GitHub adds 4 million CPU cores to cope with a wave of AI agents

GitHub, the Microsoft platform where millions of programmers store their code and work on software together, is growing very fast because of AI agents: AI programs that write and change code on their own. According to GitHub, 7.38 billion commits were made in September 2026. A commit is a saved change to the code. That is more than five times as many as a year earlier.

To keep up with that growth, GitHub has added 4 million processor cores, 160 petabytes of fast storage and 50 terabits per second of network capacity this year. That is what Jay Parikh, who leads the CoreAI division at Microsoft, writes in a blog post for GitHub's annual Universe conference. 160 petabytes is roughly as much as 160,000 large hard drives of 1 terabyte each.

Everything is growing at once

It is not only the number of commits that shot up. According to GitHub, the number of pushes, the moment when saved changes are sent to GitHub, grew in one year from 0.69 billion to 3.35 billion a month: almost five times as many. The number of merged code proposals (pull requests) grew almost fourfold. And GitHub Actions, the service that tests and builds code automatically, ran 3.26 billion times in September, more than four times as often as a year earlier.

This is down to the way agents work. An agent saves its work after almost every step. So the speed at which GitHub can process a single change decides how quickly the agent can carry on. On top of that, every change leads to thousands of read requests, for example from test systems and security scanners that fetch the same code again and again.

That flood of work caused problems. According to The Register, which reconstructed data from GitHub's status page, availability over ninety days fell to around 84 per cent on 7 May 2026, with outages in months including February, April and August. Senior GitHub staff apologised and promised improvements.

Separating storage and computing power

GitHub is rebuilding its systems while the platform keeps running. The core of the work is separating storage from computing power. At the moment, full copies of each code folder (repository) sit on the disks of several servers, which store the data and answer all requests at the same time. Every extra copy makes saving a change slower.

In the new set-up, the data is stored centrally in Azure Blob Storage, Microsoft's storage service. Light computing servers with their own cache handle the read requests, and more of them can be switched on when it is busy. Maintenance work, such as clearing out old data, moves to separate servers. According to internal tests, the new set-up can process up to 35 times as many changes per second. GitHub is also moving to several regions in Azure. Parikh expects that move to be finished early in 2027.

At Universe, GitHub also announced Copilot Fusion, a service that chooses for each task which AI model, or which combination of models, does the work. According to GitHub, this should give better results at lower cost. Customers with GitHub Enterprise can now use all features of the AI assistant Copilot without a separate subscription; they pay for their AI usage.

The growth figures come from GitHub itself.

What does this mean for the road to the singularity?

This news shows what AI agents already mean for the physical foundations of software development. Work that a year ago was mostly done by people now calls for millions of extra processor cores and huge amounts of storage. As agents write more and more code, the limit shifts from the number of programmers to computing power, storage and networks. GitHub is now building an infrastructure that can grow along with that demand. If that succeeds, agents will soon be able to build software even faster and on an even larger scale.

What is the technological singularity? Read the explanation →