BestHub
Sign in
AI Engineering
AI Engineering
Jan 24, 2026 · Databases

Can One PostgreSQL Instance Power 800 Million Users? OpenAI’s Trade‑offs Unpacked

OpenAI runs a single primary PostgreSQL with about 50 replicas to serve roughly 800 million ChatGPT users, employing PgBouncer connection pooling, cache‑lock mechanisms, query and schema optimizations, workload isolation, multi‑layer throttling, and cascade replication, while acknowledging the architecture’s limits and the complexity of full sharding.

Azure CosmosDBCascade ReplicationOpenAI
0 likes · 8 min read
Can One PostgreSQL Instance Power 800 Million Users? OpenAI’s Trade‑offs Unpacked
Radish, Keep Going!
Radish, Keep Going!
Jan 23, 2026 · Databases

How OpenAI Really Scaled PostgreSQL for Hundreds of Millions of Users

The article debunks OpenAI's sensational claim of handling 800 million ChatGPT users with a single PostgreSQL instance, revealing a pragmatic hybrid architecture that combines many read replicas, Azure CosmosDB for write‑heavy workloads, and top‑tier hardware, while highlighting cost and complexity considerations.

Azure CosmosDBDatabase ArchitecturePostgreSQL
0 likes · 6 min read
How OpenAI Really Scaled PostgreSQL for Hundreds of Millions of Users