Lex in depth: Anthropic at $2tn isn’t far-fetched
Lex walks through how to value Anthropic ahead of a possible ~$2tn IPO — and whether that price is hallucinatory or conservative. Foley notes Claude’s unofficial annualised revenue around $65bn by August (kid gloves advised), versus Anthropic’s own 18-month-ago 2027 revenue expectation of just $12bn; some investors now talk of US$207.1B run-rate by end-2027, which would put $2tn at ~7× 2028 sales (below Microsoft; SpaceX trades ~16×, implying ~$5tn on that peer). A TAM framing borrows SpaceX’US$16.6T “enterprise apps” claim and Anthropic/WSJ talk of ~$30tn (Morgan Stanley’s knowledge-work TAM up to ~$60tn) to show how small share assumptions can justify multi-trillion equity values — while Foley stresses TAM history (Uber, WeWork) is littered with overreach. The counterweight is “commodity cognition” (Palantir’s Karp): if models become fungible amid China undercutting, Meta/Alphabet subsidies, price cuts, routing software and Nvidia’s Hugging Face open-weights push, labs may keep only thin rents while apps/data owners capture value. Recursive self-improvement is framed as both moat and unpriceable risk that could scramble asset prices. Foley notes labs moving up the stack (Anthropic legal plug-in / SaaSpocalypse; OpenAI’s Astra for Law) and that Anthropic/OpenAI, unlike SpaceX, are essentially pure-play model companies — so valuation hinges on whether they stay toll-keepers or get commoditised.
Why it matters
Dealroom-core capital-formation thesis: the most rigorous public attempt yet to underwrite Anthropic’s ~$2tn IPO maths — multiples vs SpaceX/Microsoft, TAM inflation, and the commodity-cognition risk that decides whether frontier-lab equity or application/infra layers capture AI returns for European and US investors.
Executive takeaways
Revenue trajectory disputed: unofficial ~$65bn annualised by Aug vs Anthropic’s prior $12bn 2027 plan; bull investors float ~$320bn run-rate by end-2027 → $2tn ≈ ~7× 2028 sales (vs MSFT; SpaceX ~16× → ~$5tn analogy). | • TAM games: SpaceX $22.7tn enterprise-apps claim; Anthropic filings may cite ~$30tn (WSJ); MS knowledge-work TAM up to ~$60tn — small share × 10× multiple + discount can spit out $4.5tn–$10tn paper values; Foley flags Uber/WeWork TAM cautionary tales. | • Commodity cognition risk: China (Kimi/DeepSeek) price undercut; Meta/Alphabet can subsidise; routing away from expensive tokens; Nvidia×Hugging Face open-weights — Karp argues value accrues to data/apps/deployers, not model smiths. | • RSI double-edged: intelligence-explosion moat vs unpriceable systemic risk that could invalidate today’s asset-price frame. | • Up-stack pivot: Anthropic legal plug-in (SaaSpocalypse trigger); OpenAI Astra for Law — pure-play labs more exposed than diversified SpaceX.
What Financial Times may be missing
Little primary disclosure from Anthropic itself (relies on unofficial run-rates, WSJ/investor colour); limited Europe/Asia LP allocation detail; thin treatment of secondary markets, employee liquidity, or how SoftBank/sovereign co-investors distort comparable multiples. Commodity-cognition counterfactual is argued more than modelled against Anthropic’s actual enterprise mix/retention.
