Anthropic has filed to go public; the numbers point to operating profit, but the sources differ widely on the details.
Why this is useful to know
Anthropic, the company behind Claude, filed for a stock market listing in the US on Monday (an S-1). According to the summary by Latent Space, the document shows the first profitable quarter in Anthropic's history, with an operating profit of 974 million dollars in the second quarter. The net loss over the first half year is huge, but that comes mainly from accounting items around financial instruments that involve no cash going out, not from day-to-day costs. By 2028 Anthropic wants to deploy so-called 'virtual colleagues' (Virtual Collaborators) for more than a billion knowledge workers. For you as an SME owner, this is mainly a signal: the vendor is getting bigger and sturdier, and Claude is more and more often available through your existing cloud partners. Do read the figures critically, because the sources disagree on many numbers.
When you need this
This becomes relevant when you choose which AI vendor to use. You can use Claude directly, but also through the marketplaces of AWS (Bedrock), Google Cloud (Vertex) and Microsoft Azure. According to the source, 48 percent of revenue in the first half of 2026 ran through those channels, against 11 percent in 2023. If you already have a cloud contract, you may be able to buy AI within that contract, with one party for agreements and invoicing. That makes comparing harder, but also lowers the threshold. Hold off on large investments until prices and terms per channel are clear. Record which data goes to which vendor, also when it runs through a cloud partner.
How it works under the hood
The S-1 is mostly about finances, but three points matter for an application manager. First, distribution: according to the source, marketplace costs sit in sales and marketing, not in cost of revenue, so the headline gross margin of 85 percent is not affected by them. Work out the total cost per channel. Second, capacity: the document names compute as a lasting bottleneck and reports a postponed pre-training of a new frontier model. That can affect model versions and availability. Third, prices: a lower 'token price index' does not have to mean a price cut, it can mean a shift in model mix. So measure cost per task, not per token.
This news touches chapters 24 and 6 of the course AI(l) or nothing by Newway. If you follow the course, this subject returns there at a calmer pace.
What you can do with this
Want to know what this means for your business? Plan an introduction: half an hour, no sales pitch, but an honest answer to the question where to begin. Rather read on first? The course AI(l) or nothing is free and starts with the basics.
Written in response to a publication by Latent Space: https://latent.space/p/ainews-anthropic-s1/
