On September 11, 2026, OpenAI published "Rapidly scaling online storage to serve over 1 billion ChatGPT users", the opening episode of a series about its storage infrastructure. This is not a model launch, not a flashy demo: it is the engineering story of the layer that holds everything else up. And for anyone who has ever sized a distributed system, the published figures are worth the read.
What was announced
According to OpenAI's official post, Habitat — the company's internal storage platform — evolved from a simple Python library into a globally distributed storage platform. The system now serves more than one billion ChatGPT users, per the same publication.
The headline number: 22 million requests per second, according to the article. That is the throughput the storage layer handles in normal operation to feed conversations, uploaded files and session state.
The post is explicitly framed as "part one": OpenAI promises a series detailing the architecture. This first instalment sets the context — Habitat's trajectory from internal script to planet-scale system — without yet unrolling the full technical diagram.
Three concrete advantages
- A codebase that grew without a ground-up rewrite. Per OpenAI, Habitat started as a Python library; for platform teams, it is proof that a pragmatic prototype can become a critical production system when evolution is well managed.
- A public throughput reference: 22 million requests per second. This figure, cited in the article, gives builders a real order of magnitude to calibrate their own high-concurrency storage architectures.
- An announced technical series, not a marketing post. The "part one" format likely signals architecture diagrams and hard-won lessons to come — the raw material engineers love to dissect.
Three opportunities it opens
- A trajectory model for homegrown storage stacks. Teams hesitating between an off-the-shelf solution and internal development now likely have a documented scale-up case, from script to a billion users.
- Transferable lessons for agent backends. Agents chaining tool calls generate intensive read/write patterns; OpenAI's choices for Habitat will likely inform the design of state layers for these workloads.
- A signal for infrastructure careers. Publishing this series suggests OpenAI treats storage engineering as a strategic differentiator — useful input for anyone building a platform team or a career in the field.
What remains to be verified
The post does not yet disclose implementation details: underlying storage engine, replication strategy, or cross-region consistency handling. Those elements will likely arrive in later instalments. No cost or latency figures are published at this stage, and no date is given for the next episodes.
Which layer of your own stack would deserve this level of transparency?
If you follow the latest AI technology, I publish a deep dive every day on frontier models, hardware, robotics, automations and AI-generated music. 👉 Get the next one straight in your inbox — sign-up takes ten seconds.
Sources
- Rapidly scaling online storage to serve over 1 billion ChatGPT users (OpenAI Google News)