Listen to this article
AI may have been used to generate this audio.
Press Space or K to play or pause, and J or L to go back or forward 15 seconds.Sources checked 2 October 2026.
Anthropic announced a $65 billion funding round in May at a $965 billion post-money valuation. It also said its revenue run rate had passed $47 billion that month. These figures describe different things, and putting them in one paragraph does not make them comparable. [1]
The funding round is money raised. The valuation is the value implied by the financing terms. A run rate annualises revenue at a particular pace; it is not a completed year of revenue. None of those numbers tells you the company’s profit.
Reuters reported a target of about $2 trillion in October, rather than an agreed offering price. [4] The completed May financing and a proposed later transaction need to retain their separate labels.
The sequence matters because the number can travel faster than its label. A private round, a proposed offering and a public market price involve different transactions. To discuss whether a price is demanding, we first need to agree which price and which claim on the company we mean.
That distinction makes the argument about AI being overrated easier to handle. Someone can have a good experience with Claude and still doubt that a particular price for the company will produce an attractive return. Someone else can dislike an AI product and underestimate the value another customer gets from it.
Figure announced in May | What it describes | What it does not establish |
|---|---|---|
US$65 billion | Capital raised in the round | Annual customer spending |
US$965 billion | Post-money company valuation | Cash held by the company |
More than US$47 billion | Revenue run rate at that time | Completed annual revenue or profit |
Where the customer benefit goes
Suppose a tool saves a business ten hours of work a month. This is an illustration, not a measured Claude result. The business may pay for the tool, keep some of the saved time, reduce a backlog or ask the same staff to do additional work. The vendor captures only the portion of that benefit it can charge for.
Competition can change that portion. If customers can obtain a similar result elsewhere, the vendor may need to lower its price or spend more to keep them. A company can therefore deliver a service customers value while having a difficult time earning a large margin on it.
Costs matter too. Revenue from an AI service has to pay for the resources used to supply it, along with the company’s other expenses. Revenue growth can help cover those expenses. Whether it does so depends on the actual economics; the growth rate cannot answer the question on its own.
DORA’s 2026 work on the return from AI-assisted development provides a framework for evaluating adoption costs and benefits over time. It is a planning framework, not a guarantee of a particular return. [2] A buyer should do that work for its own organisation before treating somebody else’s impressive demonstration as a budget assumption.
Put an annual sales number beside the target
Here is a deliberately limited calculation. At a hypothetical value of $2 trillion, a price equal to 20 times annual sales would correspond to $100 billion in annual sales. At 40 times, it would correspond to $50 billion. These are divisions, not forecasts, and neither chosen multiple is being offered as a fair one.
The exercise is useful because it makes the disagreement explicit. One person may be expecting much more revenue. Another may be willing to pay a higher multiple for the same revenue. Both can sound equally enthusiastic about the technology while making quite different financial assumptions.
Now add the cost of producing that revenue. Two companies with the same sales can leave very different amounts for owners after supplying the service, paying staff and funding further development. A revenue multiple compresses those expectations into one number. It does not remove the need to think about them.
The May run rate is useful information about the pace Anthropic reported then. Comparing it with a target based on some future year would mix dates unless the assumed growth is written down. A chart that labels both simply “revenue” would give the reader more confidence than the comparison deserves.
The price adds another requirement
Now consider an investor. Even a company that eventually earns substantial profits can disappoint someone who bought a claim on those profits at a demanding price. The price paid, future ownership dilution and the time before cash reaches shareholders all affect the outcome.
There is a useful discipline here: write down the assumptions without using the word “AI”. Estimate what customers pay, what supplying the service costs and what remains after other spending. Then ask how much growth the price already assumes. If the argument becomes impossible to explain without referring to the size of the technological change, it probably needs more work on the business.
Anthropic confirmed its confidential filing for a proposed IPO in June. The announcement leaves the offering dependent on the SEC process and market conditions. [3] A proposed listing is an opportunity to examine financial disclosures as they become public; it is not evidence that any particular valuation is justified.
A useful sceptical case would therefore name what is expected to disappoint: demand, retention, pricing, costs or the terms on which new capital is raised. A useful optimistic case would show how the company can meet those demands while competitors and suppliers respond. “Everyone will use AI” leaves too much unspecified to distinguish the two.
A customer’s evidence can be quite local: a recurring task now takes less time at an acceptable cost. An investor needs an account of future earnings, ownership and price. The customer does not have to solve the second problem to keep using Claude, and the investor cannot use the first result to skip it.
Sources
One email when we publish. Research, product decisions, and what teams report back.
