GitHub recorded 7.4 billion code commits in September 2026, five times the volume that reached its servers twelve months prior, according to announcements made at the company's Universe conference in San Francisco on Wednesday. The massive increase in traffic, driven by developers running AI agents that generate expanded quantities of programming code, left the Microsoft-owned platform unprepared and triggered repeated outages and service reliability problems throughout the year.
AI-powered code commits began creating difficulties last December and sparked issues in February, April, August, and the months between those periods. The platform's 90-day uptime reached a nadir of approximately 84 percent on May 7, based on reconstructed information from GitHub's status page. Following apologies from GitHub executives and pledges to improve performance, the company has placed four million CPU cores and 160 petabytes of high-speed storage on its infrastructure roadmap, along with additional compute capacity for GitHub Actions, its continuous integration and continuous delivery platform.
The company is also redesigning its technical architecture to separate compute and storage, an approach expected to enable 35 times greater throughput. "We're also migrating to a multi-region deployment in Azure, while continuing to serve incredible levels of traffic from developers around the world," said Jay Parikh, EVP of CoreAI at Microsoft, adding that the migration should be complete early next year. At the same event, GitHub announced general availability of Copilot Fusion, its model routing service previously previewed under the name HydraFusion, which directs requests to different models to boost task-completion rates while cutting token consumption costs.
The infrastructure expansion reflects GitHub's recognition that AI-generated code is fundamentally reshaping the demands placed on code hosting platforms. The separation of compute and storage, combined with multi-region Azure deployment, aims to create the architectural flexibility needed to absorb traffic spikes that can overwhelm traditional systems. The Copilot Fusion approach represents another efficiency strategy: by evaluating what each task requires across reasoning, code generation, debugging, and tool use, then selecting how to execute across one or more models, the service reportedly scored within two points of Opus 5.5 Max at half the cost and 40 percent less time per task in the Artificial Analysis Coding Agent Index.
GitHub expects to finish its multi-region migration early next year, positioning the platform to handle continued growth in AI-driven development activity. The company is extending Copilot access to all GitHub Enterprise users without additional per-seat charges, billing customers instead for token usage, and adding AI Scans to check pull requests for vulnerabilities plus a package firewall that verifies enterprise policy settings before fetching npm packages. The combination of hardware expansion, architectural redesign, and expanded AI tooling signals GitHub's bet that code generation volumes will continue climbing as developers lean more heavily on AI agents. For platform operators across the tech industry, the GitHub outage crisis offers a preview of infrastructure stress tests to come: legacy architectures built for human-paced workflows may buckle under AI-generated loads unless fundamental separation of compute and storage happens first.