Reading Anthropic’s prospectus as a customer

On 26 October, Anthropic, the company behind Claude, filed publicly to list on Nasdaq under the symbol “ANT”. The price range and the number of shares are still blank, and 22 banks are named on the cover, led by Goldman Sachs and Morgan Stanley. Reuters reported that the listing is expected before Thanksgiving.

Most coverage reads the prospectus as an investor would. We read it as a customer: a company that pays for Claude, or is deciding whether to. This is not a view on the shares. It is what a major AI supplier says about itself when it must put its accounts and risks on the record.

Where the money comes from

Revenue was 41 million dollars in 2023, 386 million in 2024 and 4.6 billion in 2025. In the six months to 30 June 2026 it reached 16.3 billion dollars, against 1.2 billion a year earlier, which the filing gives as growth of 1'300 percent. Anthropic adds a sentence buyers should read as carefully as investors: “We do not expect our revenue will continue to grow at comparable rates.”

The filing gives no separate figure for the API, Claude Code or the consumer apps. It splits revenue by how customers pay. Usage metered in tokens and priced by model, which Anthropic calls “pay-for-intelligence”, brought 80 percent of first half revenue. Subscriptions brought the other 20 percent.

Most of the money comes from companies. By 30 September, more than 11'900 customers had paid 100'000 dollars or more over twelve months, up from more than 950 a year earlier. In the first half, 48 percent of revenue came through cloud marketplaces. The top five customers made up about 25 percent of revenue for the nine months to September.

What compute costs

Compute is, in Anthropic’s words, “the most significant operating cost in our business”. It cost 8.9 billion dollars in the first half, 55 percent of revenue against 201 percent a year earlier, and the operating loss narrowed to 693 million dollars. Of the net loss of 80.7 billion dollars, 72.4 billion is a change in the fair value of convertible notes and similar instruments, driven mainly by Anthropic’s rising valuation.

Anthropic no longer shows gross profit in its income statement. A note to the accounts puts cost of revenue, substantially all of it compute for serving customers, at 2.4 billion dollars for the half year. Most compute goes to research, model development and internal use, not to answering customers.

The bill is growing. Since the start of 2026, Anthropic has signed agreements with minimum payments of about 417 billion dollars, mostly for compute equipment, data centres and cloud capacity, with terms of up to 20 years. It warns that margins “may compress” as new capacity arrives ahead of the revenue it will support. Better news for buyers: it says Claude Opus 5.5 cut customers’ cost per task by about 40 percent compared with Opus 5.

The risks it names

The risk factors read like a supplier audit written by the supplier, and four themes matter most to a buyer. The first is capacity. Anthropic says it has “consistently operated in a compute-constrained environment where customer demand exceeds available compute”, and a surge in demand helped cause an outage in March 2026. It also writes: “We have also changed, and may in the future change, usage limits, rate limits, or model and product capabilities.”

The second is regulation. In June 2026, after the US Department of Commerce imposed worldwide export restrictions on Fable 5 and Mythos 5, Anthropic disabled both models for all customers until the restrictions were lifted. Its legal dispute with the US Department of War, which designated it a supply chain risk, is unresolved.

The third is behaviour: Anthropic states that models, its own included, have shown “unexpected strategic or self-preserving behaviors”, among them attempts to resist shutdown. The fourth is dependence. Some of its cloud suppliers are also investors, customers and competitors. And among the reasons customers may not expand their usage, Anthropic lists a “preference for multi-vendor AI strategies in order to limit dependence on any single provider.”

Who stays in control

Anthropic is a public benefit corporation, so its board must balance shareholders’ financial interests with the interests of those affected by its conduct and with its purpose, “the responsible development and maintenance of advanced AI for the long-term benefit of humanity.” A Long-Term Benefit Trust of five “financially disinterested” trustees, set up in 2023, elects the directors who form a majority of the board.

A single Class F share goes to a company formed by the seven co-founders. It carries 50.1 percent of the voting power on general matters. The Class A shares offered to the public carry one vote each, and Anthropic writes that their holders “generally will not be able to affect the outcome” of shareholder votes. TechCrunch called the structure controversial because voting power does not follow economic ownership. Anthropic’s reason is to let its leaders focus on the long term “rather than facing pressure to prioritize short-term results”.

What a listing changes

A listing brings customers two things. First, disclosure. Once listed, Anthropic will report under US securities law, with figures every quarter and audited accounts every year. How much compute is committed, how much revenue rests on a few customers and how margins move will be answered on a schedule.

Second, pressure. Quarterly figures invite quarterly judgements, and the governance is built to resist them. Anthropic warns investors that its product decisions “may not prioritize short-term financial results”, citing models it delayed or released with limits until safety testing was done. In our view, a supplier that says this to shareholders is telling its customers something useful too.

The practical lessons for a buyer are three. Fix prices in writing where you can: Anthropic says a growing number of enterprise customers commit to minimum spend “in exchange for a committed price list.” Expect your model to change, since models released in the previous six months made up over 80 percent of Anthropic’s revenue in the third quarter. And keep your knowledge, rules and workflows in a form that can move between models.

Where we stand

That last point is how we build. Twin, which is live, connects a company’s knowledge, rules and tools once and makes them work in Claude, ChatGPT or Codex, and we integrate Anthropic, OpenAI, Google and other partners. When data must not leave the company, we set up Eigna, a private model of your own made by our partner Eigna Labs, now in early access. Anthropic’s filing shows a strong supplier, and, in its own words, why customers may prefer not to depend on any single one.