Anthropic's Run Rate Reaches $65 Billion, Roughly 7x in Seven Months

Nine billion dollars at the end of 2025. Forty-seven billion in May 2026. Sixty-five billion at the end of July. That is Anthropic's annualized revenue run rate over seven months, reported by Bloomberg on August 17, and it works out to roughly a sevenfold increase.

Investor projections cited by the Financial Times put the company's full-year 2026 revenue somewhere between $100 billion and $120 billion.

What a run rate measures

An annualized run rate is not annual revenue. It takes a recent period and extends it across twelve months, which makes it a projection wearing the clothes of a result. For a company growing this fast, the run rate runs well ahead of what actually lands in the bank during the year. For a shrinking company it flatters in the opposite direction.

That does not make the figure meaningless. It makes it a measure of momentum rather than size, and momentum is the thing worth watching here: the money is coming from enterprise budgets that did not have an AI line two years ago.

Worth separating too: revenue growth is not profit. The cost side of this business is data centers, and the bill for those keeps surfacing in the news cycle. A $65 billion run rate is evidence of how much the industry is spending as much as of how much it is earning.

Comparing to OpenAI is harder than it looks

OpenAI's annualized revenue sits around $40 billion as of mid-August, doubled from $20 billion at the end of 2025. The tempting conclusion is that Anthropic is now well ahead. Be careful with it. The two companies may not calculate the metric the same way, and how committed enterprise contracts, consumption-based usage and consumer subscriptions get counted can move a number like this considerably.

On valuation, Anthropic was last marked at $965 billion in May 2026, with an IPO target reportedly north of $2 trillion.

Read this before your next renewal

Numbers this large are usually read as reassurance: your vendor is not going anywhere, the product will keep improving. Fair enough. The other reading is about pricing behavior, and the last few weeks have shown both directions at once. Competition pushed some prices down, while other providers raised API rates or quietly ended promotional periods. We covered one such case when DeepSeek raised its API prices.

Three habits protect you regardless of which way prices move. Budget AI as a variable cost rather than a software license, because it scales with usage and not with headcount. Put the end date of every promotional or introductory rate in your calendar, since most teams discover the increase on an invoice. And keep model calls behind a thin abstraction layer with a small evaluation set that runs the same task against two providers, so that switching remains a decision rather than a project.

The expensive mistake is rarely picking the wrong model. It is building in a way that makes changing your mind expensive.

Sources: TechCrunch, Fortune