A Nordic AI company with no public fundraising, facing Bolt.new and Vercel on both sides, reached an estimated $15 million in annual revenue without burning money or relying on SEO. What it got right was not technology, but a product decision most people overlook.
If you have opened Twitter in the past six months, you have probably seen the phrase “vibe coding.”
It describes a new programming pattern: you do not write code; you describe what you want, and AI builds it. The phrase originated with an Andrej Karpathy tweet, but the company that turned it into a category was a Swedish startup: Lovable.
You may not have heard of Lovable. It has not announced funding, has a team of 25, and is headquartered in Stockholm. But its numbers are striking: one million projects created, 20,000 new projects every day, and one million daily visits to apps built with Lovable.
All of those numbers come from Lovable’s website. I did not find third-party audited data. But because the product is public and projects can be browsed publicly, these claims are relatively easy to cross-check: visit Lovable and see how many public projects exist.
The question I want to unpack is not just growth speed. Any AI product can show nice numbers in a hype cycle. The deeper question is:
In a category surrounded by giants such as Cursor, Bolt.new, v0, and Replit, why did a 25-person team not only survive but move ahead?
The answer lies in one product decision.
A Counterintuitive Product Strategy: Support Only One Stack
Most AI coding tools try to support as many languages and frameworks as possible.
Cursor supports Python, Java, Go, Rust, C++, and more. Bolt.new supports React, Vue, Svelte, Angular, and others. v0 focuses on Next.js but still works with multiple UI patterns.
Lovable chose: React plus TypeScript plus Supabase, which uses PostgreSQL.
That looks like a deliberate narrowing of TAM. In AI products, it may be the smartest move.
The reason is simple: AI-generated code quality depends on model certainty. The more frameworks a tool supports, the more context the model must handle and the higher the error rate.
Lovable flips the problem around. Instead of forcing the model to adapt to every scenario and doing none of them well, it lets the model become excellent in one scenario.
The result: while Bolt.new sometimes generates Vue components with errors and Cursor requires repeated prompt tuning, Lovable users often report that the first generated code is mostly usable.
This is not a technology advantage. Lovable, Bolt.new, and Cursor use many of the same underlying models, including Claude and GPT. It is a product-strategy advantage.
Narrowing is not about avoiding competition. Narrowing is about building absolute certainty on the battlefield you choose.
The Second Key Decision: Credit-Based Pricing
Lovable’s pricing looks like this:
| Plan | Price | Usage |
|---|---|---|
| Free | $0 | 5 daily credits, roughly 30 per month |
| Pro | $25/month | 100 monthly credits plus 150 shared daily credits |
| Business | $50/month | 100 monthly credits |
| Enterprise | Custom | Bulk credits |
Two design choices matter.
First, team-shared pricing. The Pro plan is $25 per month without limiting team seats. A personal project can invite colleagues, then the whole team starts using Lovable, then the team upgrades to Business. There is almost no organizational-decision friction. Compared with Cursor’s per-seat model, Lovable’s pricing makes team adoption easier.
Second, credits align with AI cost. The largest cost in AI products is inference. If pricing is a fixed monthly fee, heavy users can erode margin. Credit-based pricing fixes the mismatch: the more users generate, the more they pay, and Lovable’s inference cost is covered.
Many AI products treat pricing as the commercialization team’s job. Lovable’s credit system is part of the product itself. Every generation consumes credits, which gives users usage awareness and gives Lovable a cost-recovery mechanism.
Pricing is part of the product. Good pricing is not making users feel everything is cheap. It is making heavy users pay more and light users pay less.
The Third Flywheel: Every App Is a Billboard
Lovable’s free plan has one critical limit: all projects must be public.
That is intentional.
When a free user creates and deploys an app with Lovable, that app becomes a living advertisement. Visitors may ask, “What was this built with?” and discover Lovable.
One million projects and one million daily visits mean Lovable receives one million free impressions every day.
This is pure PLG: the product itself is the acquisition channel.
The flywheel also accelerates itself:
- More users -> more apps -> more exposure -> more users.
- Better models -> better apps -> better word of mouth -> more users.
- More users -> more revenue -> more resources to improve product -> better product.
Lovable does not need to spend on SEO or paid ads. Its customer acquisition cost is close to zero.
If your product output can be seen or used by other people, design a mechanism that turns every output into your advertisement.
What Can Be Copied and What Cannot
When analyzing any AI product, keep asking: what can competitors copy tomorrow, and what can they not copy?
Copyable:
- AI code generation. Bolt.new, v0, and Replit all do it, using similar underlying models.
- Supabase integration. Any competitor can add it.
- Credit pricing. It can be copied directly.
- One-click deployment. Vercel and Netlify APIs are public.
Hard to copy:
- One million created projects and one million daily visits. The flywheel is already turning; later entrants cannot skip time.
- Category definition for “vibe coding.” Lovable is one of the earliest teams to productize “chat to full-stack app.” When people say “vibe coding,” many think of Lovable. Category mindshare is hard to replace.
- Engineering productivity from a 25-person team. Swedish engineering density plus a small-team minimalist culture cannot be bought with funding alone.
Lovable’s moat is not technology alone. It is category ownership, data flywheel, and community asset together.
The lesson for AI builders: when entering a market, do not only ask whether your AI is better than competitors’. Ask whether you can define a flywheel competitors cannot easily enter.
But I Need to Be Honest
All data in this article, including one million projects, 20,000 daily new projects, and one million daily visits, comes from Lovable’s website. I do not have access to its backend database and cannot independently verify the claims. Lovable has also not disclosed ARR. The $15 million to $17 million estimate comes from third-party organizations such as Sacra and may be wrong.
Lovable also faces risks:
- Model dependency: it relies heavily on Claude and GPT APIs. If model providers raise prices or restrict usage, the cost structure will be affected.
- Competitor ecosystems: Bolt.new has StackBlitz’s browser sandbox, v0 has Vercel’s Next.js ecosystem, and Replit has more than 5 million online IDE users. Whether Lovable’s independent path can continue remains open.
- Quality ceiling: current AI-generated apps are better suited to MVPs and prototypes. Whether they can meet production-grade reliability is still an open question.
Three Final Lines for Builders
Lovable taught me three things.
First, in AI products, doing less can be more valuable than doing more. Certainty in one stack beats breadth across ten stacks.
Second, pricing is not sales strategy. It is product strategy. Making users pay more when they use more matters more than making the product feel cheap.
Third, make your product output inherently distributable. Every Lovable app is a billboard. You do not need an SEO budget if users create things that others can see.
All data sources in this article are marked by confidence level. Company self-reported data is labeled as not independently audited. Third-party estimates are labeled as inferred estimates. The author has no economic interest in Lovable.
The research brief with full data sources, claim layers, and source list is available in the original source file: ai-product-scout-lovable.md.