GitHub is rebuilding its Git infrastructure as AI agents drive rapid growth in platform activity. Storage and compute are being decoupled, with Azure Blob Storage serving as the foundation. Internal benchmarks show the new architecture achieves up to 35 times higher write throughput.
GitHub commits are rising rapidly. In September, developers and agents collectively made 7.38 billion commits, more than five times as many as a year earlier. Between September 2025 and August 2026, total Git activity grew from 218.2 billion to 473.3 billion events per month.
A median repository processes about 25 requests per month; the busiest one handled roughly one billion in August. According to GitHub, this is driven by large engineering teams with busy CI pipelines and a growing number of AI agents.
Why are agents putting so much strain on the system? An agent runs in a tight loop, committing or checking out after virtually every action, turning latency, which a human would never notice, into the limiting factor. Furthermore, CI jobs clone or fetch the same branch tip thousands of times per minute. GitHub Actions ran 3.26 billion times in September, more than four times as many as a year earlier.
Replicas slow down writes
The current system, Spokes, stores each repository by default on the local disks of five file servers. A three-phase commit protocol with quorum ensures that CI, the web UI, and API clients see a consistent state. That setup currently serves one billion repositories.
But the copies on disk are both the source of truth and the means of scaling. Every replica participates in every write, so a push is only as fast as the slowest replica. GitHub sums it up: “Adding replicas to absorb read load makes writes slower.” If the quorum is lost, writes come to a complete standstill.
Storage and compute separated
The new architecture separates the two layers. The authoritative repository data is stored in Azure Blob Storage, which already handles durability and replication. Lightweight compute workers cache data for read requests and scale with traffic. If a worker fails, GitHub says it’s comparable to a cache miss.
Coordination is also minimized. Only reference updates still require consensus; object storage, connectivity checks, and secret scanning largely run in parallel. Compaction and garbage collection are moved to separate workers, away from the hosts that handle live Git requests. In internal benchmarks, the new setup delivered up to 35 times higher write throughput.
GitHub emphasizes that branch protections, mandatory reviews, and audit logs remain in place. The restructuring is happening without a maintenance window, and the platform continues to run as usual.