Anthropic books $17 billion in Q3 and flags self-improving AI

The quarter-on-quarter pace has slowed. Sales were about 47 per cent above the second quarter’s, which had itself risen 144 per cent, after 132 per cent in the first. The quarter’s net loss was $23.29 billion. Over nine months, revenue reached $33.24 billion and the loss $104 billion, most of it an accounting charge on convertible notes and derivatives, CNBC says. Purchase commitments and leases total about $895.4 billion, substantially all of it for compute and infrastructure.

The filing also carries a safety warning, TechCrunch reports. The risk it names is AI speeding up the building of AI: Anthropic already uses AI in its own research, has only noticed past speed-ups once they had happened, and now has internal estimates pointing to it soon meeting its policy’s definition of “dramatically accelerated progress attributable to AI”. For now, the company adds, none of its models is designated as past that line, and it is weighing policy changes so that people stay in charge of how its frontier models are built.

Among the products Anthropic sells are the subscriptions we track: Claude Pro at $20 a month, Max at $100 and Ultra at $200 on our 2 November check.

CNBC’s account of the filing does not split the quarter’s $17 billion by product, so it gives no way to tell how much came from the $20 to $200 plans on our sheet.