How Anthropic Calculates Its $3 Billion Run-Rate Revenue
A Reuters Breakingviews source reveals the formula Anthropic uses to convert recent sales into an annualized figure it calls run-rate revenue.
The $3 billion run-rate revenue figure that Anthropic has circulated in recent months does not come from audited accounts, because they don't exist: the company remains private. What does exist is a specific methodology for constructing that figure, and until now nobody had explained it in detail. Karen Kwok at Reuters Breakingviews, citing "a person familiar with the matter", did so in early March.
The mechanics, later documented by Simon Willison on his blog, work as follows: take the last 28 days of sales from consumption-billed customers, multiply by 13 to get an annual equivalent, and add the total monthly subscription income multiplied by 12. Both figures are summed and the result is what Anthropic presents as its run-rate.
A formula that blends two different business models
The method has its own internal logic: API and enterprise customers typically pay for tokens consumed, while Claude.ai Pro and Team users pay fixed monthly subscriptions. Since billing cycles are different, Anthropic has chosen to annualize them with different multipliers rather than applying a single one.
Using a 28-day period instead of a calendar month is not arbitrary: months have between 28 and 31 days, which introduces noise when comparing periods. With 28 days the denominator is always the same, and the multiplier 13 (rather than 12) adjusts the difference so the total equals a full year of 364 days. This is standard practice in SaaS financial analysis with irregular billing.
Why understanding the formula matters
Without access to audited accounts, investors, competitors and analysts depend on the figures the company itself chooses to share. The problem with run-rate as a metric is that it amplifies the present: if the last four weeks have been exceptionally strong (from a large contract, a renewal cycle, or a product launch), the annualized figure can significantly overestimate the actual trend.
This doesn't mean Anthropic is lying. Run-rate revenue is a common convention among high-growth startups to signal current business pace, distinct from ARR (annual recurring revenue), which only counts signed contracts. But the difference matters, and the specific formula (with its 28 days and multiplier of 13) was not publicly documented until Kwok published it in Reuters in late March.
Who should pay attention to this
For teams making decisions about Claude adoption in enterprise, the run-rate figure itself doesn't change anything about the product. But it does offer context about the health of the business sustaining model development: an Anthropic with genuine revenue growth has more room to maintain its launch cadence and infrastructure investment.
For secondary round investors and funds that participated in recent funding rounds (a valuation of $61.5 billion in 2025), understanding what's behind the figure is essential for gauging whether the valuation has a floor or not.
And for competitors and industry analysts, the simple fact that the formula came to light via an anonymous Reuters source says something about the state of financial communication in the sector: major bets are being made on partial information and without verifiable financial statements.
---
From our perspective, the reading is pragmatic: that Anthropic has a high and growing run-rate is a good sign for the integrator ecosystem, but it's worth not confusing a momentum metric with a solidity metric. These are different things, and in a market that remains unconsolidated, the difference ends up mattering.
Sources
Read next
Andrew Yang Bets on Startups to Lower the Cost of Living
American entrepreneur and politician Andrew Yang highlights housing, food, and telecom as sectors where startups have real potential to reduce what citizens pay.
SpaceX IPO Has Nothing to Do With Claude
The submitted article covers SpaceX's IPO. ClaudeWave covers the Claude AI ecosystem. There is no justifiable editorial overlap.
Google sues Chinese criminal network that used AI to defraud hundreds of thousands
Google has filed a lawsuit against 'Outsider Enterprise,' a criminal organization that used AI to send 2.5 million fraudulent SMS messages in just two weeks.