Anthropic's Claude Code subscription may consume up to $5,000 in compute per month while charging the user just $200

An internal analysis by coding AI startup Cursor suggests Anthropic is burning through cash to subsidize its coding services.

According to Cursor's internal analysis reported by Forbes, Anthropic's $200-per-month Claude Code subscription "could use" up to $2,000 in compute last year. Today, that same plan is "able to consume" about $5,000, according to a separate person who has seen analyses of the company's spending patterns.

According to Forbes, Cursor subsidizes its own users too, though not nearly as aggressively. Consumer subscriptions run at negative margins, while business plans are profitable. The company offers a Teams plan aimed at startups and negotiates enterprise contracts with larger organizations.

When your model provider is also your competitor

The numbers spotlight a tough reality for companies building on top of a model provider that's also chasing the same customers. There's barely any margin left to capture. Worse, as AI models gain increasingly agentic capabilities, tools like Cursor's code editor could become redundant altogether.

To hedge against that risk, Cursor is betting on its own models. A team of about 20 AI researchers is developing the Composer model family, built on open-source foundations like DeepSeek, Kimi, and Qwen, then fine-tuned with proprietary data and reinforcement learning. Version 2.0 marked the release of Cursor's first homegrown coding model. Composer 1.5 is fast, the second-most popular model on the platform, and significantly cheaper for Cursor to run than paying for Anthropic's large models.

And despite the competitive pressure, Cursor is growing fast. Annual revenue shot from $100 million at the start of 2025 to over $2 billion, doubling in just the last three months, Forbes reports. The company is also pushing hard into enterprise sales, where contracts are slower to close but far stickier. Large customers now include Meta and Nvidia.

Even if Cursor's estimates are off—we don't know what Anthropic actually pays to run its own models—the broader point remains: once generative AI is deeply embedded in daily work, prices will go up. That goes for Claude Code just as much as ChatGPT, Copilot, and the rest of the pack.