OpenAI Is Building Its Own Exabyte Storage System
OpenAI published Part II of its Habitat engineering series today, covering how it scaled Azure Cosmos DB, the database layer under ChatGPT, the API, Codex and its internal services. Read it for what it is: OpenAI's own account of OpenAI's own infrastructure, not an independent evaluation.
Transcript
OpenAI published new storage engineering details today. The company says it is now building its own storage system.
Habitat is the storage platform behind ChatGPT. OpenAI says it holds over five hundred petabytes.
Its largest Azure Cosmos DB account serves more than twenty million queries per second. Two earlier outages forced OpenAI and Azure to operate it jointly.
OpenAI says global data lookups got roughly twelve times faster this year. It is now building a custom storage system at exabyte scale.
These figures come from OpenAI describing OpenAI infrastructure. Every number is self reported, with no published benchmark method. No outside audit exists.
Inboxsmith builds AI phone and email agents for small business. Please like and subscribe for more news.
Sources
Every claim in this video comes from the top ranking coverage of this topic. The claims and where each one came from:
- Every factual claim in this script comes from this single first-party OpenAI engineering post, including the five hundred petabyte platform size, the largest Azure Cosmos DB account serving more than twenty million queries per second, the two late regional outages that led to joint operation with Azure, the global median read latency falling from roughly one hundred twenty milliseconds to roughly ten milliseconds, and the custom in-house exabyte-scale storage system now in development. No third-party corroboration or video transcript was available.(OpenAI engineering blog post titled "Rapidly scaling online storage to serve over 1 billion ChatGPT users (Part II)", by Jon Lee, Hrachya Hakobyan, and Chaomin Yu, Members of Technical Staff, published on OpenAI's own first-party website.)