Cursor (Anysphere): the AI-native code editor that went from launch to $2B ARR in three years

The 2023 launch and product bet

Anysphere was a small team that graduated from OpenAI's startup accelerator in 2023. Rather than build a new IDE from scratch, the team forked Visual Studio Code — the dominant open-source editor — and rebuilt the AI assistance layer around modern frontier models. The structural bet was that AI capabilities had reached a point where the productive use case was not autocomplete (the GitHub Copilot model) but full-file and full-codebase reasoning: refactor this function, explain this code, write this test, debug this stack trace.

The product launched in March 2023. By the end of 2023 it had approximately 30,000 daily active users and an $8 million seed round led by the OpenAI Startup Fund. The growth was driven almost entirely by developer word-of-mouth and the underlying capability of the integrated frontier models (initially OpenAI; later expanded to include Anthropic Claude and Anthropic-specific model integrations).

The ARR build

Cursor crossed $100 million in ARR in January 2025, approximately 20 months after launch — the fastest path to $100M ARR ever recorded in enterprise SaaS, reportedly without any marketing spend. By June 2025 ARR had reached $500 million. The $1 billion ARR milestone was crossed by late 2025 alongside the Series D announcement. By April 2026, Anysphere had surpassed $2 billion in annualised revenue.

The driver was a combination of two factors. First, individual developers adopted Cursor through pro subscriptions ($20-200 per month per developer). Second, enterprise procurement caught up: by 2025 large engineering organisations were standardising on Cursor at the team and company level as the AI-coding tool of choice, paying enterprise seat licenses. Net retention rates were reported above 150 percent as customers expanded usage over time.

The competitive and dependency questions

Cursor's position has structural risks. First, the underlying frontier models are built by Anthropic and OpenAI — Cursor pays both for API inference and is exposed to pricing and capability decisions those labs make. If a model provider competes more directly in the IDE layer, Cursor faces a buyer-as-supplier problem. Second, GitHub Copilot (the incumbent) has been responding with deeper AI integration including its own Copilot Workspace and Copilot Edit features. Third, AI-native competitors (Replit, Cognition's Devin agent, Codeium/Windsurf, others) have been building features that target the same workflows.

Fortune coverage in early 2026 highlighted these concerns. Anysphere's strategic response has included building proprietary models for specific coding tasks (reducing third-party model dependence), and expanding enterprise sales motion. The competitive question will be resolved over the next several quarters as enterprise procurement cycles complete and the underlying frontier-model landscape continues to evolve.

How RGM thinks about AI-tool category creation

When clients ask about AI-tool category creation, the Cursor case is the defining recent reference. Three structural lessons. First, the product bet was specific: not autocomplete but full-codebase reasoning. The Copilot incumbent was occupying the autocomplete category; Cursor created a new category by reframing what AI in the IDE was for. Second, the product's viral developer-led adoption is what built ARR ahead of any meaningful marketing spend — the developer audience trusts other developers' recommendations and the product delivered enough productivity gain to be worth talking about. Third, the dependency on frontier model providers is the structural risk; Cursor's ability to expand into proprietary models for specific tasks may be the most important strategic move of the next 18 months.

The pattern is hard to copy in categories without a comparable developer-led adoption motion. AI tools targeting non-developer audiences (designers, lawyers, finance professionals) generally cannot rely on viral end-user adoption to drive enterprise procurement at the speed Cursor achieved. We tell clients targeting non-developer AI workflows to expect substantially slower ARR ramp and to plan their marketing and sales motion accordingly.

Frequently asked questions

Related