Anthropic told investors during a pre-IPO investor day on Wednesday that its quarterly revenue rose by nearly 50%, reaching $16.8 billion in the third quarter. Details of the presentation were reported by the Financial Times and attributed to people with knowledge of the matter.
Quarterly breakdown
- Q1 (January–February–March): about $4.8 billion in revenue.
- Q2 (April–May–June): $11.5 billion in revenue.
- Q3 (July–August–September): $16.8 billion in revenue.
The company’s disclosed figures show a sharp acceleration in revenue between Q2 and Q3, representing close to a 50% quarter-on-quarter increase.
Operating metrics and cost efficiency
Anthropic said its adjusted operating margin (excluding stock-based compensation) was 11%. The company also reported that it generated $1.70 in revenue for every $1 spent on computing power.
Distillation: how rival usage affected revenue
The startup said it had significantly clamped down on “distillation” — a controversial practice in which rival AI labs train their models on the outputs of more advanced systems. According to the presentation, distillation accounted for 5–7% of revenue, and the company materially reduced that activity in the third quarter, people with direct knowledge of the matter told reporters.
Using the Q2 figure as an example, if distillation represented 5% of Q2 revenue, then 5% of $11.5 billion equals about $575 million, implying rival labs may have spent hundreds of millions of dollars distilling Anthropic’s models in that quarter.
Why this matters
The disclosures highlight both rapid top-line growth and shifts in the composition of Anthropic’s revenue. Reducing distillation could lower external usage-related revenue but increase control over model outputs and usage patterns. The 11% adjusted margin and the $1.70 revenue-per-dollar-of-compute metric provide indicators of the company’s cost efficiency and the economics of its model deployment, according to the information presented to investors.
The details above are drawn from the investor presentation reported by the Financial Times and attributed to multiple people familiar with the matter.
