GitHub’s Rubicon architecture separates repository data storage from compute to handle agent-scale traffic

GitHub says internal tests of its new Git storage architecture handled 5,000 concurrent agents and 7,200 pushes per second to one repository while maintaining consistently low-latency reads. The system stores Git objects in Azure Blob Storage and keeps references and pack manifests in a separate replicated log, allowing read, write, and maintenance workloads to scale independently; Git clients keep using the same workflows and protections.

For teams building high-write repository services, separating immutable Git objects from the small, ordered reference updates offers a way to add write capacity without making every storage node coordinate. GitHub reports internal test results, not production limits or independent benchmarks, so the article leaves open how performance and latency hold up under sustained real-world load.