What Anthropic's S-1 Actually Tells Us, Past the 13x Headline
The number everyone quoted was 13x. The number that matters is what it costs to grow that fast, and what happens when the model that got you here keeps getting cheaper to run.
The number everyone quoted from Anthropic's S-1 was the 13x, first-half revenue of 16.3 billion dollars against the prior year. The number that matters is what it costs to grow that fast, and what happens when the model that got you here keeps getting cheaper to run.
A filing like this is the first honest look at AI economics at scale, and the honest parts are uncomfortable. Revenue this concentrated on a few foundation-model providers means the buyers, the startups and enterprises building on top, are exposed to price and policy moves they do not control. When one vendor cuts a price in half, an entire layer of products built on its old price has to reforecast overnight.
The interesting tension is on the cost side. The same week as the filing, GPT-6.1 Terra cut the price of an agent step roughly in half. Falling unit economics are great for adoption and terrible for the story that this is a high-margin business forever. The market is being trained to expect intelligence to get cheaper, which compresses the margin that justifies the valuation.
What I take from the S-1 is not the size of the win but the shape of the risk. The companies that survive the next two years will be the ones whose revenue does not evaporate the moment a competitor opens a stronger model at a lower price. Read the filing for the customers, not the multiple.