How Anthropic Passed OpenAI on Revenue by Selling to the Enterprise

Anthropic has overtaken OpenAI on revenue, and the more useful question is not whether it happened but why. On the latest disclosures, Anthropic is running at a ~$47B annualized run-rate (a May 2026 reading), roughly 35% ahead of OpenAI's most recent $25-33B figure for 2026. The headline is a reversal; the mechanism underneath it is a divergence in business model that has been building for more than a year. Anthropic sells mostly to enterprises and developers. OpenAI sells mostly to consumers. In 2026, those two curves stopped tracking each other.

Roughly 80% of Anthropic's revenue is enterprise plus API, led by Claude Code and agentic coding workloads that bill by consumption and scale with usage rather than seat count. OpenAI's base is consumer-heavy ChatGPT, where reported paid conversion sits near 5-6%. That difference compounds. Enterprise and API revenue expands as customers deploy models into production, wire them into workflows, and let agentic systems run longer jobs — each of which raises token consumption without a new sales cycle. Consumer subscription revenue, by contrast, is gated by conversion and churn: you monetise a small slice of a very large free base, and growth means persuading more individuals to pay.

The customer roster shows where the money concentrates. Per Fortune, 8 of the Fortune 10 and roughly 70% of the Fortune 100 are Claude enterprise customers, and more than 1,000 companies each spend over $1M a year. That is a different revenue quality than a consumer app: fewer, larger, stickier contracts with expansion built in. It is also why Anthropic's trajectory looks near-vertical — reportedly ~$10B end-2025, ~$30B in April, ~$47B in May 2026 — a pace that works out to adding roughly $96M in annualized revenue per day at the top of the curve.

The contrast is easiest to read side by side. The point is not that one number is larger; it is that the two companies monetise fundamentally different surfaces:

Read that way, the run-rate gap is a downstream effect of the mix, not an independent fact.

The comparison deserves a discipline that most coverage skips. These are self-disclosed annualized run-rates, not audited GAAP revenue. A run-rate takes a recent period — often a single month — and multiplies it out to a full year; it captures momentum but assumes the latest month holds, which for a company adding tens of billions in months is a strong assumption. The two companies also disclosed in different windows, so this is not a clean same-day A-versus-B. And the $47B May figure is partly a Sacra forward-extrapolation, while the ~$30B April reading sits in a separate reporting window. The honest framing is 'ahead on revenue run-rate, on the latest disclosures' — not 'earns more than OpenAI, audited.' None of this makes the lead illusory. A run-rate that has roughly quintupled inside six months, backed by named Fortune customers and consumption-based billing, is a real signal of demand. It simply is not the same object as a booked annual figure, and analysts should hold both truths at once: the direction is unambiguous, the precision is not.

For a company that has filed confidentially for an IPO targeting around October 2026, the mix matters more than the headline run-rate. Enterprise and API revenue is what public-market investors reward: it is recurring, expansion-driven, and easier to underwrite than consumer subscriptions that hinge on conversion campaigns. Anthropic's reported Q4 2025 profitability is the more consequential data point than any single run-rate reading, because it tells prospective shareholders the unit economics can hold even while the company reinvests hard in compute. When the S-1 becomes public, the numbers will convert from run-rate to audited GAAP, and the enterprise concentration that powers the growth will be re-priced as both a moat and a dependency.