders ~$1.029T.
Coding TAM $1T/year (software developers/creation). Could double to $2T as coding 10-100x.
Frontier labs: Anthropic porcupine (focused), others fox (nano banana, Sora, image gen, fantasy chatbots = waste). Market now coding/Co-work/agents. Others pivoting: OpenAI Codex on GPT 5.5 (new base Spud), accelerating. Google strong in coding. Elon/Cursor. Leader maintains inertia, others change.
SpaceX: $20B this year. Hyperscalers: $300B revenue, $4-5T market cap. Elon core: factories (data centers), energy (batteries, SolarCity). $5B EWS neo-cloud. Compute in Teslas/Powerwalls, distributed home-to-home, space. Announcement hints space interest. Pulte Homes/NVIDIA/Span deal: GPU clusters beside homes.
Span: smart power panels (breakers in app), pivoted to this. Base Power similar.
SpaceX IPO 40-50x revenue (next year $40-50B → $2T). Elon future pipeline/LATAM commands premium vs MAG5 25x earnings. Tesla same Elon variable (2-4x market). Apple/MAG devalued for lack of innovation/incrementalism. Tesla/SpaceX merge to Elon Corp by end-year/mid-next: premium asset as others stopped innovating (e.g., Tim Cook's Apple TV = Netflix knockoff).
Debate on whether certain companies are innovating in AI or self-driving cars. Critics argue they shut down innovative projects and won't take risks, leading to no premium in valuation. Counterarguments highlight incredible valuations by every metric, especially compared to others. No world exists where Google, Meta, Apple, and Amazon are viewed as penalized in valuation, while Elon gets a massive premium for innovating.
Participants agree they're saying the same thing but debate semantics. Reference to SaxxyPoo as captain of debate club and master debater, with jokes about a 20,000-word article.
Tech markets historically consolidate into monopolies or duopolies. Currently, only two companies make substantial AI revenue: Anthropic and OpenAI. OpenAI growing at 3-4x, Anthropic at exponential 10x a year. If Anthropic continues for 18 more months, it becomes the most valuable company in history with unprecedented control over the most important technology.
John D. Rockefeller, most successful ruthless monopolist, poor at PR (e.g., There Will Be Blood). Imagine if he rebranded Standard Oil as Safe Oil (kerosene dangerous: lights houses, burns them down, makes bombs). He could call for government agency for rigorous testing, licensing, common-sense regulation on safety standards like wick thickness, blocking independent refiners. Public gets distracted from monopoly-building, might even call him effective altruist for safety concern.
Ridiculous to see as monopoly. Annual run rate revenues, but on GAAP basis, Anthropic matches OpenAI in March. Five months ago, OpenAI thought to run away with it. Google has substantial AI revenues. Google, Amazon produce $100 billion free cash flow to justify investments. Startups like these are fledgling, fragile. Best competition in AI, at frontier, leading world. Want DC out of way, no roadblocks to competition. Translation: Don't mess with Brad's paper (bets on these).
Nobody wants vigorous competition more. Premise of action plan last year: bring out best in everyone for America to win AI race with five major labs. Competition is good, can have guardrails. Washington wakes to monopolies late (e.g., 80% market share). Not monopoly yet, but 18 months trajectory leads there. Don't get distracted by safety rhetoric—Rockefeller could use it too. Safety-ist policies seek regulatory capture, stronger moat for monopoly/duopoly, hinders competition. Be skeptical of altruistic claims.
Can't believe talking monopolies before leaving starting gate. Only two with revenue. DC preemptively picking winners/losers at AI starting line would be disastrous.
Monopolists win, stop competition via regulatory capture, ban competitors from products. Anthropic banning OpenAI from using its models is anti-competitive—worthy of double-clicking.
All-In Summit selling out fast, Sept 13th-15th in Los Angeles. Top speakers (Freebird, Sax). Networking software for meeting people. Playbook: learn 1-2 things, meet someone new, great food/fun—get 2-3, return. Liquidity sold out, waitlist closed.
Reports: White House considering FDA for AI to vet new models for safety. Trump considering executive order for AI working group (tech execs, officials) to examine oversight, including review process for new models. Catalyst: Anthropic's Mythos model spooked White House (avoid political repercussions from AI-enabled cyber attack).
Kevin Hassett (National Economic Council director) on Fox Business: Studying executive order for clear roadmap on future AIs creating vulnerabilities—process like FDA drug approval, proven safe before release.
"We're studying possibly an executive order to create an executive order uh to give a clear roadmap to everybody about how this is going to go and how future AIs that also potentially create vulnerabilities uh should go through a process so that you know they're released in the wild after they've been proven safe just like an FDA drug."
Scott Bessent: Step change in LLM power last month, will see from others. US government works with AI companies to innovate but maintain safety—calculus between innovation and safety.
Kevin Hassett doesn't think FDA right analog—wants models shown for coordination, harden systems, update intelligence. No approval regime (pre-approve every model in Washington)—disaster. Leads to three bad things: (1) Washington picking winners/losers; America winning, don't change; (2) burden with more democracy; (3) improve coordination framework (finite feedback time, build government capacity for cyber review). Continue current path. No FDA for models.
Agree no FDA for AI, but not total free-for-all—need guardrails. Vibe shift on tech oligarchs/AI from Main Street to Washington. Regulations coming (worse under Democrats, some under Republicans). Why? Horrible messaging on AI positives, no broad support. Fear of few winners, many losers disconcerting. Tech leaders not reinvesting/uplifting America at scale—D- trending to F. Buildup of antibodies. Glass half-full: demonstrate investment, poor communication. Response: three trillion-dollar net worths controlling keys.
Easy to imagine bad AI outcomes (deepfakes, robotics, self-driving jobs, bioweapons)—humans biased toward safety. No one articulating positives.
Fake news on FDA for AI—no senior official supports (spoke to Hassett). Trump most pro-innovation president; Susie Wiles (chief of staff) statement shoots it down. Originated from NYT/Andrew Ross Sorkin commentary, took life. Visceral reaction—damaging to innovation.
Straw-manning Trump AI first year as laissez-faire/no guardrails. Actually, March 20th White House national AI regulatory framework (Sachs worked on): four-page bulleted list of legislation for Congress. Specific solutions to problems, not giant Washington power grab.
Legitimate issue: Mythos/cyber capabilities (OpenAI matches, all frontier labs/Chinese in 3-6 months). Need system hardening, scan/patch codebases before hackers (AI good at cyber offense/defense; open-source already has).
Government/private sector cooperate. Leverage US cybersecurity industry (CrowdStrike, Palo Alto Networks, startups in AI-powered pen testing). Get tools (Mythos, OpenAI models) to them as force multiplier for companies/vendors/IT depts.
KYC (know your customer) wrapper: identify users before access to Mythos (not state actors/bad guys). Anthropic/OpenAI acted responsibly—self-regulated, didn't release super powerful models. Solves non-problem of pre-release approvals. For super powerful API, basic KYC makes sense to know recipients. Rapidly get tools to good guys—KYC predicate. Log usage for security (not public release due to privacy).
PolyMarket: 21% chance Trump orders federal review of AI models by May 31st. Don't bet.
For the preview period, implement KYC. Post-preview in general release, KYC matters less due to widespread use. All labs track API use and flag suspicious activity amid major anti-distillation efforts. Extensive coordination with government on API monitoring; suspicious activities shared. Allow some use to observe extractions. Better coordination needed.
Pre-release approvals solve a non-existent problem; Anthropic and OpenAI weren't rushing releases. AI doomers exploit cyber issues using 'never-let-a-crisis-go-to-waste' strategy. Harden systems over next 3-6 months, but avoid permanent Washington infrastructure.
Bernie Sanders supports FDA for AI, saying he likes Trump administration idea for first time; claims AI out of control, takes jobs and summer homes. Sanders wants to stop progress, ban data centers, bought into doomer narrative to kibosh innovation.
Two strategies: major giving from Nvidia, SpaceX, Anthropic, OpenAI IPOs—e.g., 5% or 1% to every American via Invest America (IPOK, IPO for kids). Giving pledge was virtue signaling; propose real 1% stock annually over 20 years.
AI massive for health, life extension, reducing suffering; contribute to basic science, education, lowering costs. Bottom half anxieties: income, healthcare, housing, kids' education.
Study minimum wage raises in New Zealand, Sweden, Switzerland, Australia—consumers spend, don't save. Opt-in company-by-company raises (Amazon, Target, restaurants) to lift bottom third. Capitalists should discuss universal healthcare, minimum wage—good for capitalism, consumer-driven economy (60-70-80% GDP). More spending on Netflix, Amazon.
"If we increase the minimum wage just modestly each year, and we opted into doing that, and we figured out a way to give universal healthcare, companies wouldn't have to deal with universal healthcare, and we would have customers."
AI black eye from wealth polarization, job fears. Privately, support universal healthcare if feasible; issue is cost, not desirability. PJ O'Rourke: "If you think healthcare is expensive now, just wait until you make it free."
AI unpopular per polls but ranks 29th out of 39 issues—not top of mind. Top issues: cost of living (#1), economy (#2). AI deflationary, drives 75% of Q1 GDP growth. Construction boom, blue-collar boom, 25-30% wage increases for construction workers. Healthcare also prominent.
Media should report AI economic boom honestly. Bernie Sanders data center moratorium would cause negative GDP, stock market down 15-20%, rising unemployment. 1929: 4% unemployment to 23% in 3 years due to government regulation.
AI delivers net benefits: low unemployment, economic/productivity growth. Deliver net benefits like every American AI investment account. Data centers in Abilene, Texas—make local household electricity free.
As capitalist/innovator: society $38 trillion in debt, can't afford top-down solutions. Markets solve better; government involvement inflates healthcare, education. Entering age of abundance with AI coaches. Let markets work, government ensure safety.
Founders avoid housing, education, minimum wage, healthcare due to regulation—kryptonite. Remove regulation to innovate.
AWS: $150B run rate, 28% growth. Azure: $108B, 39% growth (includes software). GCP: $80B, 63% growth (includes Google Workspace). ARR multiples: Google Cloud 10x, AWS 10x, Azure 9.5x—$30B collectively.
Markets at all-time highs: Mag7, Uber, Disney strong growth. Consumer/tech strong. Inflation up slightly, hand-wringing on wars, oil.
Despite criticism (tariffs hyperinflation, Venezuela/Iran conflicts), accelerating GDP, 10-year at 4.3%, inflation controlled, AI/compute leadership. S&P up 8% YTD—not bubble. Valuations: Meta 17x, Nvidia 19x, Microsoft 20x, Google 24x fully-taxed GAAP earnings. Memory: SK Hynix 5x, Samsung 6x, Micron 7x (25% portfolio).
OpenAI + Anthropic: $30B combined revenue start of year, now $80B four months later. Administration policies working; US AI gap growing. Went all-in on market end of last year after bearish stance—saw AI revenue in Dec/Jan. 80%+ exposure in compute/AI/memory. Private markets inform public.
Without $100B war and unconstitutional tariffs, economy stronger. Goldilocks: geopolitics reset, discount rates down, trillion-dollar privates (SpaceX) going public. USA winning; American exceptionalism (SpaceX, Google). Administration business-friendly but tariffs poorly executed, war mistake—seek quick resolution.
AI boom thanks to Trump rescinding Biden policies first week: model approvals, GPU sales licensing. Declared win AI race, unleashed companies. Energy: "drill baby drill," AI companies as energy cos for data centers (own power, not grid). Blue-collar construction boom.
Alternative: Bernie Sanders—raise taxes, ban data centers.
Markets up for while, then down. Bullish catalysts: new things (Nvidia, memory, Anthropic, SpaceX, OpenAI) demonstrate value. Headwinds: no evidence AI lifts S&P 500 operating margins. Fork in 2-3 years (500 days): OPEX shrinks (margins up, workforce down) vs. revenue growth/margins expand/OPEX flat/up. Paying for tokens must see benefits; be net long now, reckoning soon.
Model revenue proven (from capex question), now ROI on tokens/applications. Enterprises buying coding tokens month-over-month expect ROI. ROI trickling: infrastructure → models → apps → end-user economic boom.
Deja vu: PC/internet/cloud/mobile revolutions—hand-wringing to payoff. Using agents, building code; 3 people at 22-person firm ship products vs. SaaS. ROI fait accompli in startups: build/ship with half employees for same capital.
Enterprise picture needed. No S&P profits yet—e.g., Anheuser-Busch sell more beer, Nike more shoes, medical devices company.
Companies must demonstrate direct ROI from AI investments: spend X to generate Y where Y > X and margins improve to accelerate the flywheel. Currently, experimentation is rampant but unproven. No visible impact yet in global GDP, global productivity, or S&P 500 profit margins.
Counterarguments highlight cost reductions and efficiency:
- Ads more effective, hiring paused.
- Nike generates app imagery without photographers.
- Startup aids brands like Nike; another helped DoorDash replace food photos with AI, slashing costs.
- AI ad creatives double-digit more effective at half the cost.
Data points:
- Azure grew 39% last quarter.
- Google Cloud grew 63%.
- MAG5 headcount growth ~3% over 3 years.
- Operating margins expanding.
- S&P 500 Q1 '24 margins 11.8% (up from 11% in '23), now 13%—200 basis point improvement.
Debate: Is this AI-driven or financial engineering/post-COVID layoffs? Consensus forecasts continued margin expansion next 2 years. Anecdotes show top-line acceleration without headcount growth.
Operating efficiency gains occurred with unemployment at historic lows (4-5% full employment threshold; actual ~4.2%). Unemployment for recent college graduates dropped, countering AI job loss narrative—now easier for them to find work (WSJ: "College graduates are finally catching a break in this job market"). No evidence of AI-driven downsides like mass unemployment; possible AI-native skills help.
Conflicting evidence: Labor participation rate 61.9% in March (pre-COVID 63.3%). Mixed job outcomes by degree; too early to conclude.
Heated exchange on data trust, Fed role (accusations of calling to "abolish the Fed" denied).
Host ends debate, congratulates innovators, references Elon and Dario (Anthropic) "Rockefeller" deal. Invites Dario on program. Praises US AI ecosystem (Elon, Anthropic, OpenAI, Google, Amazon) for competition, model-sharing, innovation. US has best framework; stay the course—"America for the win."
Playful jabs:
- Brad Gerstner teased as future Senator or Secretary (Treasury/State); suggests ditching red glasses for tortoiseshells.
- David Sacks: Secretary of Cashmere and Wine.
- Chamath Palihapitiya: Secretary of Cashmere? (misheard).
Affirmation of US innovation despite poking fun; happy companies are American. "Congratulations on your monopoly, Dario."
Media critique: Fake news ignores positive trends; good coverage of "micro looting" (NYT). Let winners ride.
Keep All-In Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
