Anthropic’s IPO shows that companies pay for AI that works, not for chats

Anthropic, the company behind the AI assistant Claude, is going public. On 26 October, it published its prospectus with the US Securities and Exchange Commission (SEC), and an amended version followed on 2 November. The shares are to trade on Nasdaq under the ticker ANT and will also be offered in some European countries and in Switzerland. The date and price range haven’t been set yet.

The prospectus is interesting for another reason: for the first time, one of the big AI providers has opened its books. The figures show what businesses really spend money on when it comes to AI. Whether the stock is worth buying is not something I assess here.

Ahead of its IPO, Anthropic grew its revenue twelvefold in one year

In the third quarter of 2026, Anthropic generated 17.0 billion US dollars in revenue, compared with 1.4 billion a year earlier. Since the start of 2026, revenue has exceeded the cost of compute and infrastructure, its biggest cost item, in every quarter.

The pace is slowing, though. In the second quarter, revenue grew by 144% on the previous quarter, in the third by 47%. Anthropic isn’t profitable yet either: from January to September, it posted an operating loss of 1.3 billion US dollars. Most of the net loss of 104 billion comes from changes in the value of convertible notes and similar financing, which are accounting entries, not money spent.

AI makes money where it gets work done

Four in five dollars come from usage-based billing. Customers pay for the computing work Claude does for them, for example when companies build the model into their own software, customer service or workflows. Subscriptions at a fixed monthly price, as many people know them from the chat window, account for only just over a fifth.

One comparison in the prospectus makes this especially clear. Users of Claude’s enterprise version who didn’t use the coding assistant Claude Code spent less than 75 dollars a month on average in the third quarter. The top 5% of users with Claude Code spent around twenty times that. If you let AI work on its own, you use it far more than if you only ask it questions. Anthropic leads by example: in September, Claude wrote more than 90% of the code contributions inside the company itself.

Three quarters of revenue come from large business customers

At the end of September, more than 11,900 customers were paying Anthropic over 100,000 dollars a year, up from 950 a year earlier. These customers account for 74% of revenue. They include all ten US companies with the highest revenue and more than half of the 500 largest. From January to September, customers outside the US accounted for 29% of revenue.

Germany shows the same pattern. According to the Federal Statistical Office, 57% of companies with more than 250 employees used AI in 2025, compared with 23% of those with 10 to 49 employees. You can read more in my article on AI in small businesses and the trades. Anthropic has just founded its own services company for mid-sized businesses, together with private equity firms.

Why Anthropic needs money from the stock market

Anthropic has made commitments of 895 billion US dollars for data centres and compute, due through 2045. That is more than six times the capital the company has raised so far. In the twelve months after the end of September alone, 69.8 billion fall due.

For customers, this arms race has a pleasant side: the work is getting cheaper. According to the prospectus, the models released in September do the same task 25 to 40% cheaper than their predecessors, partly through lower prices and partly because they work more efficiently.

What small businesses can take from the prospectus

The tools large corporations use are open to small businesses too. The difference is whether AI only runs in a chat window or is built into your own workflows.

1. Give AI a fixed task. For example, sorting enquiries, drafting quotes or confirming appointments, rather than just asking it the odd question.
2. Keep your data in one place. AI can only work with what it finds: customer data in a CRM, services and prices clearly structured on your website.
3. Think in tasks, not subscriptions. What does a handled enquiry cost you today, and what would it cost with AI? The cost per task is falling noticeably.
4. Check the results. Even good models make mistakes, and customer data only belongs in tools with clear data privacy rules.

I use AI every day for writing, analysis and automation. On the websites and in the CRM systems I build for businesses, content and customer data are organized so that AI can work with them. If you want to know where AI saves your business the most time, write to me.

Sources

1. Anthropic, PBC (2026): Registration Statement Form S-1/A (Amendment No. 1), prospectus filed with the SEC, 2 Nov 2026. sec.gov
2. Anthropic, PBC (2026): Registration Statement Form S-1, SEC, 26 Oct 2026. sec.gov
3. Federal Statistical Office of Germany (2025): Companies using artificial intelligence technologies by size class, table 2025 (in German). destatis.de