The GitHub platform has officially announced its new infrastructure for Git version control storage, which has been specifically designed to handle thousands of writes per second while providing scalable reads on demand. Internal tests have demonstrated the new architecture's ability to support 5,000 AI agents working concurrently on a single repository, sending 7,200 pushes per second, while maintaining low and consistent latency for read requests.
This step comes in response to the radical change that software agents have brought to the nature of working on busy repositories over the past year. The new architecture relies on two fundamental principles: reducing coordination between components, and separating storage from compute by relying on Azure Blob Storage object storage service and a dedicated reference consensus service named Ref-CAS.
What changes in practice?
In the previous traditional architecture, storage operations, request services, and repository maintenance were closely and locally linked. As for the new architecture, it separates reference state from object data, where Git packs and read accelerators are stored in cloud storage, while the Ref-CAS service handles the management and updating of repository references via a Compare-and-Swap mechanism supported by a quorum of replicas.
Maintenance, compaction, and downsizing operations have also been separated from the nodes that serve daily traffic, as independent compaction nodes work in the background directly against cloud storage to merge packs and remove unused objects without consuming the resources of the nodes dedicated to users.
Why does this news matter?
This development represents the technical foundation to enable the GitHub platform to support the era of AI agent-driven development for hundreds of millions of developers. As AI tools and autonomous agents shift to working around the clock, repositories face massive rates of concurrent updates and associated requests, such as running continuous integration systems and code testing. The new architecture ensures the platform's stability and its ability to scale horizontally and vertically with the same capacity and efficiency.