Anthropic now uses so much of its own AI that, priced at list rates, the monthly bill would be larger than its payroll before stock pay. That line first appeared in the S-1 registration statement the company filed with the US Securities and Exchange Commission on 26 October, and it is repeated in the amended version filed on 2 November, ahead of its planned share listing. The amendment adds third-quarter revenue of $17.0 billion and a new warning about the pace of AI progress. The numbers come with real caveats, and they also give other companies a useful way to think about their own AI use.
What the amended filing adds
An S-1 is the document a company files before selling shares to the public, and an S-1/A is an amendment to it. The original S-1 of 26 October left the third-quarter revenue figure blank and based its growth comparison on the quarter to 30 June. The amendment of 2 November fills in the quarter that ended on 30 September:
- Revenue was $17.0 billion for the three months to 30 September 2026.
- Revenue grew 1,127% year over year in that quarter, while compute and infrastructure expenses grew 382%. In the original filing, the same comparison for the quarter to 30 June was 1,368% against 300%, so the gap between revenue growth and compute growth narrowed in the third quarter.
Both versions of the filing carry the figures on Anthropic's own use of Claude:
- The value of Anthropic's own internal Claude usage at list prices grew more than 250 times between January and September 2026.
- In April 2026, that monthly value first exceeded monthly personnel expense, excluding stock-based compensation.
- Over 90% of contributions to Anthropic's codebase in September 2026 were written by Claude, and average daily code per engineer in the third quarter was over 45 times the 2024 average.
The caveats that matter
These are Anthropic's own figures, in a document written to support a share sale, and several qualifications are worth keeping in view. The internal usage figure is valued at list prices, which is what an outside customer would pay, and the filing does not say what that usage actually cost Anthropic to run. The personnel figure excludes stock-based compensation, and that is a large exclusion. Of $4.9 billion in personnel expense in the first nine months of 2026, $3.3 billion, about two thirds, was stock-based compensation. And "contributions to our codebase" is the company's own measure, so it does not tell you how much of that code is routine or how much human review it needed.
The filing also shows the scale of the costs behind the growth. Anthropic reported a net loss of $104.0 billion for the nine months to September. Most of it, $94.3 billion, came from changes in the fair value of its convertible notes and similar instruments as its valuation rose, and the operating loss for the same period was $1.3 billion. The filing also lists purchase commitments and leases worth about $895.4 billion, substantially all for compute and infrastructure, with about $447.5 billion expected before the end of 2030.
A company that still hired
The filing does not describe Claude replacing staff. Headcount grew from about 940 at the end of 2024 to about 5,300 by 30 September 2026, and the company says trailing twelve-month revenue per employee rose from $1.4 million to over $6.6 million over the past year. Anthropic's argument is that each person does far more with Claude, which it calls operating efficiency, while it keeps hiring.
The amended filing also carries a warning that was not in the original. It says internal estimates and expectations for future models, rather than current ones, indicate that Anthropic "may imminently meet" the definition of "dramatically accelerated progress attributable to AI" in its Responsible Scaling Policy, the framework it uses to decide which safety measures a model needs. The company says it has not designated any current model as exceeding that threshold and is still assessing the data.
Why this matters outside Anthropic
We think the useful part for other companies is the comparison itself. Few teams can say what their AI use would cost at list prices, or how it compares with the hours it saves. Anthropic has picked a simple yardstick, its own payroll, and reported it to investors. A finance lead or a team manager can borrow the same approach on a small scale, as long as they also measure the value created and not only the usage. Our earlier piece on the original Anthropic S-1 covers the customer and revenue figures from the first filing.
What to do this week
- Estimate your team's monthly AI use at list prices, using the usage pages in each tool or your API dashboard.
- Next to it, write down the hours that use saved, even as a rough estimate per person.
- If you cannot measure either number yet, set up the tracking now, because budget conversations about AI next year will ask for both.
- When you quote Anthropic's figures, quote the caveats with them: list price, no stock pay and a company filing.
For a plain guide to Claude's plans and models, see Claude in Learn.