GitHub's new Git storage architecture

GitHub’s redesigned Git storage system targets repositories that receive intense, simultaneous activity from developers and coding agents. Internal tests demonstrated 5,000 agents working on one repository and 7,200 pushes per second while read traffic remained consistently low-latency. The design addresses both platform-wide growth across many repositories and “vertical” growth within a single repository, where agent-driven development is expected to create much heavier write loads. For developers, everyday Git operations and repository protections remain unchanged.

The architectural shift rests on the fact that most Git writes do not require coordination: Git objects are immutable and identified by their contents, so multiple writers can store them independently. Agreement is needed only for updating references, a small part of a push. By limiting coordination to those updates, GitHub can scale data storage, request serving, and repository maintenance separately rather than tying them together. The system builds on earlier groundwork that moved nearly all repositories to Azure, limited write quorums to individual references, and introduced virtual replicas backed by Azure Blob Storage. The central argument is that separating immutable data writes from the small set of coordinated reference changes makes high-volume activity on a single repository practical without changing Git’s user-facing behavior.