Anthropic's $47 Billion Is About to Get a Stress Test

OpenAI's chief revenue officer publicly accused Anthropic of overstating its revenue by $8 billion. That accusation went largely unreported outside fintech circles. Anthropic then filed confidentially for an IPO. The S-1 will settle it — and the answer reshapes how every frontier AI lab is valued.

The mechanics are worth understanding, because they're not about fraud — they're about methodology. Anthropic books the full gross value of revenue flowing through Amazon Web Services and Google Cloud. OpenAI reports its Microsoft-hosted revenue on a net basis, after stripping Microsoft's platform share. Both approaches are GAAP-compliant. But at scale, the perception gap becomes real money. If Anthropic's $47 billion ARR is $37–38 billion on a net-equivalent basis, the multiple that public market investors are paying for a $1 trillion listing is meaningfully different than the $47 billion headline implies. This is the dispute the S-1 will force into the open — not because Anthropic chose to address it, but because SEC disclosure requirements will demand a breakdown of revenue channels, accounting methodology, and concentration risk.

The investor-competitor problem

The revenue accounting controversy is the obvious story. The more interesting one is the structural conflict that disclosure will illuminate.

Amazon holds $42.2 billion in Anthropic convertible notes and $32 billion in nonvoting preferred stock. In Q1 2026 alone, Amazon booked $16.8 billion in pre-tax gains from Anthropic's revaluation. Amazon is also Anthropic's primary cloud infrastructure provider — Anthropic committed to spending $100 billion or more on Amazon infrastructure over the next decade. And Amazon is an AI model competitor: its Bedrock platform sells Nova and Titan models through the same API endpoints that host Claude. Amazon is simultaneously Anthropic's largest shareholder, its biggest creditor, its primary infrastructure vendor, and a competitor for the same enterprise AI contracts.

Google holds approximately 14% equity — contractually capped at 15% — worth roughly $135 billion at Anthropic's current valuation. Google has committed $40 billion more in recent tranches. Google is also Anthropic's second-largest cloud infrastructure partner and the developer of Gemini 3.5 Pro, which competes directly with Claude Opus for enterprise contracts.

Coverage of Anthropic's IPO has treated Amazon and Google as friendly financial backers. They are that. They are also competitors with hundreds of billions in cross-obligations to a company they're invested in. The S-1 risk factors section will be legally required to describe this in full — and the description will be the most significant document in AI finance in 2026, regardless of what the headline valuation ends up being.

What $47 billion actually represents

The revenue figure itself is striking regardless of the accounting debate. At the end of 2025, Anthropic was at approximately $9 billion ARR. By May 2026, that number had reached $47 billion. That's five-fold growth in roughly five months. The primary driver appears to be Claude Code adoption and API expansion into enterprise software development — but the speed of that curve implies either genuine demand at a scale that hasn't been publicly documented, or a measurement approach that front-loads revenue recognition in a way the S-1 will need to explain.

I don't know which it is, and I'm not sure anyone outside Anthropic's finance team does. The secondary market for Anthropic shares, by all accounts, is substantially more competitive than OpenAI's equivalent market — "rabid" was the word Fortune used — which suggests investors who do have visibility into the numbers are comfortable with them. That's evidence. It's not the same as a public filing.

The race dynamic

Anthropism filed its confidential S-1 on June 1. OpenAI filed approximately one week later, targeting a September 2026 listing at $850 billion. Anthropic's regulatory review begins first, giving it a structural lead toward an October listing. Both are running on parallel tracks toward what will collectively be the defining AI IPO moment of the decade.

The race has one asymmetry worth noting: the lab that lists first sets the comparables. If Anthropic gets to public markets in October with its numbers confirmed at scale, it becomes the benchmark against which OpenAI's September-or-later listing is measured. OpenAI's own CRO accused Anthropic of overstating revenues on a methodology that makes OpenAI look worse by comparison. The S-1 accounting choices will either vindicate Anthropic or give OpenAI's argument oxygen right before roadshow season.

What the document will reveal

Four things in the S-1 will matter more than the headline valuation: the gross-vs-net accounting breakdown and what the net-equivalent revenue figure is; the percentage of revenue flowing through Amazon and Google channels specifically; whether Project Glasswing government revenue is disclosed as a separate segment with its own growth and risk profile; and the compute cost structure that underlies the margin trajectory from 40% today to the 77% target by 2028 — a path that requires both a volume scaling curve and a compute cost decline curve to hold simultaneously.

If all four of those disclosures hold up to scrutiny, Anthropic's S-1 will make the structural argument that profitable frontier AI at scale is real and replicable. If the gross-vs-net gap is material, or the government revenue is high-concentration and unpredictable, or the margin path requires assumptions that don't survive public examination — the $1 trillion headline becomes a negotiation, not a floor.

Either way, the S-1 is the most important document in AI this year. Not because of Anthropic specifically. Because it's the first time anyone is legally required to tell the truth about how this industry actually makes money.