Anthropic Is Selling 2028 by the Quarter

Anthropic’s prospective IPO valuation hinges on a forecast of $190 billion to $200 billion in revenue by 2028, Reuters reported, citing sources. That range gives the company a deadline as well as a valuation argument. It asks investors to pay now for customer spending that must materialize over the next two years.

The forecast looks like a line on a chart. Read from the buyer’s side, it is a stack of purchase orders. Enterprises must subscribe, expand usage and renew. Cloud capacity must be reserved. Data centers must receive electricity. Sales teams must close contracts, and support workers must keep customers from abandoning expensive deployments when a model fails inside an ordinary workflow.

Software revenue does not generate itself, especially at this scale. Anthropic would need a vast pool of paying demand, enough computing supply to serve it and pricing that leaves money after inference and service costs. It would also need customers to treat AI spending as a durable operating expense rather than an experimental budget that can vanish after a disappointing pilot.

Buyers may be asked to accept longer commitments, minimum spending levels or deeper integration with Anthropic’s products. Those arrangements can make revenue more predictable. They can also transfer risk to customers, who may pay for capacity before workers find reliable uses for it. A contract can prove that a company bought AI. It cannot prove that the purchase improved the company.

The labor beneath the forecast is easy to misplace. Engineers adapt systems to company data. Security teams approve access. Procurement staff compare vendors. Customer-support workers investigate failures. Managers decide whether the tool saved time or merely moved review work downstream. Anthropic books the sale; its customers often supply much of the implementation.

Investors therefore need more than a growing revenue total. Quarterly disclosures should show how much growth comes from new customers versus expanded contracts, how often customers renew, whether usage follows committed spending and how computing costs change per unit of revenue. Concentration matters too: a forecast supported by a few enormous buyers carries a different risk from one supported by broad, repeated use.

The useful test is not whether $200 billion sounds visionary or absurd. Watch what customers actually buy each quarter, what it costs Anthropic to serve them and how much human support is required to keep deployments useful. If renewals, usage and margins strengthen together, the forecast is becoming a business. If commitments rise while adoption and service economics remain obscure, 2028 is still doing sales work in the present.