OpenAI's Identity Crisis, Datacenter Wars, Market Up on Iran News, Mamdani's First Tax, Swalwell Out
Core four + Travis Kalanick on the Mamdani pied-à-terre tax and the London-style hollowing-out it'll cause; OpenAI's leaked Dresser memo attacking Anthropic's $30B run rate as 'inflated', the secondary-market flippening where Anthropic now prices higher than OpenAI; codex vs claude for long-horizon work; the quiet violence spike (Altman's house firebombed, Ken Griffin dog-whistled); and the network-effects gap that should worry OpenAI.
Key points
- Mamdani's first tax: a ~3.9% annual levy on pied-à-terre / second homes above $5m across Manhattan. Target is elastic demand (non-residents); predicted to crash the high-end market the way London's stamp tax did, with wealthy capital redirecting to Zurich, Lugano, Milan.
- Mamdani publicly pointed at Ken Griffin's Manhattan apartment in a video that has been viewed 30-50m times — 'a dog whistle to crazy people' in the same week a Molotov cocktail and bullets landed at Sam Altman's house. Both sides need to think about what political rhetoric produces.
- Counter-analysis: Austin has doubled in population while rents have fallen because Texas lets you build on your own land. The affordability gap in blue cities is a buildability problem — the people blocking supply are the ones claiming to care about affordability.
- OpenAI's chief revenue officer Denise Dresser sent a leaked four-page memo attacking Anthropic: claims $30B run rate is inflated by ~$8B of revenue-share with model providers, positions Anthropic as 'fear, restriction, and the idea that a small group of elites should control AI.' Clear escalation.
- Secondary markets now price Anthropic above OpenAI for the first time — 'the flippening'. Some investors argue OpenAI would need to IPO at $1.2T for the $850B round to make sense; there are reportedly no buyers at $850B today.
- Code product split: Codex is better than Claude for long-horizon complex tasks. Claude ensemble is more reliable for everyday work. Implication for OpenAI: if they separate consumer and enterprise, consumer alone could be $3-4T and enterprise via Codex another $2-3T — $7-9T in the fullness of time.
- Travis Kalanick's framing from Uber: at the same size, being out-grown 3-5x means losing network effects around compute, tokens, and reinforcement-learning data. If you're OpenAI you should be very worried about Anthropic's growth rate.
- OpenAI hired Peter Steinberger (architect of OpenClaw) — positioned by cynics as a way to internalise his next innovations rather than have them remain open-source.
- Coming: new xAI model ('very cool stuff' per Chamath's visit), new OpenAI model codenamed Spud (~75% probability released next week per Polymarket), Meta's Muse Spark from Alex Wang's Superintelligence Labs.
- Swalwell out of Congress (implied from title).
Notable quotes
Anthropic's story built on fear, restriction, and the idea that a small group of elites should control AI.
You have ChatGPT, a 1 billion user business growing 50 to 100% a year. What are you doing talking about enterprise and code? It's a deeply unfocused company.
Prisoners don't create growth. Prisoners, prisoners don't create growth.
If you let people build to satisfy the demand, you won't have this problem.
Themes
- Blue-city real-estate flight and the London precedent
- OpenAI's identity crisis and the Anthropic flippening
- Political rhetoric and the security of tech executives
- Network-effects gap between OpenAI and Anthropic