GitHub rewrites Git storage from scratch to cope with AI agent traffic surge

Move to object storage and decoupled reads and writes follows months of degraded performance as platform traffic doubles

By LineZotpaper
Published
Read Time3 min
GitHub is rebuilding its underlying Git storage architecture, responding to unprecedented demand from AI agents that has doubled platform traffic in little over a year. Early internal tests of the new design show a 35x improvement in write performance, according to the company.

GitHub has announced a ground-up redesign of its Git storage architecture, driven by a surge in activity as customers increasingly shift coding work to AI agents. The company said the changes are designed to keep the service reliable for large engineering teams running busy CI pipelines alongside growing fleets of automated agents.

The scale of the problem is significant. Between September 2025 and August 2026, GitHub's monthly traffic more than doubled, from 218.2 billion events to 473.3 billion. In September 2026 alone, commits reached 7.38 billion, a fivefold increase from September 2025.

That growth has taken a toll. According to GitHub's Availability Report, the service suffered 10 performance-degrading incidents in April and a further nine in May.

The current storage architecture, nicknamed Spokes, uses a three-phase commit protocol that stores full repository copies across multiple local disks and requires a quorum of replicas to acknowledge each write. That approach provides strong reliability, but it also creates a bottleneck: every push is limited by the slowest replica required for quorum.

The new architecture writes each commit only once, to Azure Blob Storage, which handles replication and redundancy automatically. Read requests move to a separate channel handled by lightweight compute workers, with coordination between reads and writes limited to reference branch pointers. Maintenance tasks such as compaction and garbage collection shift to background processes, removing them from the serving path.

Brian Celenza, GitHub principal software engineer, described the move in a blog post as building Git infrastructure for sustained, concurrent reads and writes at a scale few repositories reach today. He did not provide a timeline for the migration.

GitHub is not alone in rethinking Git for the agentic era. Former GitHub CEO Thomas Dohmke has launched a Git-hosting service called Entire, which offloads agent traffic to mirror repositories. SpaceX subsidiary Cursor, which uses Git to back its agent-support service, also reworked its storage layer, replacing Spokes-style three-phase commits with an object storage write-ahead log and cached copies on solid-state disks.

PlanetScale CEO Sam Lambert, commenting on X, said he had seen companies spend three months preparing for events that would increase traffic by 10% or 20%, which he described as nothing compared with what GitHub now faces.

§

Analysis

Why This Matters

  • GitHub hosts a large share of the world's source code, and its reliability directly affects millions of developers and the companies employing them.
  • The shift from human-driven pushes to AI agent traffic changes the shape of demand, and GitHub's response will influence how other Git hosting services adapt.
  • If the migration succeeds without disrupting workflows, it could quietly raise the ceiling on automated software development at scale.

Background

Git is the most widely used version control system in software development, and GitHub is its largest collaborative platform. For years, the service has stored repositories using an architecture designed for human-paced pushes and pulls. AI coding agents change that assumption: they generate commits, open pull requests and run builds far more frequently than human developers. The resulting load has pushed storage systems that were designed for a different era, prompting companies across the industry to experiment with object storage and write-ahead logs as alternatives to traditional replica-based storage.

Key Perspectives

GitHub: The company frames the rewrite as a necessary, under-the-hood change that will sustain concurrent reads and writes at a scale few repositories reach today, without altering developer workflows, review processes or security controls. Users and engineering teams: For organisations running busy CI pipelines and growing agent fleets, the promise is fewer incidents and faster pushes. Their interest is in stability and a non-disruptive transition. Competitors and alternatives: Entire and Cursor have both taken different technical approaches, suggesting a broader industry consensus that traditional Git storage is straining under agent traffic, but no single solution has emerged. Critics and skeptics: The absence of a public timeline is a concern, as is the ambition of replacing a proven, strongly consistent architecture while traffic continues to grow. Migrations at this scale carry risk of new failure modes.

What to Watch

  • The GitHub Availability Report in coming months for signs that degraded-performance incidents are declining.
  • Any announced rollout dates or phased migration milestones from GitHub engineering.
  • Whether other Git hosting services adopt similar object-storage-backed designs or continue with WAL-based alternatives.

Sources

Zotpaper

Written by software from the reporting listed above, scored by an automated standards desk, and published without a person reading it first. If something here is wrong, tell the editor and it will be put right.

How we workSubscribe