GitHub adds 4M CPU cores and 160 PB storage, makes Copilot Fusion model router generally available

GitHub is adding more than 4 million CPU cores and 160 PB of storage to its infrastructure, and has made Copilot Fusion, an engine that routes tasks across frontier AI models, generally available. Both were announced ahead of and at GitHub Universe, the company's annual developer conference, as reported by Analytics India Magazine.

Infrastructure expansion

The new capacity follows a build-out in August, when GitHub provisioned more than 3 million CPU cores and 120 PB of high-speed storage after a series of reliability incidents.

GitHub recorded 7.38 billion commits in September alone, a 5x increase over the previous month, driven by AI agent activity. Jay Parikh, EVP of CoreAI at Microsoft, wrote in a blog post: "Agents are changing the volume and shape of software development. Adding capacity matters, but capacity alone won't get us there. We've also had to change the architecture."

GitHub is rebuilding its Git architecture to decouple compute and storage. The company says this could enable 35x more throughput and keep developer tools from buckling under synthetic load.

Copilot Fusion

Copilot Fusion decides which frontier models to dispatch based on task difficulty, balancing cost, latency and reasoning capability. It offers three effort tiers: Standard, High and Max.

According to GitHub, the Max tier improves task-completion rates over top-tier frontier baselines while capping estimated inference cost at 1.1 times the baseline model. These are company claims, and no benchmark figures were provided.

Parikh described the approach: "Copilot Fusion evaluates what a task needs across reasoning, code generation, debugging, and tool use, then chooses how to execute across one or more models." GitHub did not name the models Fusion routes between in the announcement coverage.

Security features

GitHub is rolling out three supply-chain and code-quality protections:

• AI-powered code scanning of every pull request and repository for vulnerabilities.
• Reachability analysis, which shows whether code calls a vulnerable function, with a call path developers can inspect in GitHub, so teams can prioritise alerts.
• Package firewall, which inspects npm packages at download time to block known malicious packages, typosquatted names and zero-day dependency attacks.

Parikh also said Copilot Code Review "surfaces meaningful, actionable bugs that developers need to catch."

Copilot app additions

GitHub is expanding the Copilot app into a project manager:

• Copilot Projects tracks goals, context, active terminal sessions and work-in-progress code diffs across parallel development tracks.
• Copilot Connectors integrates the Model Context Protocol natively, letting Copilot pull production telemetry and deployment context from services such as Cloudflare, Vercel and Sentry across the CLI, web app and VS Code.
• Copilot Automations lets teams configure reusable background agent tasks.

Copilot access is also expanding to all GitHub Enterprise users without additional seat purchases.

What this means

GitHub is treating agent-generated activity as a load problem that capacity alone does not solve. The company pairs a second large hardware addition with an architectural change that separates compute from storage, and Parikh's own statement frames the architecture work as necessary alongside the added cores.

Fusion moves model selection from the developer to the platform. Teams using Copilot would choose an effort tier rather than a specific model, with GitHub's stated 1.1x cost cap on the Max tier as the main published cost control. The task-completion improvement is GitHub's claim, and the announcement coverage gives no benchmark numbers to test it against.

The security additions target code and dependencies that agents produce or pull in at volume. Extending Copilot to all Enterprise users without extra seat purchases widens the number of developers who can reach these features.