SAN FRANCISCO, 29 AUG 2026 — Anthropic is preparing to tell prospective IPO investors that its total addressable market exceeds US$30 trillion, a figure above the US$28.5tn SpaceX put to investors before its own offering. The number is not a forecast, and it is not really a market either.
It is an arithmetic exercise, and understanding how it is built is the only way to read it.
What a total addressable market actually measures
Total addressable market is the maximum annual revenue a company could earn if it captured every possible customer, at the prices it charges, with no competitor taking any share.
Anthropic's version is reportedly built on the full scope of work that could be performed using AI models. That is a defensible construction of the metric and it produces a number with a specific property: it counts the value of the labour, not the price anyone currently pays software to do it.
For scale, US gross domestic product is in the region of US$30tn. The claim, read literally, is that the work AI could do is worth about as much as everything the United States produces in a year.
The gap between the number and the business
Anthropic more than doubled revenue to US$11.6bn in the second quarter — a remarkable trajectory by any standard other than the one the pitch invites.
Annualise that and the company is capturing roughly one part in six hundred of the market it describes. Its own projection of US$190bn to US$200bn by 2028 would take it to under one per cent.
Neither figure is embarrassing. That is the point of a TAM this large. It is not a claim about what the company expects to earn, and reading it that way produces nonsense. It is a statement that the ceiling on growth is not the problem.
Why the SpaceX comparison is the weakest part
Coverage has led on the deal topping SpaceX's US$28.5tn. That framing treats two marketing numbers as though they were comparable measurements.
They are not measurements of the same thing. SpaceX and Anthropic each constructed their own figures, using their own definitions, for their own offerings. There is no common standard and no auditor. Ranking one above the other says something about who was more expansive in defining the denominator, and nothing about which business is larger.
Aswath Damodaran of NYU, assessing the SpaceX figure before that offering, described the AI TAM as reaching the end of what is plausible and pushing beyond. That was the judgment on the smaller of the two numbers.
What the prospectus will and will not have to defend
This distinction matters more than it might appear. A pitch to investors and a registration statement are different documents with different consequences.
Forward-looking statements in a prospectus carry liability. They are accompanied by risk-factor disclosure, and a company that publishes a projection it cannot support has a problem. A TAM is market context rather than a forecast, which is precisely why it can be this large.
So the number to watch for is not the US$30tn. It is whatever revenue projection appears in the filing itself, because that is the figure Anthropic will be held to. We have already noted that the filing is reported to name public opposition to AI as a risk factor, which is the same document being candid in the other direction.
What the number would have to assume
Building a US$30tn addressable market from the value of AI-addressable work requires several assumptions, none of them stated.
The first assumption is that work done by people becomes a billable service rather than software a customer builds once and runs on open weights. Where the value then lands is a separate question, since it could flow to the application on top, to the cloud underneath, or to the customer as savings that are never spent, and only one of those routes reaches a model provider. Underlying both is a bet on price holding, in a market where DeepSeek is raising to compete on cost and capable open-weight models are released monthly.
Each of those is arguable in Anthropic's favour. A TAM computed on the value of labour silently assumes the most favourable available answer to all three.
Why the framing is being used now
A US$2tn valuation needs a story that a discounted cash flow cannot tell, and a very large market is that story.
This is a familiar move rather than a new one. Databricks raised at a valuation supported by a comparable narrative, and its chief executive's claim that AGI had arrived and changed nothing was the counter-observation from inside the same industry. The technology being real is not in dispute. What is in dispute is how much of that value reaches a model provider as revenue.
The timing supports the reading. Anthropic is expected to publish prospectus documents shortly, with a debut possible as early as September or October, and a TAM in circulation before the filing sets the frame that the filing's more constrained numbers will then be read inside.
How to read it if you are considering the offering
Three questions cut through this, and none of them involves the trillion-dollar figure.
Start with the revenue run-rate and how fast it is growing. Then gross margin, which for a model provider comes down to inference cost against price. Then find the projection in the filing itself, which is the only number in the document that carries consequences if it proves wrong.
A TAM tells you that a market is not too small. It never tells you what share a company will take, what it will cost to take it, or what anyone else will charge. The answers to those questions are elsewhere in the document.
