Anthropic's $2T IPO: Pushed to November, Leaked S-1 Shows $4.6B Revenue Against $518B of Compute Commitments

Anthropic's $2T IPO: Pushed to November, Leaked S-1 Shows $4.6B Revenue Against $518B of Compute Commitments

The open question from the Sept 28 thread — what a reported ~$2T Anthropic IPO implies for the AI complex — now has hard specifics, and they are extraordinary on both sides of the ledger.

The filing and the timing. Anthropic confidentially submitted a draft S-1 to the SEC on June 1, 2026, with an IPO target window of Q4 2026, possibly October (per CNBC reporting). Per The Motley Fool via Yahoo (Oct 6), the listing has slipped: "Anthropic Pushes Its IPO to November as Investors Eye a $2 Trillion Valuation... The AI giant is expected to fetch a jaw-dropping $2 trillion valuation, about double the $965 billion valuation set in May of this year." Reports put the raise at up to $100 billion.

What the leaked draft filing shows. Per ValueAdd VC's read of the leaked prospectus: "an $8 billion operating loss against $4.6 billion in revenue and plans to spend roughly $500 billion on [infrastructure]¹." Per TEMA ETFs' analysis of the same filing: "The filing disclosed $518 billion in infrastructure commitments for the years ahead, approximately 80% of which are non-cancellable." That's the number that reframes the whole AI-capex debate (see The Capex Answer Is In: No Slowdown — Big Four at ~$725B and Rising — as the Debate Shifts From "Will They Spend" to "What Return on $1.75T of Debt"): a single model lab carrying $518B of compute lease commitments — ~$414B of it non-cancellable — against $4.6B of trailing revenue. The $2T valuation "assumes that growth continues, customers stay, contracted capacity is used, and margins rise enough to generate tens of billions of dollars in free cash flow a year" (PitchBook via Morningstar).

The bear read is already on the record. New Constructs, via CNBC (Oct 7): "While Anthropic offers more to society than WeWork ever did, at a $2 trillion valuation, its IPO presents far bigger risks and is positioned to be a far bigger rip off of U.S. capital markets... the IPO's purpose isn't to provide wealth for public markets investors, but rather liquidity for the company's Wall Street backers."

Why this matters for the complex:

• It's the AI-trade's biggest near-term supply event. A ~$2T market-cap debut (roughly a third of NVDA's $5.54T, more than Intel and Qualcomm combined several times over) lands in November into a market that just showed (see The AI Trade Passes Its First Live Stress Test: OpenAI's $20B Revenue Whipsaw) how violently it trades on OpenAI revenue credibility. Anthropic's leaked $8B operating loss on $4.6B revenue is the same credibility question, formalized in an S-1.
• The comps get a new anchor. At $2T, Anthropic would debut at ~435x trailing revenue. Every AI-complex multiple — NVDA at 29x earnings, MU at 13.8x — gets measured against a private-market mark that assumes tens of billions of future FCF from a currently loss-making lab.
• The $518B non-cancellable commitment stack cuts both ways. Bulls read it as contracted demand locking in the compute oligopoly's revenues for a decade; bears read it as the clearest number yet on how leveraged the AI buildout is to a handful of customers who themselves aren't profitable. Either way, it's the largest single datapoint yet on the $1.75T debt-funded capex question.

The timing slippage from October to November — into a market showing AI-fatigue jitters, with an Nvidia-backed Firmus Grid already pulling its own ASX IPO on "recent market volatility" — is itself a signal about the window.