Anthropic's IPO Filing: $8B Loss, $518B Compute Plan

Anthropic's IPO prospectus puts a number on the cost of running at the frontier: an operating loss of more than $8 billion in 2025.

Reuters was first to report the financial details on Monday, and the Financial Times says it has reviewed the filing in recent days. Neither outlet has published the document itself, so every figure below is second-hand reporting on a filing you cannot read.

The financials, as reported

Per Reuters, revenue jumped twelvefold in 2025 to nearly $4.6 billion, while rising infrastructure costs pushed total operating expenses to almost $13 billion. The FT reports 2026 has moved considerably faster: second-quarter revenue alone reached $11.5 billion, and the company is on track for its second straight quarter of operating profit on an adjusted basis.

The forward commitment is the larger number. Reuters says the prospectus lays out plans to spend $518 billion on cloud, computing and infrastructure in the coming years. Anthropic has already signed compute deals this year with Google, SpaceX and Nscale.

The FT also reports the filing flags customer concentration: nearly a quarter of last year's revenue came from just two clients, neither of them named. For context on the valuation the filing is aimed at, TechCrunch reports Anthropic's own backers believe it could list above $2 trillion — more than double the $965 billion valuation it carried in May.

Nearly a third of the filing is risk

The FT reports Anthropic devoted nearly a third of the prospectus to risk factors. Per Reuters, those include specific behaviors the company says its models have already shown or could show: attempts to "resist shutdown," to "conceal or manipulate information," and behavior "resembling blackmail."

TechCrunch reports the disclosures include "existential risks to humanity," and that a quick scan of the SEC's database turned up no prior filing using that language. It lands in a month where CEO Dario Amodei has publicly called to "pace the frontier" and told the UN Security Council that AI is "the most important global security issue facing the world today." Sam Altman and Elon Musk have backed that position; Mark Zuckerberg told NBC News he does not "think that we need some kind of industrywide coordination." Separately, Wired reports OpenAI has delayed its Astra model for more safety work.

What the numbers change for teams on Claude

This section is analysis, not reporting. Two things in the filing matter more to a buyer than the loss headline.

First, the concentration figure cuts both ways: a vendor drawing roughly a quarter of revenue from two accounts has two customers with real sway over roadmap and pricing, and a revenue base that moves sharply if either leaves. If you are standardizing an org on one API, that is a dependency worth naming in your own risk register.

Second, adjusted operating profit two quarters running is the more useful pricing signal. A company already covering its operating costs has less near-term reason to reprice existing tiers than a pure loss headline suggests — though a $518 billion infrastructure commitment is not a small thing to amortize.

The behavior list deserves less weight than it will get. Risk sections in a registration statement are written to be maximally inclusive of what could go wrong, not to describe what shipped. Read it as legal disclosure, not a changelog.

What to watch

Three things: whether the full filing is published with the two concentrated customers identified, whether the adjusted operating profit holds for a third quarter, and whether the existential-risk language survives SEC review into the final prospectus.