OpenAI explains how it scaled Habitat, its storage platform, on Azure Cosmos DB

In the second part of an engineering series dated October 23, 2026, OpenAI describes Habitat, the online storage platform behind ChatGPT, the API and Codex. OpenAI says Habitat serves over 1 billion users across 40 regions and holds more than 500 petabytes of data, and that its summary of the post claims 12× faster reads for ChatGPT.

OpenAI describes three phases on Azure Cosmos DB: relying on the database's own limits, which led to noisy neighbors and shrinking throughput as partitions split; giving each new use case its own container, which made capacity planning a burden for every product team; and returning to one shared multi-tenant container with hot-key throttling, request-priority tiers and a new serverless mode that Azure built for it. OpenAI says it has had no provisioning or noisy-neighbor incidents in the months since.

OpenAI also says its largest Cosmos DB account stores tens of petabytes and serves more than 20M queries per second, and recounts two outages in late 2024 (a regional power failure and a connection stampede during failback) that led to joint testing with the Azure team. The post also covers global caching and a Rust proxy to cut overhead. All figures are OpenAI's own.