Cursor makes roughly $8.6M in revenue per employee. Almost no software company in history has matched that.
It also loses money on its individual developers, the exact people who made it famous. About 60% of its ~$3B revenue now comes from large enterprises instead.
So which Cursor is the $29.3B one?
And what happens to the other?
Model breakdown 001 Cursor snapshot
01. The Tech
Start with what Cursor actually is now. It’s an agentic IDE. With Cursor 3.0, the product went agent-first: you run fleets of agents in parallel, and the company says over 30% of its own code is already written by them.
To run all that, Cursor made a bet that most of its rivals haven't. It decided to own its models instead of renting them. The Composer models now handle most of the coding work; the frontier APIs, such as OpenAI, Claude, and Gemini, are the fallback.
There's one catch. Training frontier models takes enormous compute, and that's the one thing a 350-person company can't simply buy. Which is what the SpaceX deal is really about: access to xAI's Colossus cluster in Memphis, one of the largest in the world.
02. The Business Model
The pricing is simple enough: free, a $20 individual plan, $40-per-seat teams, and enterprise contracts.
The headline numbers hide a split. Enterprise accounts are gross-margin positive. Individual accounts are not, and the free tier is the obvious suspect. But it isn't the whole story.
The $20 plan loses money, too. A flat subscription competes with usage-based inference costs, so a heavy individual user incurs more in model calls than $20 covers. Cursor only crossed into gross-margin profitability in April 2026 because the enterprise mix and its own cheaper Composer models finally outweighed the individual drag.
The people who made Cursor famous are the ones it still can't make money on.
Cursor's problem isn't with the free users. It's that even the ones paying $20 a month cost more than they pay.
03. The Real MOAT
The enterprise business has a real moat. Workflow lock-in, used by roughly 64% of the Fortune 500 companies, and growing cross-IDE reach. Once a team adopts Cursor as their standard tool, they strongly stick with it and rarely abandon it impulsively.
The consumer business has almost none. Individuals switch tools over a weekend. Copilot, Claude Code, and Codex are one download away.
So being famous and being defensible (hard to beat) are different things. Cursor's goal is to turn its fame into a defensible business quickly, before its profits run out.
MY TAKE
Cursor looked at its one fatal flaw, which is renting intelligence from the companies trying to kill it and spent two billion dollars and a model team closing the gap before it mattered. The April margin flip says the plan is working.
The risk nobody is pricing is that Cursor is now two companies wearing one logo. A loss-making consumer brand and a profitable enterprise vendor. At $29.3B, you're paying for the enterprise business and subsidising the consumer one. If Composer ever falls behind the frontier models, both halves get expensive at the same time.
04. ONE THING TO WATCH
Whether Composer can match the frontier models on hard reasoning. If developers keep switching back to Claude and GPT for the difficult work, the margin gains leak straight back to the API bill.
That's the teardown.
Reply and tell me your feedback and which AI startup you want torn apart next. I read every one.
- Abhishek
P.S. If you found this useful, forward it to a founder who's tired of the hype.