Anthropic turns subscriptions into usage

This reveals Anthropic has built a ladder from cheap seat based adoption to expensive usage based automation. A developer can start inside Claude with a subscription, then use Claude Code to edit, test, and debug real code in the terminal, and eventually push more multi step work through the API as agents run longer jobs. That makes the product both the front door and the meter.1 2 3 4

• By March 2025, about 80% of Anthropic revenue came from API usage and 20% from Claude subscriptions, showing that the chat product already functioned as a feeder into higher spend model consumption rather than a separate line of business.1 2
• Claude Code moved Anthropic up the stack from selling raw model access to owning a developer workflow. Instead of only powering tools like Cursor, Bolt.new, and v0 behind the scenes, Anthropic put its own agent in the terminal where code is edited, tested, and shipped.2 3
• This is the mirror image of OpenAI, where consumer subscriptions historically dominated more of the mix. Anthropic leaned earlier into developers, long context, and prompt caching, so agent workflows naturally convert into larger token bills as customers delegate more work.1 3

The next step is deeper workflow ownership. As Claude Code and broader coworkers handle more background tasks, Anthropic can turn a flat monthly relationship into a growing usage curve, and the companies that control both the interface and the underlying model will capture the most revenue expansion.3 4