How AI-Driven Coding Overwhelmed GitHub’s Infrastructure
GitHub’s AI-driven capacity crisis is the rapid overload of its code-hosting and collaboration infrastructure as AI coding agents trigger far more commits, requests, and background operations than its engineers forecast when planning the platform’s move to Microsoft Azure. This surge has pushed GitHub into repeated performance issues, outages, and emergency scaling decisions. GitHub’s own availability reports show a platform under strain. In May, the company logged nine incidents that degraded performance, following a similar pattern in April. Executives say structural changes are underway, such as isolating the primary database cluster and separating user, authentication, and authorization domains to stop failures from cascading. Yet the central problem is volume. AI-assisted development now drives huge spikes in pull requests, repository activity, and traffic from agentic workflows, turning GitHub into both a code archive and an active execution surface for AI systems that read, write, and modify code continuously.
From 10x to 30x: Forecasts That Fell Behind AI Reality
GitHub’s recent history shows how quickly AI adoption has outpaced even aggressive scaling plans. In October 2025, GitHub aimed for a 10x capacity increase, but by February 2026 engineers concluded they needed 30x expansion instead as AI workflows drove an unexpected flood of pull requests, commits, and new repositories. According to GitHub’s leadership, the platform served only 8 percent of its monolith traffic from Azure in February; by May that had reached 40 percent, with Git traffic at 30 percent and repository replication at 99 percent. They say this has more than doubled effective capacity in four months, yet availability issues persist. The numbers behind AI usage explain why. GitHub reportedly handled 1 billion commits in the whole of last year. Now it sees 1.4 billion commits every month, and internal projections suggest 14 billion commits in 2026. That leap reflects not gradual adoption but a structural shift in how code is produced.
Why Microsoft Turned to AWS Despite Owning Azure
The most striking twist in GitHub’s AI surge is Microsoft’s decision to bring Amazon Web Services into the picture. GitHub had long planned a full migration to Azure by 2027, and Microsoft has marketed Azure as the natural home for advanced AI workloads. Yet the speed and scale of AI-driven development turned those plans into a bottleneck. Business Insider reports that Microsoft is adding overflow GitHub capacity on AWS after AI coding tools swamped internal resources. A Microsoft spokesperson confirmed that GitHub is now using multiple cloud providers, saying the “incredible spike in agentic development” has tested infrastructure limits and that the company is exploring a multi-cloud strategy for future capacity and elasticity. For Microsoft, this AWS GitHub partnership is awkward but pragmatic. Azure remains the strategic priority, but during traffic spikes and GitHub infrastructure outages, engineers need reliable compute immediately, regardless of which logo is on the data center wall.
Agentic Development, Downtime, and the New AI Platform Scaling Playbook
The root cause of GitHub’s instability is not only more developers but more autonomous software. Early tools like autocomplete Copilot suggestions were relatively light on infrastructure. Agentic development is heavier: AI agents scan repositories, plan work, modify code, run tests, and open pull requests, generating many more operations per feature change. GitHub has leaned into this shift with Copilot Pro Plus and Enterprise offerings, adding coding agents from OpenAI and Anthropic’s Claude across GitHub, GitHub Mobile, and Visual Studio Code. When outages hit, they now stall human developers and AI agents at once, compounding delays for pull requests, CI pipelines, and code reviews. This pressure has already pushed GitHub to pause new Copilot subscriptions at times and move Copilot pricing to usage-based GitHub AI Credits. Both reliability incidents and billing changes point in the same direction: AI activity is no longer a side feature—it defines GitHub’s infrastructure and economic model.
Cross-Cloud AI Partnerships: From One-Off Fix to Industry Trend
Microsoft’s decision to extend GitHub capacity through AWS underlines a broader shift in how AI platforms scale. When AI demand grows faster than any one cloud’s build-out, cross-cloud arrangements move from unthinkable to unavoidable. GitHub’s multi-cloud strategy is one example; similar patterns are emerging elsewhere as AI compute and storage needs spike. Business Insider points to other deals that show the same dynamic, including Google Cloud selling AI compute to Anthropic and Google agreeing to pay SpaceX USD 920 million (approx. RM4.3 billion) a month for AI capacity from October 2026 to June 2029. These moves suggest that even the largest providers cannot assume their own infrastructure will always be enough. For developers and enterprises, the lesson is clear: AI platform scaling now depends on flexible, cross-cloud capacity planning. For providers, traditional rivalries may matter less than keeping AI services responsive when usage jumps overnight.
