How AI labs make money: Anthropic and OpenAI revenue, explained

AI labs make money by selling access to their models: monthly subscriptions for people, per-seat plans for companies, and metered API access for developers, much of it resold through big cloud providers. As of October 2026, the latest reported number is Anthropic's: it told investors its revenue was $16.8 billion in the third quarter of 2026, according to people familiar with the matter cited by CNBC on 15 October.

What do AI labs sell?

Both labs sell the same broad mix of products: subscriptions, business seats, API access, cloud resale and coding tools.

• Consumer subscriptions. A free tier, then paid plans. Anthropic's Pro plan costs $20 a month billed monthly, and Max starts at $100 a month. OpenAI sells ChatGPT in Free, Go, Plus and Pro tiers.
• Business seats. Companies pay per user. Anthropic's Team plan starts at $20 per seat per month billed annually; its Enterprise plan charges $20 per seat plus usage at API rates, with a 20-seat minimum. OpenAI sells ChatGPT Business and Enterprise.
• API usage, priced per token. Developers pay for what their software sends and receives. A token is a small chunk of text, roughly part of a word. Anthropic lists Claude Opus 5.5 at $4 per million input tokens and $20 per million output tokens. Output costs more because the model has to generate it.
• Cloud partners. Claude models are sold through Amazon Bedrock and Google Cloud, where the cloud provider bills the customer, and through Microsoft Foundry, billed via the Azure Marketplace.
• Coding tools. Anthropic includes Claude Code in its paid plans, and OpenAI bundles Codex into ChatGPT. OpenAI's finance chief, Sarah Friar, told staff in August that its coding and work product had reached 20 million weekly users.

What is the difference between revenue, run-rate and profit?

They measure different things: money earned, current pace, and what is left after costs.

Revenue is money earned in a period, such as a quarter. Anthropic's reported figures are $4.7 billion in the first quarter of 2026, $11.6 billion in the second and $16.8 billion in the third.

Annualised run-rate takes a recent period, usually a month, and multiplies it up to a year. It is a snapshot of pace, not money already earned. Anthropic told investors its run-rate hit $65 billion at the end of July.

Profit is what is left after costs. Anthropic reportedly said it was profitable on an adjusted basis in the third quarter. "Adjusted" means the company chose which costs to leave out, so it is not the same as profit under accounting rules.

A fourth question sits under all three: whose sales count. TechCrunch reported that Anthropic counts sales made by its cloud partners in its annualised revenue and OpenAI does not. That difference explains much of the confusion over OpenAI's numbers in October.

What does it cost to run an AI lab?

Running these businesses mostly means paying for computing power. That covers training (building a model, a large one-off cost for each generation) and inference (running the model each time someone uses it, which grows with every customer).

Gross margin is revenue minus the direct cost of serving customers, as a share of revenue. Neither lab publishes an audited figure. The closest public signal is Anthropic's: it reportedly told investors it made $1.70 in revenue for every dollar spent on computing power in the third quarter, a figure first reported by the Financial Times.

On OpenAI, TechCrunch reported that leaked 2025 financials showed revenue of about $13 billion, with spending well above that.

What revenue figures have the labs reported, as of October 2026?

The biggest are Anthropic's $16.8 billion third quarter and OpenAI's run-rate of about $50 billion.

None of these comes from an audited filing. Both companies have filed confidentially with the SEC for a listing but have not made the financials public, so every number above is either something the company told investors or something a news outlet reported from documents or people familiar with them.

OpenAI's $50 billion run-rate was lower than a $68 billion figure widely reported in late September. A person familiar with the matter told CNBC the higher number included gross revenue from OpenAI's partners. In the same investor update, OpenAI reported 77% run-rate growth in its third quarter, and 107% for its enterprise business.

Our take

Anthropic's quarterly figures show growth that is still fast but slowing in percentage terms: revenue more than doubled from the first quarter to the second, then rose about 45% into the third. The $16.8 billion quarter is close to a quarter of the $65 billion July run-rate, so those two claims hang together.

The OpenAI episode is the warning. A $20 billion gap opened up not because demand changed, but because two sets of people counted partner sales differently. Until both labs publish audited accounts, compare like with like: quarter with quarter, run-rate with run-rate, and check whether cloud resales are in or out.

What to watch:

• The prospectuses. A public S-1 filing would give the first audited revenue, costs and margins. Reuters has reported an Anthropic listing in November; OpenAI's finance chief has pointed to 2027.
• Gross versus net revenue. Whether filings book cloud-partner sales at full value or only Anthropic's share.
• Adjusted versus reported profit. Which costs the "adjusted" figure leaves out.
• Compute spend. Whether the $1.70-per-compute-dollar ratio rises as models get more efficient, or falls as each new model costs more to train.