Cursor just announced 1 million daily active users and fifty thousand businesses, including Stripe, Figma, and Shopify. It has a fresh 900M dollar valuation and is building its own model. Meanwhile, I'm sitting in a terminal running claude with no syntax highlighting, no sidebar, no cute inline diff preview. Just text in, text out, code changed. I've never shipped faster in my life.
Cursor is a good product, probably the best AI-native editor today. Tab completion understands your codebase, inline edits respect conventions, and Composer handles multi-file refactors. If you're coming from vanilla VS Code with Copilot, Cursor feels like going from bicycle to motorcycle. A million users didn't happen from marketing. Credit where it's due.
But 1M DAU sounds massive until you remember there are 28 million developers worldwide. That's 3.5%. Copilot still has more users. VS Code still has more. Vim still has more (yes, counting the ones who can't quit). Cursor is winning among developers who want AI deeply integrated into their editor and will pay 20 bucks a month for it. That is a real market, but a narrow one to call a war won.
The battlefield: Cursor: 1M DAU, best inline experience, building own model. GitHub Copilot: largest user base, Microsoft plus OpenAI backing. Windsurf/Codeium: the "like Cursor but different" play. Zed: blazing fast, Rust-based, devs insufferable about it. Claude Code: CLI, agentic, no editor, just vibes and stdout. Aider: open source, terminal, doing this before it was cool. Codex CLI: OpenAI's terminal entry. With seven contenders still landing on the beach, this is a long way from over.
Cursor building its own foundation model is either the most ambitious or most delusional move in dev tools history. Building a competitive foundation model costs hundreds of millions. OpenAI spent 10 billion plus. Anthropic spent billions. Cursor raised 900M, a rounding error in the foundation model game. Like showing up to F1 with a go-kart with FAST spray-painted on it. But they have something the labs don't: a million developers generating edit sessions daily. A data flywheel that's hard to replicate.
I chose a terminal for the enterprise software I build: Clean Architecture, 252 use cases. I need to talk through an entire architectural layer and refactor across 22 files without breaking interface contracts. Autocompleting a for loop does little for that job. The dirty secret of AI editors: most flashy UX is optimized for the wrong unit of work. Tab completion equals lines. Inline edits equal functions. Composer equals files. I optimize for features. A feature touches 8 files across 4 layers. The terminal's mental model is "you're running a command" and the command can be anything.
The uncomfortable economics: Cursor charges 20 bucks a month. Their cost is API calls to Anthropic and OpenAI. Every power user is a margin risk. That's why they're building their own model, to stop paying Anthropic. But if their model is worse, power users leave. If it's just as good, building it costs more than licensing. Meanwhile Claude Code charges based on actual token usage, without that subscription arbitrage.
I expect the models to become commodities and agents to beat autocomplete. The 1M token context windows will make a big difference in what those agents can handle, and the terminal will have its moment. Cursor will survive, I think, without monopolizing the market.
The competition will change when models understand architecture and business rules well enough to implement features across entire codebases. The request becomes "build this module" instead of "help me write this line." We are in the Model T era, and Cursor has the best paint job on the lot. I am waiting to see how it handles the next road.