Mr. Latte News


Anthropic's amended S-1 shows $895.4 billion in compute commitments running through 2045

This English page is a machine translation of the Korean original, so some phrasing may read awkwardly.

Anthropic's S-1 amendment filed on November 2 discloses approximately $895.4 billion in spending commitments and take-or-pay terms that bill the company whether or not it uses the capacity.

Read the original at U.S. SEC EDGAR ↗

We rarely get to see where an AI model company has tied up its money, or how much. A company preparing to go public, however, has to put those numbers in front of regulators. The amended S-1 that Anthropic filed with the U.S. Securities and Exchange Commission on November 2 shows spending commitments of approximately $895.4 billion as of September 30, 2026.

Almost all of that money is tied to compute and infrastructure. The company says it expects to spend approximately $447.5 billion of it before the end of 2030, with substantially all of the remainder falling between 2031 and 2045. Approximately $69.8 billion is due within the twelve months following September 30, 2026.

The filing also describes the shape of those commitments. The company explains that its long-term compute agreements include minimum purchase commitments and take-or-pay provisions, so it has to pay regardless of actual utilization. Some of its power and data center arrangements have initial terms of 20 years with annual escalators, and fuel cost increases pass through to the company without a cap.

The revenue underneath those commitments has grown quickly. Revenue was $17.0 billion for the three months ended September 30, 2026. According to the filing, that quarter's revenue grew 1,127% year over year, while compute and infrastructure expenses grew 382%. Customers spending $100,000 or more a year grew from more than 950 at the end of September 2025 to more than 11,900 a year later.

The question a developer asks while reading this document changes, too. Until now, the main criterion was which model is smarter. Once commitments stretching across decades are public, you also start looking at what cost structure the API you are wiring in actually sits on.

If capacity has to be paid for whether it is used or not, every idle unit is a straight loss. A company holding contracts like these therefore has an incentive to fill spare capacity, and that incentive can turn into pricing that developers feel directly, such as batch discounts or off-peak rates. This part is not a fact from the filing. It is my inference from the structure of the commitments.

The same structure reads the other way as well. Fixed spending that runs through 2045 stays in place even in a year when demand softens, and the filing does not say where the company would cover the cost in that case. The company itself describes these commitments as substantial fixed obligations and lists them as a risk factor.

So the job right now for anyone building a product on a single company's API is not to prepare to switch models. It is to write down your assumptions about price and capacity. If you note in one line under what conditions today's rates could rise or fall, and how far your product can hold up when they do, you will have a baseline to recalculate against when the next filing arrives.

Sources

  1. Anthropic, PBC Form S-1/A, U.S. Securities and Exchange Commission EDGAR, filed November 2, 2026. Spending commitments, quarterly revenue, customer counts and contract terms are in the Prospectus Summary and Risk Factors sections.
  2. Anthropic, PBC Form S-1, initial filing on October 26, 2026.

A question to think about

Can you write down, in one line, the conditions under which the API rates your product relies on would go up or down?

Get new posts by email

We email you the news we pick. You can unsubscribe anytime from the link in any email we send.

Get new posts by RSS

Add the feed address to your RSS reader to follow new posts.

Open feed ↗