Building Turbopuffer: Engineering Databases for the AI Era

The Engineering Mindset: From Shopify to First Principles

Simon Eskildsen’s career is defined by a deep-seated insecurity regarding his lack of formal computer science education, which drove him to obsessively deconstruct systems. During his eight-year tenure at Shopify, he scaled infrastructure from 1,000 to 1 million requests per second. His approach to engineering was rooted in "napkin math"—a methodology of calculating theoretical performance limits based on hardware specs (DRAM bandwidth, SSD latency, S3 throughput) rather than relying on misleading benchmarks. This practice allowed him to identify when a system was underperforming due to poor implementation rather than hardware limitations.

The Genesis of Turbopuffer

Turbopuffer emerged from the realization that vector search infrastructure was unnecessarily expensive and complex. Eskildsen and his co-founder, Justine, applied their experience with high-scale infrastructure to build a database that treats S3 as a primary storage layer. By optimizing how data is batched and persisted, they achieved a cost structure where a million vectors cost roughly one dollar—a massive reduction compared to industry standards at the time. The project was validated early by Cursor, which became their first major customer, proving that the architecture could handle real-world AI workloads.

Navigating the Cloud and Venture Capital

Eskildsen offers a pragmatic, often contrarian view on the cloud ecosystem. He notes that while GPUs are currently the focus of massive investment, the fundamental bottleneck for many AI applications remains CPU and storage efficiency. He recounts a meeting with Jensen Huang, who challenged his initial business model, forcing a pivot toward more sustainable infrastructure economics. Eskildsen also maintains a skeptical stance on venture capital, arguing that founders should only raise when necessary to avoid unnecessary dilution of employees and to maintain focus on building a sustainable business rather than chasing vanity metrics.

Culture and Remote Operations

Building a company in the current era, Eskildsen emphasizes the "campfire" concept for remote-first teams. This involves creating digital spaces that mimic the casual, high-bandwidth communication of an in-person office. He prioritizes hiring engineers who share his trait of "peeling back the layers"—those who cannot help but investigate how a reverse proxy works or why a specific database driver fails under load.