OpenAI and Anthropic count revenue differently, and investors are confused

OpenAI says its annualized revenue run-rate was approaching $50 billion as of late September 2026. A lot of investors had a different number in their heads: $70 billion.

That $20 billion gap did not come from fraud, a leak, or a typo in a pitch deck. It came from two companies using two different methods for counting money that flows through cloud partners, and from investors who tried to make the figures comparable.

Reports from the Financial Times and Bloomberg on October 8-9, 2026 laid out the confusion. The timing is awkward. Both OpenAI and Anthropic are being watched closely ahead of potential IPOs, and the market is trying to decide which one is bigger.

Where the $70 billion came from

Annualized revenue run-rate, or ARR, is a simple idea. You take a company’s current revenue pace and project it across a full year.

The trouble starts with what counts as revenue in the first place. Both OpenAI and Anthropic sell their AI models directly. They also sell through cloud platforms such as Microsoft Azure and Google Cloud, where a business customer buys access through its existing cloud account.

When a sale passes through a partner like that, a company has a choice to make. It can book the gross amount the customer paid. Or it can book the net amount, meaning only what it keeps after the partner takes its share.

According to the research findings, the gross-versus-net choice alone can move ARR by up to $8 billion between otherwise similar firms.

The $70 billion figure for OpenAI was not something the company put out. It was built by investors who took Anthropic’s accounting approach and applied it to OpenAI’s revenue, with the goal of comparing the two on equal terms. The reports describe that application as incorrect, and OpenAI’s own figure lands closer to $50 billion.

Two GAAP-compliant answers to the same question

Neither company appears to be breaking any rules. Both are described as GAAP-compliant, meaning they follow the generally accepted accounting principles used for US financial reporting.

The split comes down to interpretation. GAAP asks, in effect, who controls the customer relationship when a sale goes through a partner. OpenAI and Anthropic reach different conclusions on that question, and those conclusions flow straight into the top-line number.

For reference, Anthropic reported a $65 billion ARR in July 2026. Set next to OpenAI’s figure of roughly $50 billion in late September, that comparison looks like a clear lead for Anthropic. Set next to the investor-built $70 billion estimate, it looks like the opposite.

Why the AI revenue race keeps getting muddier

The cloud partnerships make this especially tricky. Microsoft Azure and Google Cloud are major sales channels for AI models, because enterprises like buying through platforms they already use. The more revenue that runs through those channels, the more the gross-versus-net decision matters.

What this means for investors and the IPO path

The immediate fallout has been some volatility in AI-related stocks, according to the research findings.

The $70 billion estimate was reportedly an attempt to compare OpenAI and Anthropic on equal footing before either one lists. Investors want to know which company is truly ahead, and right now the answer changes depending on the accounting method.

There is also a lesson here about borrowed math. Adjusting one company’s numbers to match another’s method sounds rigorous. In this case, it produced a figure that the company itself does not report and that the reports call incorrect.

For anyone tracking the AI sector, the practical takeaway is to ask a basic question before comparing headline revenue: gross or net? The answer can account for billions of dollars, and it is not always disclosed alongside the number.

For now, OpenAI has put its own figure on the table: approaching $50 billion. Anthropic’s last reported mark was $65 billion in July. The gap between those numbers, and the gap between OpenAI’s figure and the $70 billion estimate, is less a story about who is winning and more a story about how the score gets kept.