The Idea: In 2022, Simon Eskildsen was a freelance "angel engineer" - a Dane who'd spent nearly a decade at Shopify scaling databases from hundreds to over a million requests per second. One of them was Readwise, the read-it-later app. They wanted to build article recommendations using vector embeddings. But the economics were absurd: vector search on their 100 million documents would cost over $20,000 a month - roughly quadruple what they paid for their entire relational database. So they shelved the feature. Simon couldn't let it go though. A useful feature, killed purely by infrastructure cost? That couldn't be a law of physics. The insight that eventually clicked was about storage: in-memory vector databases ran at $2+ per GB, but object storage like S3 ran at $0.02. For search, slightly slower writes didn't matter. The whole cost equation could be rewritten. This was the beginning of Turbopuffer.
The Execution:
- Mid-2023: What began as a blog post became a project, then a startup built out of co-founder Justine Li's living room in Ottawa. Simon locked in for four months, doing three complete rewrites to nail the architecture.
- October 4, 2023: Turbopuffer launched publicly, what Simon called the first truly serverless vector database. Priced at $1/month per million vectors, claiming it was 10–70x cheaper than alternatives. The launch post pulled nearly 400,000 views on X.
- November 2023: It clicked immediately for a scrappy ~8-person startup in Hayes Valley running one of the largest search workloads Simon had ever seen. He flew out, shipped the two features they needed, and days later they migrated production over - a 95% cost reduction, as promised. The company was Cursor, which today searches 1T+ chunks of code on Turbopuffer.
- January 2024: Turbopuffer raised a small round - under $1M total, from a single investor - to fund one year of R&D, with an explicit plan to shut down by year-end if they hadn't found product-market fit.
- Late 2024: With Cursor, Notion, Linear and Superhuman aboard, VCs lined up for a Series A. Simon turned them all down, running profitably by a personal framework of exactly six legitimate reasons to raise capital — "ego" being the one he warned against.
- December 2025: Turbopuffer finally took outside money - an undisclosed seed round from existing backer Lachy Groom (ex-Stripe) and new lead Thrive Capital (Joshua Kushner's firm) - atop a customer list that now included Anthropic and Atlassian.
- March 2026: Turbopuffer crossed a $100M run-rate - just 19 months after hitting $1M - still profitable, still having raised under $1M, on a team of 34. Customers now span Cursor, Anthropic, Notion, Cognition, Harvey, Bridgewater, Ramp, Linear, Legora, Superhuman, Atlassian, and Granola.
The lesson? Simon didn't set out to revolutionize databases. He obsessed over one customer's shelved feature until he realized hundreds of companies had the same wound. He bootstrapped to profitability, then used that profitability as leverage to say no to the money everyone else was desperate to take. While competitors raised mega-rounds to fund growth, Turbopuffer let a 95%-smaller bill do the selling - and went from $1M to a $100M run-rate in 19 months on under $1M raised. Profitability isn't the boring alternative to venture capital. It's the thing that lets you treat venture capital as optional.