Bolt.new: a near-dead dev-tools startup that reached ~$40M ARR in five months

What it is

Bolt is a web-based AI coding agent from StackBlitz: you prompt an app into existence and it runs in the browser on StackBlitz's WebContainer technology (Lenny's Newsletter, Mar 2025). StackBlitz introduced it as "combining AI and WebContainers" to create Vite apps (StackBlitz blog).

Timeline

• Seven years before launch: StackBlitz builds WebContainer, "a browser-based operating system developed over seven years" (Lenny's Newsletter).
• Oct 2024: Bolt is announced in the StackBlitz keynote at ViteConf; by 22 October usage had "exceeded everyone's expectations" (StackBlitz blog).
• First two months: "we went from zero to 20 million of ARR", Simons says (podcast transcript).
• Mar 2025: nearly $40M ARR, 3 million registered users and 1 million monthly active users, with 15 to 20 people (Lenny's Newsletter).

Why it worked

• Years of deep tech met the right moment. Lenny's summary credits Claude 3.5 Sonnet as "the critical breakthrough that made AI-generated code production-ready" (Lenny's Newsletter).
• Low expectations, low burn. Before launch the hope was modest: "This would be amazing if this added a 100K of ARR over the next couple of months" (transcript). Keeping burn "extremely low" gave them time to keep trying (Lenny's Newsletter).
• An audience they didn't plan for. Simons said "67% of our users are not developers, at this point" (transcript).
• A simple launch. It launched with a post on Twitter (Lenny's Newsletter).

What to learn

1. Look at what you already built. StackBlitz's pivot used tech it had for years; your old project may fit a new model capability.
2. Survive long enough to catch timing. "Surviving long enough to catch market timing can transform a struggling startup" (Lenny's Newsletter).
3. Know your tool's limits. Simons: "if you have a large existing code base, you're going to need something like Cursor" (transcript).
4. If you build with Bolt, plan before prompting. Its docs recommend Plan mode and version history to save tokens (Bolt docs).

Risks and what didn't work

• Error loops cost money. "each attempt uses tokens. Avoid clicking Attempt fix over and over, hoping for things to eventually work out" (Bolt docs).
• Model dependence. By the founder's account, the product took off when a third-party model got good enough (Lenny's Newsletter); its quality and cost are not in StackBlitz's control.
• Self-reported numbers. ARR and user counts come from the founder on a podcast, not audited filings. The quotes above come from a community-published transcript of that episode.
• Crowded market. Lovable, v0, Replit and Cursor compete for the same users; Lovable's founder claimed to be "winning" in comparisons against bolt and others (Show HN).