Part one of OpenAI's engineering series on Habitat, the storage platform behind ChatGPT (1B+ weekly users, 500+ PB stored, 22M requests per second at the edge, 70M+ per second across all services, 100k+ requests per second during storage failure recovery). The writeup covers the migration from a shared Python library into a globally distributed service: asyncio tuning to bound scheduling delay under load; fixing an aiohttp LIFO-connection-pool metastable failure by moving to FIFO; a custom HTTP/2 fan-in front end on Envoy; an internal NoSQL layer with a TAO-style graph API; and Rockset-based CDC for online/offline migrations with a disaster-recovery escape hatch. The kicker: in Q2 2026, Habitat v3 — a complete rewrite in Rust — was built by 2 engineers in months using Codex and GPT-5.5, replacing a Python service second only to Kubernetes in core share, at ~6x CPU and ~15x memory efficiency, now serving ~95% of production traffic.
OpenAI's Habitat storage platform: 70M req/s lessons and a 2-engineer Rust rewrite
OpenAI described Habitat, the storage platform behind ChatGPT, as a 500 PB service spanning 40 regions. Part I covers its Python-to-Rust rewrite, HTTP/2 fan-in, connection-pool failures and change-data-capture migrations. Part II adds Cosmos DB co-managed multi-tenancy, hot-key throttling, placement and residency controls, selective replication and integrated caching. Scale and efficiency figures are vendor-reported.
Imported from the existing research archive. Migration preserves the recorded judgment and is not a fresh verification.
What this means for you
The architecture lessons (metastable failures in connection pools, asyncio tail latency, HTTP/2 fan-in economics) apply to any high-scale Python-to-Rust migration. But the headline data point is organizational: a 2-engineer, agent-assisted full rewrite of a top-2-by-cores OpenAI service at 6x/15x efficiency — currently the strongest public evidence for AI-assisted rewrites of business-critical infrastructure, and a benchmark for what 'AI-native engineering team' now means in practice.
Evidence & related material
OpenAI publishes Habitat multi-tenancy, caching and global placement details
Subsequent changes
OpenAI publishes Habitat multi-tenancy, caching and global placement details
Technical details & scores
| Scale | 1B+ weekly users; 500+ PB stored; 22M req/s at the edge; 70M+ req/s across services; 100k+ req/s during failure recovery; regional failover in minutes with multi-region replicas; recovery within seconds at the storage layer |
|---|---|
| Architecture | asyncio scheduling-delay tuning; aiohttp LIFO->FIFO connection pool (metastable failure fix); Envoy-based HTTP/2 fan-in front end; TAO-style graph API NoSQL layer; Rockset-based CDC; S3-less design with regional replicas |
| Habitat V3 | Q2 2026 full Rust rewrite by 2 engineers with Codex + GPT-5.5 in months; replaced the Python service second only to Kubernetes in core share; ~6x CPU and ~15x memory efficiency; ~95% of production traffic |
| Note | page og metadata shows an earlier publishedTime; dated here by the official RSS item (Fri, 11 Sep 2026 10:00 GMT) |
| Part Two | OpenAI says Habitat stores 500 PB across 40 regions and uses Cosmos DB co-managed multi-tenancy, hot-key throttling, selective replication and integrated caching. |
| Evidence Boundary | Architecture and performance figures are vendor-reported. |
| Time Basis | OpenAI RSS published 2026-10-23T17:00:00Z. |