A thirty-trillion-dollar market still needs a billable unit
Anthropic’s reported market estimate raises the useful enterprise question: which unit of work becomes revenue, at what price and with whose costs?
Fortune reported on August 26 that Anthropic was preparing a total-addressable-market estimate above $30 trillion. That is a market estimate.
I would not put it in a valuation-multiple calculation. A market estimate, a company’s annual revenue and the price of its equity are different quantities. Dividing whichever two produce the loudest number does not make the result economically useful.
My first question is smaller: what is the billable unit inside the proposed market?
That question matters well beyond an IPO discussion. An enterprise team can make the same mistake when it treats the cost of every potentially affected employee as revenue available to an AI product. There are several commercial decisions between work that could change and money a particular supplier can collect.
Start with a unit someone can buy
Consider a hypothetical document-processing service. The addressable work might include receiving a document, checking it, resolving exceptions and updating a business system. Counting the full cost of everyone involved gives one view of the economic activity.
It does not establish the supplier’s price. The buyer may retain the exception team, require human approval and pay separately for system integration. Some saved time may become spare capacity rather than an immediate reduction in expense. The service can still be valuable; its revenue model needs to specify how that value becomes a purchase.
I would define an accepted processing outcome and ask who pays for it. Then I would write down the price, the expected number of purchased outcomes and the costs still borne by the customer. That turns a broad opportunity into an inspectable business model.
A billable unit might instead be a seat or a consumption allowance. That is fine. It simply changes the arithmetic. A seat-based business needs assumptions about buyers and retention. A usage business needs assumptions about volume, unit price and how efficiency affects the bill.
Make the capture assumptions visible
For illustration, suppose a workflow contains $10 million of annual activity. Only half is technically suitable for the proposed automation, and half of that suitable work is commercially reachable in the planning period. That leaves $2.5 million of relevant activity under these invented assumptions.
If the supplier captures 10% of that activity’s value as revenue, the result is $250,000. Those percentages are not estimates for Anthropic or any real company. They show why naming the starting pool is only the first line of the calculation.
I would stress-test each assumption separately. Technical suitability can fall when exceptions are counted. Reach can shrink because the buyer cannot change a critical workflow this year. Price can move because competitors offer substitutes. Delivery cost can rise because the promised outcome includes support the model cannot provide alone.
The useful plan tells the team which assumption to test next. A market total that absorbs every uncertainty into its largest number cannot do that job.
This is not an argument that large markets are imaginary. It is an argument for using them at the right level. They can describe ambition and opportunity. A product investment still needs a path from a specific buyer’s problem to a repeatable purchase with workable economics.
Before arguing about the size of AI’s market, name the unit the customer will buy and the assumptions that turn it into revenue.
