Scaling PostgreSQL to power 800 million ChatGPT users

Source: OpenAI Date: 2026-01-22 URL: https://openai.com/index/scaling-postgresql

Summary

OpenAI infrastructure engineering post from January 2026 describing how the platform scaled PostgreSQL to handle data storage and retrieval for 800 million ChatGPT users. The post covers the specific technical challenges of running relational database infrastructure at this user volume — sharding strategies, connection pooling, read replica architecture, and the specific PostgreSQL extensions and configurations that made it work. OpenAI publishing infrastructure deep-dives is relatively rare; this one signals engineering pride in an unglamorous but critical system.

Implications

800 million users as a scale reference point. The user count disclosed here — 800 million — is the most concrete public user metric OpenAI had released for ChatGPT. For context: WhatsApp at 800M users was considered a major social platform. ChatGPT at this scale in roughly 2-3 years of operation represents one of the fastest-growing consumer applications in history.

PostgreSQL at hyperscale as an architectural choice. The choice to scale PostgreSQL rather than moving to a distributed NoSQL database reflects OpenAI’s engineering pragmatism — PostgreSQL is well-understood, has excellent tooling, and works reliably when tuned correctly. The blog post implicitly argues that relational databases can scale to these levels without exotic infrastructure, which is useful for the engineering community.

Infrastructure transparency as talent recruitment. Detailed engineering blog posts about hard infrastructure problems attract the kind of database and platform engineers OpenAI needs. This is as much a talent marketing post as a technical contribution.

Watch: Whether OpenAI’s infrastructure stack continues to favor PostgreSQL as user count grows further, or whether eventual scale forces a migration to purpose-built distributed systems.