Back to list
Industry NewsOpenAIChatGPTCloud Infrastructure

Scaling Online Storage for 1 Billion Users: How OpenAI Evolved Habitat to Handle 22M Requests per Second

OpenAI has shared insights into how it rapidly scaled its online storage architecture to support over 1 billion ChatGPT users worldwide. At the core of this engineering milestone is Habitat, an internal system that began as a Python library and subsequently evolved into a globally distributed storage platform. Today, the platform reliably sustains an unprecedented throughput of 22 million requests per second. This development illustrates the immense computational and data storage demands required to power large-scale conversational AI applications, emphasizing the critical evolution of foundational infrastructure from simple software utilities into mission-critical, worldwide distributed storage networks.

OpenAI Blog

Key Takeaways

  • Architectural Evolution: OpenAI successfully transitioned its Habitat system from a modest Python library into a full-scale, globally distributed storage platform.
  • Massive User Reach: The underlying online storage platform reliably serves more than 1 billion ChatGPT users worldwide.
  • Extreme Operational Throughput: Habitat sustains a continuous processing workload of approximately 22 million requests per second (RPS).
  • Engineering Documentation: This release marks the first installment in a series detailing OpenAI's storage infrastructure scaling journey.

In-Depth Analysis

From Python Library to Global Distributed Platform

The technological journey of Habitat highlights a foundational shift in OpenAI's engineering approach. In its early stages, Habitat functioned as a localized Python library designed to address internal data handling requirements. However, as user demand for ChatGPT experienced unprecedented exponential growth, single-environment libraries and localized data paradigms quickly became insufficient to meet runtime and reliability expectations.

To bridge the gap between rapid application development and high-availability operations, OpenAI re-engineered Habitat into a globally distributed storage platform. This structural redesign enabled the system to distribute storage state and handle requests across worldwide deployments, ensuring that ChatGPT's operational footprint could expand seamlessly without being constrained by the limits of initial software designs.

Sustaining 22 Million Requests per Second at Scale

Operating at a scale of 22 million requests per second represents an extraordinary threshold for online data storage. In conversational AI systems, storage layers must handle constant state updates, context retrieval, and real-time interaction logs while keeping latency low for interactive user sessions.

Scaling an online storage engine to process tens of millions of queries concurrently requires eliminating central bottlenecks, minimizing network round-trips, and ensuring consistent distributed throughput across all operating regions. The ability of Habitat to handle 22 million RPS signifies that the platform has matured into a resilient, high-concurrency storage tier capable of supporting interactive workloads with minimal disruption.

Supporting Over 1 Billion ChatGPT Users Worldwide

Serving an active base of more than 1 billion users places unique pressures on data partition strategies, consistency models, and global data locality. User accounts, settings, interaction histories, and active sessions generate massive data volumes that must remain reliably accessible regardless of the user's geographic location.

By evolving Habitat into a globally distributed platform, OpenAI established a shared infrastructure layer capable of managing the data footprints of one-eighth of the global population. This level of scale demonstrates how critical robust, distributed online storage is to ensuring uniform performance, data durability, and responsive service delivery for consumer-facing artificial intelligence services.

Industry Impact

OpenAI's account of scaling Habitat underscores a major challenge currently facing the broader artificial intelligence sector: the AI infrastructure race is as much about data storage and distributed systems engineering as it is about model training and compute clusters. As frontier AI models expand into everyday productivity and enterprise workflows, the underlying systems must support the same operational standards as traditional hyperscale web platforms.

The evolution of Habitat offers a clear benchmark for organizations building large-scale AI applications. It demonstrates that initial software solutions—such as standard Python libraries—must inevitably undergo architectural overhauls to become distributed, high-throughput platforms once consumer adoption reaches hundreds of millions or billions of active participants.

Frequently Asked Questions

What is Habitat and how did it originate?

Habitat is an online storage system developed by OpenAI to power ChatGPT. It originally originated as a Python library before being re-engineered into a globally distributed storage platform.

What scale does OpenAI's Habitat storage platform currently handle?

Habitat currently supports over 1 billion ChatGPT users and reliably processes 22 million requests per second across its distributed infrastructure.

Why was it necessary to evolve Habitat into a globally distributed system?

A localized library cannot sustain the extreme latency, durability, and throughput requirements of a global service. Transforming Habitat into a globally distributed platform allowed OpenAI to manage the massive concurrency and data requirements of over 1 billion users.

Related News

How Google Duplex Evolved to Call Businesses on Your Behalf in Gemini
Industry News

How Google Duplex Evolved to Call Businesses on Your Behalf in Gemini

A Google Research Blog entry titled “How Google Duplex evolved to also call businesses for you in Gemini” describes its subject as the evolution of Google Duplex to support calls to businesses on a user’s behalf in Gemini. The entry is dated November 20, 2026, and the supplied article text consists only of the label “Generative AI.” The headline identifies the capability and connects it to Duplex, but the available text does not explain how the system evolved, how users request a call, or where the feature is available. The headline establishes the topic, but the supplied material does not provide a technical account or rollout details.

How AI and Lean Are Helping Automate Formal Verification to Fight Software Slop
Industry News

How AI and Lean Are Helping Automate Formal Verification to Fight Software Slop

In Latent Space, Richard MacManus examines formal verification as an emerging way to fight what the headline calls “slop.” The central claim is that proving software correct with mathematics has traditionally required specialist work, and that AI and Lean are now helping automate that process. The distinction is important: the reported change concerns how the work of proving correctness is done, not what counts as a proof. The supplied excerpt presents automation as assistance, without saying that it replaces specialists or completes proofs independently. It also does not specify which parts of the process AI and Lean help automate, how much work they save, or which software has been verified this way.

AI and Human Claim Solutions to the 167-Year-Old Riemann Hypothesis
Industry News

AI and Human Claim Solutions to the 167-Year-Old Riemann Hypothesis

TechCrunch AI reports that AI claimed a solution to the Riemann hypothesis, a mathematical problem the headline describes as 167 years old. A human also put forward a competing solution. The report, written by Tim Fernholz and published on November 20, 2026, describes mathematicians as astonished by the AI-driven result. The supplied excerpt does not explain either proposed solution, identify the AI system or the human claimant, or say whether either claim has been verified. The distinction between a claimed solution and a confirmed one is central to reading the report: the excerpt establishes that two claims are being reported, but it does not establish that either has resolved the hypothesis.