What each company has actually disclosed
● Anthropic: a regulatory filing — $81bn run-rate revenue, 11,900+ large customers, audited numbers in an S-1.
● OpenAI: reported figures — About $10bn Q3 revenue, up nearly 50%, with large operating losses, per The Information.
● These are not comparable — One is a filed annualised run rate, the other a reported quarter. Treating them as a league table is the mistake.
● The number that is comparable — Customer concentration. 74% of Anthropic revenue comes from customers paying $100k+.
Anthropic’s IPO filing is the first time either company has put audited numbers in front of a regulator. That makes comparison tempting and mostly wrong, because the two sets of figures measure different things over different periods from different sources.
What is on the record
|
Anthropic |
OpenAI |
| Source |
Form S-1/A, filed 10 Nov 2026 |
Reported by The Information |
| Headline revenue |
~$81bn run-rate at 31 Oct 2026 |
~$10bn in Q3 2026 |
| Growth |
+$10bn run-rate in one month |
Up nearly 50% |
| Profitability |
Not summarised here |
Large operating losses reported |
| Large customers |
11,900+ paying $100k+ TTM at 30 Sep |
Not disclosed |
| Concentration |
74% of revenue from those customers |
Not disclosed |
| Verification |
Regulatory filing |
Journalism, not a filing |
A run rate annualises a recent period; a quarterly figure does not. Putting $81bn next to $10bn without saying that is how most coverage this week has misled.
The comparison that does hold
Customer concentration. Anthropic disclosed that more than 11,900 customers pay $100,000 or more a year, up from more than 950 twelve months earlier, and that those customers produced $24.5bn in the first nine months of 2026 — about 74% of revenue.
That is an enterprise company. The consumer subscription is real, and it is a quarter of the business. Any read of Anthropic as a chatbot company with an API attached has the proportions backwards, and it explains why the product roadmap has gone where it has — Whiteboards, Sheets and Data, warehouse connectors, Team and Enterprise admin controls. None of that is built for individual subscribers.
What neither has disclosed
- OpenAI has filed nothing. Its figures come from reporting, which may be accurate and is not auditable by you.
- Neither publishes per-model economics. Nobody outside knows what inference costs on any specific model.
- Neither publishes churn. Growth in large customers tells you about acquisition, not retention.
What this means if you are buying, not investing
- Enterprise concentration cuts both ways. A vendor earning 74% from large contracts builds for large contracts. If you are one, good. If you are a solo developer, your priorities are not the ones shaping the roadmap.
- A filed number beats a reported one when you are assessing whether a vendor will still exist in three years.
- Neither disclosure changes the model choice. Price and benchmarks decide that. We compared the current models here.
Frequently asked questions
Is Anthropic bigger than OpenAI?
The public figures do not answer that. Anthropic filed an $81bn run-rate as of 31 October 2026; OpenAI is reported at roughly $10bn for Q3 2026. A run rate annualises a recent period and a quarterly figure does not, and only one of the two comes from a regulatory filing.
What is a run-rate revenue figure?
A recent period annualised — typically the most recent month or quarter multiplied out. It describes current momentum, not money already received, and it is not the same as annual revenue.
How concentrated is Anthropic revenue?
More than 11,900 customers paying $100,000 or more a year produced $24.5bn in the first nine months of 2026, about 74% of total revenue.
Does OpenAI publish comparable figures?
No. OpenAI has not filed with a regulator and does not publish audited revenue. The figures in circulation come from journalism.