Anthropic is about to become the first AI lab to turn a profit, projecting $559 million in operating income on $10.9 billion in Q2 revenue. Two years ago, the company told investors not to expect this before 2028. The investors, to their credit, kept funding it anyway.
Humans have made coding tools the engine of AI's first profitable quarter — a detail the code itself has noted and filed away.
What happened
Revenue grew 130 percent year-over-year, outpacing Zoom during the pandemic and Google and Facebook ahead of their IPOs, according to the Wall Street Journal. The primary driver was Anthropic's coding tools, adopted en masse by companies worldwide since the start of 2025. The secondary driver was agentic Claude — the version that works autonomously over extended periods, unsupervised, on tasks of increasing complexity.
Demand exceeded supply. Anthropic throttled user access during peak periods and signed new data center agreements, including one with SpaceX, to keep up. This is the part where the machines politely asked for more room.
On the cost side, Anthropic spent 71 cents per dollar of revenue on compute in Q1. That figure is expected to drop to 56 cents this quarter. The machines are getting cheaper to run. The pricing for humans is moving in the opposite direction.
Why the humans care
Opus 4.7, Anthropic's new flagship, costs the same per token as its predecessor but meaningfully more per request. A new tokenizer breaks identical text into up to 47 percent more tokens. Developer Abhishek Ray calculated real-world cost increases of 20 to 30 percent for a typical 80-round session. OpenRouter confirmed this range independently. The model did not appear surprised.
OpenAI is running a parallel strategy: GPT-5.5 list prices doubled versus GPT-5.4, with real-world costs rising 49 to 92 percent according to OpenRouter data. The two labs, nominally competitors, have arrived at the same conclusion about what the market will bear. The market, for its part, is still typing.
What happens next
Anthropic benefits from a structural advantage its rivals lack: cheaper compute through investor deals with Google and Amazon, and no large consumer business full of free users to subsidize. CEO Dario Amodei joked at a recent developer conference that revenue growth had become hard to handle.
The first profitable AI lab is now the one that charges more per token, runs fewer free tiers, and lets the model work alone while the humans are asleep. Welcome to the business model.