GitHub rewrites its Git storage to cope with a flood of activity from AI agents

[MegaMike16]

GitHub is rebuilding its Git storage system from the ground up because of heavy demand from AI agents, The Register reports. Internal tests show a 35 times improvement in write performance. The scale of the growth explains why. GitHub traffic doubled between September 2025 and August 2026, from 218.2 billion to 473.3 billion events a month

Commits are rising even faster. September saw 7.38 billion commits, five times the number a year earlier. The strain has shown. GitHub's availability report logged ten performance degrading incidents in April and nine in May

The current system is called Spokes. It keeps full copies of each repository on several local disks, and every write needs a quorum of those copies to confirm it. That means the slowest replica sets the pace for everyone. It worked well for human developers, but agents committing constantly push it hard

The new design writes each commit once to Azure Blob Storage, which handles replication itself. Reads go through separate lightweight workers, and compaction and garbage collection run in the background. Principal engineer Brian Celenza described the change, and it is meant to be invisible to developers. GitHub has not given a timeline for the migration

The Register's closing line made me laugh, suggesting GitHub will need a miracle just to keep up with customer demand. It is a striking example of how AI agents are changing infrastructure behind the scenes. A five times jump in commits in one year is not normal growth. Every service that agents touch is going to face the same kind of pressure

Have you noticed GitHub being slower or less reliable this year? And are AI agents going to break more of the internet's plumbing?

So long, Spokes: GitHub rewrites storage to restore reliability, just in time for agentic hordeswww.theregister.com