GitHub powering up four million extra cores to stop its service going down
Re-architects site to cope with AI slop-pocalypse
GitHub's Universe conference in San Francisco unveiled plans to boost its infrastructure by adding four million CPU cores and 160 petabytes of high-speed storage. This expansion aims to handle the significant surge in demand from AI agents, which resulted in 7.4 billion code commits in September 2026, five times more than in the previous year.
The increase in traffic over the past year was due to developers running AI agents and generating more programming code, leading to outages and service availability issues. AI-driven code commits started causing problems in December, with notable outages in February, April, and August. On May 7, GitHub's 90-day availability dipped to around 84%, according to reconstructed data from its status page.
The company's executives apologized and promised improvements, which now include redesigning its technical architecture to separate compute and storage, aiming for 35 times more throughput. Additionally, GitHub is migrating to a multi-region deployment in Azure while continuing to serve high levels of traffic from developers worldwide.
The company also announced the general availability of Copilot Fusion, a model routing service that routes requests to different AI models to improve task completion rates and reduce token consumption costs. Copilot Fusion evaluates tasks across reasoning, code generation, debugging, and tool use, choosing the best model for execution.
GitHub is further enhancing its Copilot app with features like Copilot Projects for coordinating parallel work, Copilot Connectors for simplifying connections to MCP servers, and Copilot Automations for reusable agent tasks. Additionally, GitHub is extending Copilot access to all GitHub Enterprise users without extra charges, while still billing customers for the tokens they use.
The company is also expanding its suite of AI services for developers with AI Scans, which checks pull requests for vulnerabilities in languages not covered by rule-based scanning and provides reachability analysis tools for focusing on potentially exploitable issues.
Written by urgent.news from The Register's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
