GitHub is rebuilding the storage system behind its Git repositories after coding agents drove a surge in commits, pushes and pull requests, the company said in a post on the GitHub Blog Tuesday. The new design moves repository data to Azure Blob Storage and limits real-time coordination to the step that updates branch references.

What Happened

The post, by principal software engineer Brian Celenza, laid out how quickly load has grown. Monthly Git events rose from 218.2 billion in September 2025 to 473.3 billion in August 2026. Commits reached 7.38 billion in September 2026, more than five times the figure a year earlier. Monthly pushes went from 0.69 billion to 3.35 billion, a 4.9-fold increase, and pull request merges grew nearly fourfold. GitHub Actions logged 3.26 billion runs in September, more than four times the year-earlier level. The single busiest repository handled about 1 billion requests in August.

GitHub's current architecture, known as Spokes, keeps a full copy of each repository on local disks, by default across five file servers, and uses a quorum-based three-phase commit so the copies agree. That keeps data consistent, but every replica takes part in every write. Adding replicas to absorb heavier read traffic therefore makes pushes slower, the post said.

RelatedClaude Code Desktop Preview Flaw Could Have Exposed Developers’ Cloud Keys, Researchers Say

The replacement separates storage from compute. Durable repository data will live in Azure Blob Storage, and lightweight compute workers will serve and cache requests. Losing a worker becomes a cache miss rather than a risk to the data. Only the reference update at the end of a push needs agreement across the system; writing objects, validation, scanning and maintenance jobs such as compaction and garbage collection move to parallel or background workers.

In internal benchmarks, GitHub said the design delivered up to 35 times higher write throughput, with read capacity that scales independently.

Why It Matters

Coding agents work differently from people. They open many branches, commit in small increments and trigger CI runs on each change, so the same number of developers can generate several times the traffic. GitHub's numbers put a figure on that shift for the largest code host: write-heavy activity such as pushes is growing faster than overall events.

A design in which reads and writes are tied together breaks down under that pattern, because the replicas that help serve clones and fetches also slow the pushes agents depend on. Decoupling them lets GitHub add read capacity without penalizing writes.

GitHub said branch protections, required reviews, access controls, audit logs and repository visibility stay as they are, and that branching, review and merge workflows will not change for users. It is migrating while the service stays up, without a maintenance window.

Context

GitHub, owned by Microsoft, has been adding agent features at a steady clip, including computer-use capabilities for Copilot that let it operate desktop applications. The infrastructure post is a rare look at what that activity costs on the back end. Moving repository data into Azure's object storage also ties GitHub's core service more closely to its parent's cloud.

What to Watch

GitHub said the next post in the series will go deeper into the future architecture and how the team arrived at it. Developers will watch for any change in push latency or availability during the migration, and for whether GitHub adjusts pricing or rate limits as agent traffic keeps climbing.