Rapidly scaling online storage to serve over 1 billion ChatGPT users
OpenAI has transitioned its internal Habitat storage system from a Python library into a globally distributed storage platform. This architecture now supports 1 billion ChatGPT users and handles a throughput of 22 million requests per second.
Verified State Diff
Impact & Verification Analysis
OpenAI infrastructure engineers and high-scale platform architects.
It demonstrates the infrastructure maturity required to support massive user bases and provides a blueprint for handling extreme-scale, low-latency data access in AI-driven applications.
Full Fact Overview
The transition of Habitat from a localized Python library to a globally distributed storage platform represents a significant architectural shift in OpenAI's infrastructure stack. By moving away from a library-based model to a distributed service, OpenAI has optimized for high-concurrency, low-latency data retrieval required to maintain state for 1 billion users. The reported 22 million requests per second indicate a massive scale-out strategy, likely involving sharding, geo-replication, and a custom storage engine designed to handle the specific I/O patterns of large-scale LLM interactions.