Why database access became a platform service
OpenAI published the first engineering account of Habitat on September 11. Habitat is the online storage platform behind ChatGPT, the API and Codex. The company says it processes more than 70 million requests per second, serves over 500 petabytes and supports products used by more than one billion people each week across almost 40 regions. It began in 2023 as a Python library in front of Azure Cosmos DB, but coordinating compatible client releases became brittle as services and regions multiplied.
The team moved storage access into a central service so routing, authorization, encryption, request shaping, connection pooling and audit controls could evolve in one place. Habitat deliberately exposes a constrained object-and-edge NoSQL API instead of arbitrary SQL and unbounded traversal. Complex analysis is isolated through change-data capture into secondary systems. Less query flexibility makes the cost and failure radius of online requests easier to predict.
The migration discipline matters more than the AI rewrite headline
OpenAI says two engineers used Codex and GPT‑5.5 in the second quarter of 2026 to rewrite the service in Rust. The new implementation now handles 95% of production requests and, in the company's measurements, is six times as CPU-efficient and 15 times as memory-efficient as the Python service. Those are vendor measurements for one internal workload, not a promised result for automated rewrites elsewhere.
The transferable lesson is sequencing: constrain the API, build observability and shadow traffic, then migrate gradually. A large rewrite still requires production evidence for behavioral parity, consistency, tail latency, fault isolation and rollback. AI assistance does not remove those controls.