Chamath: "Anthropic's IPO is where the AI build switches from debt to stock"

Chamath Palihapitiya called @nodiligence's post a "Good read…" @chamath. The post links to the article and gives its stance in one line: Anthropic is "probably the best business in AI right now," but the author "wouldn't touch it on day one" because the offering is "set up for suppliers and insiders to sell into the index buying." @nodiligence

The article draws on S-1 figures: $974m of Q2 operating profit, an 85% gross margin and $637bn of compute contracts. It adds that most suppliers are also lenders and shareholders @nodiligence. Its main argument is that the IPO marks the point where public equity, not credit, becomes the marginal source of funding for the AI build-out @nodiligence. In the author's view, the value goes to whoever gets paid in cash up front, mainly TSMC and Google. The losses fall on whoever lent long-term against short contracts: neocloud bondholders, data-centre bonds and Oracle @nodiligence. Yesterday's founders read the S-1 for operating strength. This adds a skeptical read of the financing structure.

Patrick O'Shaughnessy: Cliff Sosin, on the podcast and in an essay

O'Shaughnessy is promoting his own Invest Like the Best episode here. He calls Sosin "one of my favorite people to call, ask some big questions, and then shut up and listen." The episode covers the limits of intelligence, owning the things that complement it, Meta and the frontier labs, capital markets, and why Sosin keeps 80% of his portfolio in Carvana @patrick_oshag. It's on YouTube, Spotify and Apple @patrick_oshag.

He also quoted Sosin's essay "Maybe Intelligence Ain't All That." Its central claim: "The limit is contact with reality." A smarter reasoner fills gaps between known facts but "doesn't produce new facts," so most problems are still limited by how fast you can build something and test it @patrick_oshag.

Shorter mentions

• Brad Gerstner shared Collision's video with United CEO Scott Kirby and Ben Thompson. He said he recruited Kirby to United in 2016 and called him "one of the smartest CEOs in any industry I know" @altcap. Topics include capex, route planning, the Starlink decision and credit card economics @collision.
• Vinod Khosla called an Encord clip of Pim De Witte "a strong view on world models" @vkhosla. De Witte argues models should learn directly from data the universe produces, not from simulators, and that the training runs are "tiny compared to frontier LLMs" @encord_team.
• Tim Sweeney called the CFSG formalization, "43M lines of Lean proof," amazing. He said he wished John Conway were still alive to see it @TimSweeneyEpic.
• Farnam Street called Shane Parrish's conversation with Steven Kotler "Insanely useful" @farnamstreet. It covers flow, picking challenges that hold your attention, and focusing on work you don't feel like doing @shaneparrish. This is Parrish promoting his own show.
• Garry Tan shared @grepdotai's "THE SWARMS ARE COMING", calling agent swarms "a big unlock" because one agent "can only do so much" @garrytan.
• Andrew Chen published his own essay, "New Founder Archetypes in the AI Age" @andrewchen. It weighs reasons for and against five types: solo founders, non-technical founders, domain experts, young builders and content creators. One example of a "why not": "Hoteliers didn't create Airbnb" @andrewchen.