Anthropic Prepares $80–100 Billion IPO at Valuation of Up to $1.75 Trillion

If the information provided by investors to the Financial Times is confirmed, Anthropic is preparing for one of the largest offerings in the history of the technology market. The developer of Claude wants to go public by Thanksgiving, raise $80–100 billion and secure a valuation of $1.25–1.75 trillion.

The astronomical sum is not the only notable point. The company has reportedly agreed to lower its expectations: the most optimistic investors had previously counted on $2 trillion. Amid a sell-off in technology stocks and rising borrowing costs, the market appears to be asking even AI leaders to demonstrate more grounded economics.

What exactly the market is testing

According to the report, Anthropic’s annualized revenue, calculated from its current sales run rate, has reached approximately $81 billion, up from $65 billion at the end of July. That is impressive growth, but it still does not answer the central question: how much money remains after paying for compute, data centers, model development and customer acquisition.

Consider a typical AI service for businesses. It sells a subscription for $100 but spends $45 on model calls, $20 on infrastructure and another portion of its revenue on sales and support. Revenue looks attractive in a presentation. The public market cares more about something else: whether the cost per customer can fall faster than sales grow. Anthropic will now face that same test, only on a scale of tens of billions of dollars.

An IPO will make visible the metrics that private companies disclose selectively: revenue composition, concentration among major customers, compute costs, obligations to cloud partners and actual losses. That is why the offering matters not only to Anthropic shareholders. It could become a benchmark for valuing the entire AI business chain—from model developers to startups building products on top of their APIs.

Why this matters to developers and products

If investors begin demanding more convincing margins, model providers will have to reconsider prices, limits and discounts. This is a direct risk for teams whose products depend on a single provider. A change in token pricing or enterprise contract terms can quickly turn a profitable feature into a loss-making one.

The practical conclusion is simple: start calculating the cost of one useful user action now, rather than looking only at the price per million tokens. Retain model-level cost data, use request routing and provide a backup supplier wherever switching is technically feasible. A cheaper model can handle routine requests, while an expensive one is invoked only for complex cases.

Founders should separately examine whether product growth is being concealed behind subsidized rates from a model provider. If your economics work only with the current discount, that is not an advantage but a dependency.

What remains unknown

The original report cites investors surveyed by the Financial Times but provides neither an official statement from Anthropic nor regulatory filings. The size, timing and valuation of the offering, as well as the revenue metric, could change. The $81 billion figure should also not be confused with confirmed revenue for a completed fiscal year: it is an annualization of the current sales run rate.

The direction, however, is already clear. The era in which enormous AI investments could be justified solely by promises of future growth is approaching a rigorous test. Developers must measure the cost of every use case, product teams must prepare pricing for cost fluctuations, and founders must prove that their businesses can survive more expensive capital and compute.