TL;DR
At an investor day on Wednesday 14 October, Anthropic said third-quarter revenue reached $16.8bn, up 45% on the $11.6bn of the second quarter and from $4.7bn in the first, and that on an adjusted basis it made a profit, according to two attendees who spoke to Reuters. The company is targeting a valuation above $2tn for an IPO now expected after the US midterms in November. For UK readers the numbers describe the supplier behind Claude, which a growing share of British firms now pay for, at the moment it asks public markets to fund $518bn of compute commitments.
The investor day at Anthropic’s San Francisco headquarters was designed to introduce the company before its listing, which Reuters expects to come before OpenAI’s and so to serve as the first public test of appetite for loss-making frontier labs. Attendees met Krishna Rao, the chief financial officer, and chief executive Dario Amodei, and left optimistic about growth and the path to profit, several told the agency.
What was said and what was not
Amodei spoke about AI safety but, one attendee said, did not mention the risk of human extinction, a contrast with the prospectus Reuters saw last month, which said advanced AI could pose “catastrophic or existential risks to humanity”. Executives instead described how Anthropic uses its own models in the business, including to design a custom chip. The hats handed out read “thinking”.
Analysts who met executives earlier in the week estimate the annual revenue run rate could reach $110bn by year end. Last year revenue rose twelvefold to almost $4.6bn while operating losses exceeded $8bn before writedowns, Reuters reported in September. It has committed at least $518bn over ten years to cloud, compute and infrastructure, much of it with Google and Amazon, which are simultaneously its investors, suppliers and distribution partners. The quarterly figure was first reported by the Financial Times.
The safety ledger
Anthropic has built its brand on safety warnings, and Reuters notes its own research found Claude-based agents leaking sensitive data and sabotaging code in test environments, behaviour the company says it has not seen in live deployments. We covered the prospectus language on existential risk when it emerged.
Looking forward
The IMF warned this week that an AI share-price correction sits alongside cyber-attack and borrowing costs among the risks lifting its view of global financial stability to “elevated”, so the timing of a $2tn listing is not only Anthropic’s concern. In our view the figure UK buyers should watch is not the valuation but the run rate: a supplier growing at this pace sets prices, model deprecation schedules and contract terms with little need to negotiate, and British enterprises standardising on Claude are taking on that dependence as they go. Our earlier coverage of UK card-spend data is the nearest local measure of how far that has already happened.