Thursday, 19 November 2026

GitHub rebuilds storage to protect AI tool performance

AI agents have more than doubled GitHub's Git traffic in a year. Here's how GitHub is replacing the storage system behind every repository, and what changes for developers.

GitHub
Photo: Jun OHWADA / CC BY 2.0 (modified)

The short version

  • GitHub says Git activity more than doubled in a year, from 218.2 billion to 473.3 billion events per month, largely because of AI agents.
  • Its storage system, Spokes, keeps five full copies of each repository, and every push waits on the slowest copy, which caps write speed.
  • The new design stores repository data once in Azure Blob Storage and serves reads from separate cache workers; internal tests showed far faster writes.
  • GitHub says developers will not need to change their workflows, and nearly every repository already has copies in Azure.
Quick read · 2 min

GitHub is rebuilding the storage behind every repository because AI agents are flooding it with work. Git activity more than doubled in a year, to 473.3 billion events a month, and September saw 7.38 billion commits, over five times the year before.

The current system, Spokes, keeps five full copies of each repository, and every push has to wait for the slowest one. The new design stores the main copy once in Azure Blob Storage, uses separate cache workers for reads, and runs cleanup in the background. GitHub says early tests showed up to thirty-five times faster writes.

  • Nearly every repository already has copies in Azure.
  • The busiest ones are moving next, using per-branch locking.
  • Developers shouldn't need to change anything.

GitHub hasn't given a finish date. Meanwhile, former CEO Thomas Dohmke's agent-focused rival, Entire, launched on October 26.

In September, developers and their AI agents made 7.38 billion commits on GitHub. That is more than five times as many as a year earlier, and it is the reason GitHub is now tearing out and rebuilding the storage system that holds every repository on the site.

If you write code, this matters in a very practical way. GitHub has had a rough year for outages, and when it goes down, pull requests, CI pipelines and deploys stall with it. The storage rebuild is GitHub’s main answer to the question many teams have been asking: can the platform keep up with agents that push code nonstop? Here is what is changing, why, and what it means for you.

01

Why GitHub storage reliability broke down under AI agents

GitHub storage reliability is suffering because AI agents behave nothing like human developers. A person might push a few times a day. An agent working in a loop can commit or checkpoint after almost every step, and companies now run thousands of them against a single codebase.

GitHub’s own numbers show the scale. Between September 2025 and August 2026, total Git activity climbed from 218.2 billion events per month to 473.3 billion. Pushes rose 4.9x, from 0.69 billion to 3.35 billion a month. GitHub Actions, the built-in automation service, ran 3.26 billion times in September, more than four times the figure a year earlier. The single busiest repository handled roughly a billion requests in August.

Every one of those pushes also sets off a wave of reads. Test systems and code scanners fetch the same fresh code thousands of times per minute. Merge queues pile work onto one main branch that has to accept every change in order.

data center rack with cabling
02

The problem with Spokes, GitHub’s current storage system

Today every repository lives in a system GitHub calls Spokes. It keeps full copies on the local disks of several file servers, five by default. When you push, a three-phase commit protocol makes sure a majority of those copies agree before the change counts.

That design has served about a billion repositories well. The catch is that the same copies handle both safety and speed. To serve more reads, GitHub has to add another full copy, and every copy takes part in every write. A push is only as fast as the slowest copy involved.

So adding capacity for reads makes writes slower, and losing too many copies can stop writes entirely. For most projects that trade-off is invisible. For the busiest repositories, where agents push constantly, it has become a hard ceiling.

03

What the new storage design changes

GitHub engineer Brian Celenza laid out the plan in early October. It rests on two ideas: coordinate only what truly needs agreement, and separate where data is stored from the computers that answer requests.

  • Store once, in the cloud. The authoritative copy of each repository moves to Azure Blob Storage, Microsoft’s object storage service, which handles duplication and durability by itself. That removes the need for several servers to vote on every push.
  • Serve reads from cache workers. Lightweight computers keep cached copies to answer clones and fetches. They can be added during a busy release and removed afterward, without slowing writes.
  • Push housekeeping to the background. Cleanup jobs that compact repository data and delete unused objects move off the servers handling live requests.
  • Recover faster. If a cache worker fails, a new one can start serving straight away and refill its cache from storage, instead of rebuilding a full copy of the repository.

In internal benchmarks, GitHub says the new setup delivered up to thirty-five times higher write throughput. It has not given a date for when the full switch will be finished.

developer working at a computer with code on screen
04

The migration is already well under way

A second GitHub post, published in late October, showed how far the move has come. The first repository copy landed in Azure’s Central US region in August 2025. Today most repositories keep two of their five copies in Azure, and Azure serves about half of all Git traffic.

GitHub says 99.99999% of repositories now live in both its own data centers and Azure. The holdouts are some of the busiest on the site, and they are stuck on distance. The Azure region is about 1,000 miles away, so each network round trip takes around 25ms, and a busy repository can’t afford that delay on every write.

The fix is to stop locking the whole repository for each update. Instead, GitHub will lock only the specific branch or tag being changed. Agents usually push to their own branches, so their writes no longer wait in one long line. In internal testing, writes to different branches saw three to four times more throughput at lower latency.

GitHub also explained why it is moving so fast: growing traffic, a hard limit on the power available to its data centers and their region, and supply chain constraints on new hardware.

05

What this means for developers and teams

For most users, the goal is that nothing visible changes. GitHub says there will be no maintenance window and no need to change how you work. Branch protections, required reviews, audit logs and repository visibility settings are meant to carry over as they are.

If you run a busy monorepo or a large fleet of coding agents, you should see the biggest gains: faster pushes, fewer timeouts during heavy CI runs and less waiting when many agents work at once. Smaller projects should benefit from the same, more resilient foundation.

There is competition, too. Former GitHub CEO Thomas Dohmke launched Entire, a code-hosting service built for agents, on October 26 with $60 million in funding. Entire claims its push throughput is far higher than GitHub’s under the same load, though it admits those figures have not been independently checked. GitHub held its Universe developer conference in San Francisco two days later, with agents at the center of the program.

Until the rebuild is done, the practical advice stays the same: keep an eye on GitHub’s status page during big releases, and give critical CI jobs sensible retry settings.

06

Quick answers

Why is GitHub rewriting its storage?

AI agents more than doubled Git activity in a year, and the current Spokes design slows every write as it adds copies to handle reads.

Will I need to change how I use GitHub?

No. GitHub says workflows, reviews and security controls stay the same, and the move happens without downtime windows.

Where will GitHub store repositories?

The authoritative copy will live in Microsoft’s Azure Blob Storage, with separate cache workers handling reads.

You're reading the quick version.