Anthropic files for a $2 trillion IPO — the $80.7 billion loss is the decoy, the $450 billion compute bill is the risk

Mitrade Insights — Anthropic publicly filed its S-1 on October 26, and the headline numbers look impossible to reconcile: $16.29 billion of first-half revenue, up 1,304% from a year earlier, against a $80.7 billion net loss. Reports point to a valuation of $1.5 trillion to $2 trillion, a listing on Nasdaq under the ticker ANT as early as the week of November 16, and a potential raise of up to $100 billion. The question for traders is not whether Anthropic is growing. It is whether the AI trade can survive its first public mark-to-market.

What the non-cash charge hides

The $80.7 billion net loss looks like a cash incinerator. It is not. About $72.4 billion of it is a non-cash accounting charge tied to the revaluation of convertible notes as Anthropic's private valuation rose. Strip that out and the first-half loss is roughly $8.7 billion; the operating loss was $693 million, and operating cash flow was positive $3.93 billion. In the second quarter, Anthropic reported its first quarterly operating profit of $974 million.

That does not make the company cheap. It makes the headline misleading. The market's first instinct — "AI start-up loses $80 billion" — is the wrong frame. The right frame is a company that is already generating operating cash and is choosing to convert that cash, plus a lot of outside capital, into compute.

The $450 billion bill nobody is pricing

The filing discloses at least $518 billion of decade-long cloud and infrastructure commitments with six partners; roughly $450 billion is already signed and mostly non-cancelable, with about 80% payable regardless of usage. Reported components include about $111 billion with Google Cloud, $110 billion with AWS, $31 billion with Microsoft, up to $86 billion for Nvidia chips through neoclouds, and $160 billion+ for Broadcom-facilitated Google chips. Total future obligations are above $600 billion and could reach $750 billion with deals under negotiation — roughly $45 billion to $66 billion a year.

That is the real risk. Revenue can slow, API prices can fall, and customers can leave; the compute bill still has to be paid. The $80.7 billion loss is an accounting event. The $450 billion commitment is a cash event.

Anthropic itself is private until the IPO prices, so the tradeable read-through is the AI infrastructure chain. On October 26, the market bought it: the Nasdaq Composite rose 0.82%, the Philadelphia Semiconductor Index jumped 2.82%, ARM gained 6.19%, Intel 4.04%, Micron 2.98% and Nvidia 2.39%. SK Hynix's ADR added about 4%.

* Chart source: TradingView official chart screenshot, data from Cboe One (NVDA).

The chart tells the honest version of the story. NVDA bounced off $210.18 and closed at $216.00, but it is still 11.2% below its six-month high of $243.37, and the SOX is still down about 12% in October despite Monday's 2.82% rally. This is a rebound inside a correction, not a new high. The $222 line — the October 21 close — is the first test; a close above it would confirm that the IPO news is more than a one-day short squeeze.

The read-through is not uniform. Anthropic's largest commitments are to Google Cloud, AWS and Microsoft, but the marginal beneficiary of a $450 billion compute bill is the chip layer: Nvidia for accelerators, Broadcom for custom silicon, Micron for memory, ARM for design royalties. If ANT prices at the top of the range, the whole chain gets a valuation anchor. If it prices below $1.5 trillion, the anchor becomes a ceiling.

The strongest bear case is not the loss. It is the customer concentration: about one-fifth of revenue comes from two customers, and 47% of 2025 revenue flowed through Amazon and Google marketplaces — partners that are simultaneously investors, suppliers, distributors and competitors. Add the 80% non-cancelable structure of the compute commitments, and the company has limited room to cut costs if demand slows.

The strongest bull case is the cash flow. A company that generated $3.93 billion of operating cash flow in its first half, with a $974 million quarterly operating profit, is not the same as a cash-burning start-up. It is an infrastructure business with a fixed-cost base and a demand curve that is still steep.

The conditions that decide the trade are simple. Bull case: ANT prices at or above $1.8 trillion, revenue growth holds in triple digits, and NVDA closes above $222. In that world, the AI capex chain gets a public valuation anchor and the October selloff was a reset, not a top. Bear case: the IPO is delayed, prices below $1.5 trillion, or the filing's roadshow language emphasizes the $450 billion commitment more than the revenue growth. In that world, the chip rally fails at $222, the SOX retests its October low, and the market starts pricing AI capex as a fixed-cost problem rather than a growth story.

Those are the two sides. The $80.7 billion loss is not the test. The $450 billion bill is.