GitHub adds four million cores, redesigns architecture, launches Copilot Fusion

The Microsoft-owned code hosting business is bolstering its infrastructure with four million additional CPU cores and 160 petabytes of high-speed storage to combat overwhelming demand stemming from AI agents.

This massive influx of traffic, driven by developers utilizing AI agents to produce greater volumes of programming code, caught the company off guard, resulting in a string of service disruptions and availability issues. GitHub saw a staggering 7.4 billion code commits in September 2026, which represents five times the volume recorded a year prior. The complications related to AI-driven code commits began to surface last December, leading to problems in February, April, August, and throughout the intervening months. Data reconstructed from GitHub's status page indicates that the site's 90-day availability dipped as low as approximately 84 percent on May 7th. Executives at GitHub issued apologies, pledging to improve the situation.

To address these strains, the company has incorporated these commitments into its infrastructure plan, alongside adding extra compute capacity for GitHub Actions, its currently struggling continuous integration and continuous delivery (CI/CD) platform. Furthermore, the firm is undertaking a redesign of its technical architecture—a project previously discussed—aimed at separating storage from compute, which is anticipated to boost throughput by a factor of 35.

Jay Parikh, EVP of CoreAI at Microsoft, stated in a blog post shared with The Register, "We’re also migrating to a multi-region deployment in Azure, while continuing to serve incredible levels of traffic from developers around the world." He added, "We expect to complete the migration early next year. Those investments give us more room to grow. The next step is building the systems that handle the work."

During its Universe conference in San Francisco on Wednesday, GitHub also made Copilot Fusion generally available; this is the company's model routing service, previously previewed under the moniker HydraFusion. The purpose behind enabling GitHub to direct requests to various models is to enhance task completion rates while simultaneously lowering the expense associated with token consumption. Parikh explained, "Different tasks need different models." He continued, "Developers shouldn’t have to manage every model choice to get the right result at the right price." Copilot Fusion assesses a task's requirements across areas like reasoning, code generation, debugging, and tool utilization, subsequently deciding how to execute it using one or more models. In testing, Fusion Max achieved results within two points of Opus 5.5 Max while costing half as much and requiring 40% less time per task, according to the Artificial Analysis Coding Agent Index.

GitHub is also rolling out additional features within its Copilot application, including Copilot Projects for managing parallel efforts, Copilot Connectors to simplify how Copilot interfaces with MCP servers, and Copilot Automations, which are reusable agent tasks. In what appears to be an effort to increase the adoption of Copilot within professional settings, GitHub is extending Copilot access to all GitHub Enterprise users without demanding extra per-seat purchases; however, customers will still be charged for the tokens they consume.

Moreover, Microsoft's code collaboration venture is broadening its portfolio of AI services for developers with a new feature called AI Scans. This tool scrutinizes pull requests for vulnerabilities in languages that aren't covered by rule-based scanning, as well as in code that has already been merged. It also introduces new reachability analysis tools, allowing development teams to concentrate on issues that have a realistic chance of being exploited. Another new provision is a package firewall designed to verify enterprise policy settings before any npm packages are fetched.

Why it matters

This infrastructure overhaul directly addresses service instability caused by the 7.4 billion code commits recorded in September 2026. Developers should watch how these capacity upgrades impact the reliability of CI/CD tools like GitHub Actions. The new Copilot features also signal a major shift in how AI assists in the development workflow.