OpenAI's Identity Crisis, Datacenter Wars, Market Up on Iran News, Mamdani's First Tax, Swalwell Out
In a Nutshell
The All-In podcast critiques NYC's proposed 3.9% pied-à-terre tax on second homes over $5M as a demand-killer that won't boost affordability without building more housing, while doxxing risks like targeting Ken Griffin's unit heighten safety fears amid attacks on figures like Sam Altman. OpenAI faces an identity crisis with slowing consumer growth (ChatGPT market share slipping), unfocused pivots to enterprise amid Anthropic's faster 10x revenue surge to $30B run-rate, but both hit physical limits in compute, power, and data centers—spurring xAI/Meta's massive GPU builds (555K/150K) and warnings of growth walls without owned infrastructure. Markets shrug off Iran war news to hit highs (Shiller PE/Buffett Index at peaks), Eric Swalwell exits politics amid scandals, and AI valuations hinge on proving trillion-scale enterprise ROI despite hype, with compute crunches signaling overvaluation risks for labs like Anthropic/OpenAI.
These notes were generated by AI and may contain inaccuracies.
Welcome to the podcast with the core four and Captain Travis Kalanick. Discussion starts next door to David. Jokes about not revealing locations and Mamdani doxxing outside houses, asking to foreclose.
Mamdani's tax speculated at 3.9% per year on second homes or pied-à-terre over $5 million. No homes under $5 million in Manhattan; targets area within 15 miles of midtown. Hits second or third homes most elastically, owned by non-residents who could buy anywhere. Will crash demand for second homes, impacting whole market. Claimed to improve housing affordability, but no incentive to build more. All units matter; buying in New York turns properties into low-income housing equivalents, e.g., penthouse on 57th Street or Gramercy would need breaking into seven units.
Doxxing Ken Griffin and Safety Concerns
Mamdani video points at billionaire's home (Ken Griffin's known unit); everyone knew he owned it, doesn't live there. Not true doxing like standing outside primary residence. Still a dog whistle to crazy people, seen by 30-50 million. In week of fire, Molotov cocktail, and bullet at Sam Altman's house, deadly serious. Reverse test: Republican outside Bernie Sanders' second home would be criticized. Nobody deserves house fire-bombed or shot at.
Economic Impacts
Current owners like Ken Griffin may pay tax and keep homes, but no new buyers due to uncertainty and rising taxes. After 10-11 years with interest/inflation, $10 million unit effectively $20 million to break even. Downstream: less spending on Knicks games, restaurants, birthday parties. Not fans of pied-à-terre culture with multiple houses ruining places like London (ghost towns in Chelsea on weekdays from land banking real estate as Bitcoin-like asset).
Tax hits owners renting out non-primary residences, even short-term (two-week, 30-day, five-year). Good for hotels. London's high-end market collapsed after stamp tax; rich fled to Zurich, Lugano, Milan. UK government sees no major impact yet, but slow bleed hard to reverse.
Housing Affordability and Building
Housing crisis cap; in Austin, Nevada, Florida, building units drops rents/prices despite migration. Austin doubled in decade, one/two-bedroom rents down three years running. Democratic cities/NIMBYs block building; Republican areas allow it for affordability.
London high-end bid down, bad for city/UK. Cities like London attract money via rule of law and uniqueness. Arbitrary taxes make money flee. Second-home owners pay property taxes without using services, profitable. Developers need price-insensitive whales like Ken Griffin (paying $3-4 million/year taxes) to make projects viable, subsidizing development/cranes.
LA 5% mansion tax (exempting Beverly Hills); San Francisco transfer tax starts 5% over $5 million, 6% over $25 million (total 13% with commission). Dried up transactions, no house flipping. Property unsafe in blue states; retroactive taxes take chunks (e.g., speaker's SF house). Wealthy will move money elsewhere.
OpenAI Chief Revenue Officer Denise Dresser sent four-page memo to employees (leaked, probably intentional). $30 billion run rate cap inflated by $8 billion from revenue share/accounting with AI providers. Anthropic built on fear, restriction, elite control of AI.
OpenAI Pivot
Pivoting to business customers, winning agent platform layer. Hired architect of open-source OpenClaw (Peter Steinberger); cynical view: wants his innovations in OpenAI products. Perplexity quadrupled revenue. xAI has cool stuff coming (Elon Musk). New o1 model (spud on PolyMarket 75% chance next week). Memo undercuts Anthropic valuation.
FT cites anonymous OpenAI investors frustrated with lack of focus: > "You have ChatGPT a 1 billion user business growing 50-100% a year. What are you doing talking about enterprise and code? It's a deeply unfocused company."
ChatGPT market share down as users up (Gemini, Claude gaining; Meta first proprietary model; Apple users incoming). Gen AI website traffic shown.
Panel Discussion on OpenAI vs. Anthropic
Chamath: From 8090 team, Codex better than Anthropic for complex long-horizon coding; Claude ensemble for day-to-day. OpenAI sees $3-4 trillion consumer value, $2-3 trillion enterprise; total $7-9 trillion market cap long-term. Separate consumer/enterprise teams to avoid context switching.
Travis: Growth is king; Anthropic growing faster, priced higher than OpenAI on secondaries (flippening). $850 billion OpenAI valuation questioned; needs $1.2 trillion IPO. Network effects from compute, tokens, RL favor scale/volume. Like Uber accelerating past Lyft, DoorDash.
Freeberg: Anthropic unprecedented innovation pace; supplanted OpenClaw, new Opus model. Organization shifted 90% to Anthropic in six months (from Cursor, Gemini). OpenAI raised $122 billion largest round ever, no fundraising issue.
Sacks: Anthropic revenue growth 10x/year vs. OpenAI 3-4x; both ~$30 billion Q2 (OpenAI apples-to-apples 20% less). Anthropic: $1-10 billion last year to $30 billion Q1, $80-100 billion year-end (logarithmic straight line). Businesses pay metered for code tokens (scales like electricity); consumer low willingness-to-pay (3-4% premium conversion, $20/month all-you-can-eat, doesn't scale). OpenAI unfocused (e.g., buying podcast); should focus on enterprise/coding to catch up, or Anthropic lead insurmountable in 1-2 years. Exponential can't last forever (e.g., $100 billion to $1 trillion?).
Even if Anthropic reaches $100 billion this year, hitting $1 trillion the next seems unlikely due to new problems at scale: running out of compute, electricity, data centers, and infrastructure. Physical limits are evident; users complained Claude was thinking less, with typical prompt thinking time cut by two-thirds. Recently released Opus 4.7 replaced Opus 4.6, restoring thinking time, possibly at higher cost.
Anthropic may reconsider supporting Doomer nimbyism, as they now need to build their own data centers and regret opposing them nationwide. This could limit their growth, hoisted by their own petard.
Coding platforms improve Anthropic's products while generating revenue. However, Twitter users pan cloud desktop as "vibe coded slop." Kept on task with proper guardrails, agents are force multipliers. No one has built real products at scale using agents, redefined org structures, or budgeted properly. Uber's CTO hit token budget limits.
Frontier labs like OpenAI and Anthropic grow so fast they require own infrastructure. Early on, renting from hyperscalers is easy but becomes a dependency. Strategically, lacking own compute supply is a mistake for leaders. Doomerism tied to compute limits; when Bedrock opened capacity for Anthropic, doomerism vanished, suggesting throttling.
Frontier labs avoid tin-cupping from Amazon, GCP, Azure; prefer direct customer access. Hyperscalers control 60% of compute, so game theory favors kneecapping frontier labs to catch up, like Friendster slowed by performance, enabling MySpace and Facebook to take share.
OpenAI tries displacing Stargate spend but must build own. Anthropic too. Enterprise subscription changes end token subsidies, exploding budgets. Inspecting agent code reveals slop; no 30-50% OPEX increase for nothing. Pass-through token costs amplify this; grow into it without negative gross margins.
Colossus: Elon expanding to 555,000 GPUs across three buildings, $18 billion investment. Prometheus: Meta's planned 2026, 150,000 GPUs. Elon announced deal with Cursor; renting capacity, entering hyperscale business—use for xAI, sell excess to competitors like Cursor for training.
Overbuild capacity to privilege own models. Anthropic held back Mythos possibly due to compute shortages; model huge, 10-20x token cost of Opus. Saved compute for Opus 4.7, created scarcity/altruism impression as marketing. Mythos reveals coding vulnerabilities, giving patch time—genuinely altruistic but also compute-constrained.
Polymarket shows 37% chance Anthropic buys All-In podcast by year-end; $92 million volume, #3 market. Dismissed as troll, not real.
Allbirds pivoted from sneakers to AI (new bird ai); bought 8 H100s with $50 million convertible note. Stock up 450% last week to $14/share; sold brand assets for $39 million (10% of $350 million 2021 IPO raise). Valued at $4 billion peak on collective delusion, like late '90s .com name changes (BarnesandNoble.com). Liked in Silicon Valley as next Nike, tulips; rewarded growth ignoring gross margins, COGS.
Physical companies valued like software (zero incremental cost), ushered by Uber success. 2021 COVID zerk: massive money, no inflation yet; valued 2-3 years forward growth (e.g., $100M to $300M last year → 20x bigger).
Bird scooters: Valued $2-3 billion; cities cracked down—no dedicated lanes, picked 4 operators splitting allocation (e.g., 25% of 1,000 scooters each), killing economics, network effects, competition. Regulatory capture; watch for autonomous cars (New York, Boston medallions raise prices, stifle innovation).
Recent head-scratchers: Jane Street $1 billion investment + $6 billion compute deal in new neocloud. Capital markets recognize compute constraint; Bloom Energy stock vertical on nat gas solution for on-site power, quick clean air permits vs. years for grid.
Land/shell nuclear; approvals impossible. US sentiment shifting negative on AI: doomerism, job loss fears, inequality minting trillionaires. Towns vote down data centers (e.g., $6B approved, then board ousted to undo). Maine banned all. Not just Texas—insufficient power/grid/nat gas.
Allbirds canary for compute crunch; five-alarm fire for Anthropic, OpenAI needing land/power/shell or revenue hits Friendster wall.
Unpopular in 30 states; bans, hard approvals. Wildcat real estate developers sought entitlements without power solutions; communities oppose grid draw raising residential prices unless data centers bring own power.
Ratepayer protection pledge: Hyperscalers agree no new data centers without own power—neutral or additive to grid (give back off-peak).
Doomer astroturfing: Can't sell Terminator, but water use fears (untrue). Funded by tech billionaires.
Anthropic allied with doomers/NIMBYs, avoided own data centers to hinder OpenAI, xAI; relied on hyperscalers. Backfired; hit limits, must build own. Watch effective altruists pivot to supporting data centers.
People hate rich; data centers as temples of wealth, machinery of elites taking from poor. Populism targets them over abstract AI risks. Average consumer sees little value yet (enterprise/rebuild focus; best is ChatGPT medical advice).
Data centers portable (speed of light; cheap power, low latency). Bans push to space/Iceland/Texas. Next populism targets: private jets. Root: government debt/inefficiency destroys dollar value.
Utilities: Ratepayer pledge doesn't change model—build budget, earn ~10% on investments. Incentive: endless upgrades as local monopolies.
Utilities can bury lines underground for wildfire threats, costing $10 billion, and the PUC must approve, allowing 10% returns on that investment. The utility business model incentivizes raising prices over time, similar to insurance companies. Behind-the-meter solutions are important. Data centers create blue-collar jobs in construction, with capex increasing, not decreasing—Jensen Huang projects ongoing high spending. Tens of thousands of new construction jobs, 25-30% higher wages for electricians, carpenters, drywall hangers, concrete pourers, road builders, equipment installers.
Data center construction jobs are temporary unlike permanent fab jobs. 100 data centers contested, 40% canceled (rate doubling this year), total economic value $162 billion. Permanent jobs from energy and semiconductors powering them. Crusoe (Chase) and CoreWeave (Michael) bring their own energy (BYOE): nat gas, diesel, solar.
For data center opposition not anti-progress, build in energy-rich US allies. Worked on Gulf states building data centers with American tech for American companies; caused controversy accused of serving China. Those data centers bombed/threatened by Iran/IRGC as American/Gulf assets. Opposition was a hoax, same forces behind 'data centers use too much water' hoax. Resentments whipped up by well-funded strategic agendas. Anthropic adamantly opposed, lobbied against GCC data centers.
Top AI spokespeople: Dario Amodei (Anthropic) warns of end of days, hacking, job loss; Sam Altman called sociopath by dozen people in Ronan Farrow's 70,000-word piece. They cannot represent the industry. Focus on fixing big three: healthcare, housing, education. Need Michael Dell, Jensen Huang, Elon Musk explaining world-positive AI. AI less popular than ICE, Democrats, Iran government. Alpha School (Joe Lemond) working on education.
Game identifies overvalued startups:
- Valued at $13B, selling JPEGs (monkey images) for fake money: OpenSea (Chamath guesses correctly).
- Valued at $4B, COVID mobile talk with mid-VC advice: Clubhouse (Chamath guesses).
- Valued at $270M, juice packages for machine, 'iPhone of juice': Juicero (David Friedberg guesses).
Chamath wins trip to Temptations 2 in Cabo. Everyone forgets failed unicorns.
Eric Swalwell out of California governor's race, resigned from Congress amid allegations (innocent until proven guilty). Rumors in December/January of sending pics to employees, other issues—dismissed as rumor mongering then confirmed. Broad knowledge held back, coordinated release. Victims waited for right moment. Democratic machine siphons public money to supporters' interests.
California jungle primary: top two advance regardless of party. Republicans (Hilton, Bianca) polling 14-15%, would take runoff if held today. Democrats fragmented, needed to winnow field, avoid oppo dump later.
Similar to Biden muscled out post-debate: Pelosi told him hard way or easy way; he tweeted resignation (possibly auto-pen). Pelosi mentored Swalwell 20 years, protected him from Intel Committee ouster over Chinese spy Fang Fang. No button pushed without her sign-off.
Ro Khanna traded $600M stock, more frequently than Citadel, beat Nancy Pelosi's returns this year. Supported free speech on Twitter, diplomatic Ukraine track. Wealth tax now viable (Overton window shifted). Reg FD doesn't apply to Congress; they trade on committee info.
Nancy Pelosi's decades-long returns generational, better than Buffett post-Reg FD (bimodal: pre-Reg FD off charts, post kneecapped). Top investors: Warren Buffett, Stan Druckenmiller, Nancy Pelosi. Berkshire Hathaway sits on $300B cash amid Shiller/Buffett index peaks, market dispersion (8-9 companies at highs).
Market at all-time highs in week 7 of war, $100B into Iran military, shrugging it off. Post-Islamabad meeting, market prices war resolving soon per Trump (excursion, objectives near achieved). No deal signed (unrealistic in 24 hours after 50-year hostility), but progress impression strong week.
By Tuesday, the market recovered all losses since the war started. It hit a new high on Wednesday and fresh highs on Thursday. The stock market, as the ultimate prediction market, prices in that the war is on track to resolve. Not speaking as a member of the administration or claiming insider knowledge—just interpreting market signals and statements from the president and vice president.
Travis: Trump uses the stock market as his weather vane. VIX is high, volatility up, but S&P trades in a tight band despite events. Trump gets people nervous with moves, then returns to practical reality. He prevents S&P from dropping too low. If market rises too high, he makes it hard on himself, then brings it back. Traders are adapting. Like playing poker with Alan Keating—loses big but wins key pots to recover.
Chamath: Shiller PE at near all-time highs. Buffett Index (US equities / GDP) at all-time highs—signals risk off. But dispersion shows conflicting signals: market up 5% in first half of April (typically leads to +32% rest of year); already up 7.5% in first half. Data supports any bias, creating danger. Personally risk off, waiting for IPOs to de-lever and take chips off table. Crucial: SpaceX IPO ASAP; one of Anthropic or OpenAI to front-run IPO first.
Reasons for AI Company Valuations
Traders value AI deployment unlocking unlimited intelligence, making top employees 10-30x more productive—unprecedented (Microsoft Office: 30% efficient; internet: 50%). Nobody knows how to harness fully yet, producing slop. But 10-20% do: they dominate in organizations and 600 portfolio companies.
J-Cal: Efficiency boom at Meta, Uber, Airbnb etc. will yield insane earnings. Not seen tsunami of revenue/profit personally yet. Examples in private companies:
- MicroOne: builds dark pools of data for LLMs, uses AI to find great people for corpuses—on a tear.
- Tax GPT: 6-7% of accountants use it, whipping productivity.
Chamath: Small companies nibbling edges won't build multi-trillion caps. Big companies laggards not because dumb, but sophisticated/complicated businesses. Prove it works in big league use cases, or it's a toy.
Sax: Closer to J-Cal. People figuring out business value; change management hard, especially in big companies—many McKinsey transformation projects failing. Bottom-up activity interesting: coding models at new level, generating exponential revenue. ROI now at model layer (previously questioned due to datacenter capex). Bullish overall.
Chamath: In every wave (e.g., mobile), consumer showed up at scale with profitable businesses like Google/Facebook (profitable in first few years). Need scaled profit examples in enterprise to underpin trillion-dollar valuations—no great examples yet.
Travis: Valuation metrics high—adopt conservative posture. But if AI real, upside real for crushed traditional companies. Details matter; transforming complicated businesses hard. Change management key for autonomous enterprise—human issue with middle managers, bureaucrats, undocumented processes.
Public founder-led tech companies: CEOs fired up on productivity, faster feature rollouts via pro-AI culture. Agents not smart yet—no AGI. Need human in loop; agents do repetitive tasks, lack taste/novelty, get lost (e.g., investing agents needed training not to bet both sides).
Bonus round:
- Startup raised $900M at $9B valuation, dissolved 2018; deep-voiced founder CEO serving 11-year federal sentence. Theranos (Travis correct).
- Short-form mobile-first streaming raised $1.7B from top Hollywood/SV investors (Jeffrey Katzenberg); shut down in 6 months. Quibi (Sax wins).
Prize: pied-à-terre in East Austin by airport.
Add 50-100 seats to sold-out liquidity event ($500 cap). All In Summit tickets on sale at allin.com—don't get shut out.
Keep All-In Podcast in your library
Save the videos and channels worth coming back to, and find them again in one place.
