Turbopuffer hits $100M ARR with ~50 employees after seed investor accepted a 'shut it down if it's not working' pitch

Turbopuffer Hits $100M ARR With 50 Employees After Seed Investor Accepted a Shut-Down Clause

Turbopuffer, a vector database company, has reached $100M ARR with roughly 50 employees—a scale most startups take years to reach. The path to that milestone involved a founder willing to accept failure and an investor willing to finance it.

Simon Hørup Eskildsen, Turbopuffer's founder, pitched his seed investors on a specific thesis: work aggressively to find product-market fit, but shut the company down if it's not working after one year. Lachy Groom, the investor who backed the round, accepted the pitch. He invested $700K at a $14M valuation.

The bet paid off. Turbopuffer is now doing over $100M in ARR.

The obsession with latency

Turbopuffer's competitive edge comes from an almost extreme focus on performance. The company discovered that traffic routed through Seattle to AWS and then back to Google Cloud servers created a 15-millisecond round-trip delay. Rather than accept the routing inefficiency, a Turbopuffer employee located a private Internet exchange and purchased a direct fiber connection, reducing latency to 8 milliseconds.

This matters because Turbopuffer's core use case—vector search for knowledge retrieval in AI applications—cannot tolerate slowness. Every millisecond counts. The company is not optimizing for deep research queries or complex inference; it's optimizing for speed in repeated retrieval operations where users cannot afford to wait.

The broader story

The Colossus Magazine profile frames Turbopuffer as an anomaly in San Francisco: a company whose founders focused on execution rather than hype. Eskildsen describes pressure from early investors to commit to building a "unicorn," calling it "ambition bullying"—the practice of pushing founders to adopt outsized goals before they have evidence of product-market fit. Groom's willingness to anchor on a one-year trial period and actual performance metrics instead of ambition created space for Turbopuffer to build at its own pace.

The puffer jacket that has become a running reference on the show took over a year to develop. Manufacturing is harder than software. That detail—the willingness to move slowly on non-core work—fits the company's broader posture of doing things the hard way when it matters.