OpenAI vs Anthropic IPOs, Anthropic $3T, Zuck's Price War, China Ends Open Source?, Trump Accounts
Summary
SpaceX’s $75 billion raise at a $1.75 trillion valuation gave Anthropic and OpenAI a working blueprint for trillion-dollar IPOs. Brad called the deal “textbook,” saying SpaceX now trades around $2 trillion on roughly $35 billion of forward revenue after pioneering staged lockups and early index inclusion. Rumored year-end revenue above $100 billion could make Anthropic a blockbuster, while OpenAI’s corporate restructuring probably puts it second despite renewed momentum toward roughly $70 billion of revenue this year.
The largest pre-IPO risk is whether enterprises can prove a return on token bills that are compounding faster than productivity. Chamath’s CTO said their token costs were doubling every 45 days while downstream productivity had improved “maybe 5% max”; Brad’s Fable 5 exercise suggested the S&P 493’s actual AI-related ROI was somewhere between 0% and 2% after separating out pricing, buybacks and chip revenue. Chamath’s timing call: labs should list before buyers broadly ask, “Who is paying you this, and can they sustain paying it to you?”
Brad’s counter-case is that intelligence addresses every employee and every company, making today’s spending look early rather than excessive. Millions of customers are independently buying AI, Nvidia already uses AI throughout next-generation chip design — “the machine is building the machine” — and frontier advantages may unlock revenue rather than merely cut costs. Brad provocatively argued that a lab exiting the year above $100 billion could grow 3-5x again next year, an expansion without precedent “in the history of the world.”
Falling token prices are increasing consumption while sophisticated enterprises build routing layers to control which models capture the spend. Jason cut one workload’s token cost 95%, then moved agents from daily to hourly runs; Brad said token prices have fallen roughly 90% during each of the past two and a half years, invoking Jevons paradox. Yet Sacks said open source’s measured enterprise-spend share fell from 19% to 11% because most companies lack DoorDash- or Coinbase-level routing expertise: “The spirit is willing, but the flesh is weak.”
The likely end state is premium frontier intelligence for uncertain, consequential work and post-trained open models for mature, repeatable tasks. Decagon reportedly routes 90% of usage to customized open models, while DoorDash sends lower-level work to Kimi 2.6 and reserves Fable for harder tasks. Brad’s economic test was blunt: when replacing a $200-an-hour consultant, paying $15 instead of $3 for a “bulletproof” result is immaterial. Jason’s non-consensus possibility was that the frontier gap might widen rather than converge.
Sovereign AI, Meta’s price war and possible Chinese export restrictions are fragmenting the model market along geopolitical lines. Chamath said no country he encountered wanted unquestioned dependence on closed American models; Meta promised Spark 1.1 at very low prices, while China reportedly considered restricting overseas access to leading models. Sacks said Washington should take steps against distillation, but warned that lower-level regulators could still make “ham-fisted” decisions that weaken the U.S. position.
Trump accounts launched as both a child-investment product and a potentially enormous direct-giving platform. Brad reported more than 1.5 million accounts and $1 billion of deposits in the first 24 hours, with each account invested in the S&P 500 and designed to receive family, employer and philanthropic contributions. Dell, SpaceX and Micron commitments were presented alongside a goal of 50-70 million accounts and $100 billion of philanthropy; Sacks argued the deeper attraction is tax-advantaged lifetime compounding, not merely “the freebies.”
Deep dive
Not yet available upstream; scheduled sync will retry.