Microsoft owned GitHub is expanding servers and databases 30 fold as AI coding tools push commit volume from 1 billion a year to 1.4 billion a month.
The world's code lives on a platform designed for human programmers, and that platform is being rebuilt right now because AI coding agents have pushed it past its original design. GitHub, the code-hosting site owned by Microsoft, has gone from roughly 1 billion commits a year to 1.4 billion commits a month as AI tools began writing and pushing code with limited human review, and the company is now racing to expand server and database capacity by 30 times to keep up.
The numbers come from GitHub's May 2026 Availability Report, summarized this week by The Register's Thomas Claburn. The report attributes the surge in traffic to AI-assisted coding and agentic development workflows, and acknowledges nine incidents that degraded performance in May, one fewer than in April. Reliability, in other words, is still uneven. The fix is not a patch but a generational rewrite.
GitHub is responding with structural changes. It is expanding raw capacity. It is moving more of the platform's traffic to Microsoft's Azure cloud, with roughly 40 percent of the old monolithic core now running there. And it is splitting up the database clusters, the tightly coupled storage systems that, when they falter, take the whole site with them. The goal, as GitHub's SVP of software engineering Jakub Oleksy put it in comments reported by The Register, is structural changes that permanently remove failure modes.
Oleksy also acknowledged that the work is far from finished. "More work remains to be done on reliability," he said, a candid framing of a company that knows its own architecture has fallen behind the demand curve. The 30-fold capacity target is itself an admission of how far behind the old design had drifted.
The shape of the response matters because GitHub is not just any website. It is where the bulk of the world's open source and a growing share of private corporate code is stored, reviewed, and shipped. When the platform wobbles, so do release schedules, CI pipelines, and security workflows downstream of every pull request (the change-by-change merge requests developers send to each other). The Register's framing, that outages persist "despite" the move to Azure, captures one part of the story. The other part is that moving to Azure was necessary but nowhere near sufficient. AI-assisted coding has changed the unit economics of the platform, and the structural rewrite is the response to that change, not a substitute for it.
What to watch next: whether the May incident count, which already improved by one over April, keeps trending down as the database cluster isolation work lands; whether the remaining 60 percent of the monolithic core gets migrated to Azure without introducing new failure modes; and whether the company publishes a public status timeline granular enough to let customers tell the difference between a routine blip and a structural regression.