How AI Turned GitHub into Its Own Biggest Stress Test
GitHub’s AI adoption outpaced its infrastructure planning when agentic development and AI coding assistants created more code, more automation, and more traffic than its systems and Azure migration roadmap were designed to handle, triggering repeated downtime and forcing Microsoft to seek extra capacity from Amazon Web Services. The platform’s own reports show a growing strain: GitHub logged nine incidents causing degraded performance in May alone, even after expanding its Azure footprint and making “structural changes that permanently remove failure modes,” as SVP Jakub Oleksy put it. AI-driven tools such as Copilot and coding agents have changed usage patterns from occasional human commits to continuous machine-driven activity. Each AI-assisted workflow adds commits, pull requests, and CI jobs, multiplying load on storage, databases, and network layers. This is no longer a simple hosting service; it is an AI operations hub, and GitHub’s legacy architecture has struggled to keep up.
From 1 Billion to 14 Billion Commits: Demand Outruns Design
The scale of GitHub’s growth explains why its infrastructure buckled. Business Insider cited GitHub COO Kyle Daigle stating that commits were on pace for 14 billion in 2026, up from 1 billion in 2025. GitHub’s own availability reporting aligns with that surge: The Register notes that the platform went from handling about 1 billion commits in all of last year to 1.4 billion commits every month. This explosion is closely tied to GitHub Copilot demand and the rise of agentic development, where AI agents read repos, plan tasks, modify code, run tests, and open pull requests. Each step multiplies database calls and storage writes. GitHub had originally planned a 10x capacity increase in late 2025, but by February 2026 it concluded that a 30x expansion was needed. Those numbers show AI infrastructure strain outstripping even aggressive scaling assumptions.
Azure’s Limits and the Awkward Call to AWS
Despite public messaging about Azure’s readiness for large AI workloads, GitHub’s experience exposes Microsoft Azure capacity issues in practice. GitHub has gradually shifted its monolith and Git traffic onto Azure, reaching 40 percent of monolith traffic and 30 percent of Git traffic there, while repository replication hit 99 percent. Yet outages persisted as AI coding activity spiked. According to Business Insider, Microsoft is now adding extra GitHub computing capacity through Amazon Web Services to handle this overflow. A Microsoft spokesperson confirmed that GitHub is “tapping multiple cloud providers” to get the elasticity it needs, even as it accelerates the move to Azure. Operationally, this multi-cloud detour signals that Azure alone could not absorb the sudden AI load fast enough, contradicting the clean, one-cloud narrative Microsoft has promoted around its own infrastructure and AI offerings.
Multi-Cloud as a Pressure Valve for AI-Heavy GitHub
Microsoft’s decision to bring AWS into the mix shows how multi-cloud has shifted from a theoretical strategy to an operational necessity. Startup Fortune describes the move as a “practical answer”: keep pushing GitHub toward Azure, while using multiple cloud providers for elasticity where the AI load hits hardest. This arrangement helps GitHub contain GitHub downtime AWS risks by spreading traffic and compute across more capacity than Azure alone can currently provide. At the same time, GitHub is restructuring its own systems, isolating its primary database cluster by splitting users, authentication, and authorization into separate domains to reduce cascading failures. Together, these steps show that AI infrastructure strain is not only about more servers; it is about architectural resilience. For developers, the goal is simple: keep repositories, CI pipelines, and Copilot features available, no matter which cloud is underneath.
What the Crisis Means for AI Product Rollouts
GitHub’s turbulence is a cautionary tale for any company racing to ship AI features. The platform moved from autocomplete-style suggestions to full coding agents integrated with GitHub, GitHub Mobile, and Visual Studio Code, as reported by Startup Fortune. That shift changed both the intensity and pattern of traffic, but infrastructure planning lagged behind product ambition. The temporary halt on new Copilot subscriptions and the switch to usage-based GitHub AI Credits show that the economics and capacity planning are closely linked. AI products can no longer be treated as thin layers atop existing platforms; they reshape core workloads. For Microsoft, the episode undercuts simple claims of limitless Azure scale and shows that even hyperscale providers can be caught off guard when AI demand grows faster than planned. For the wider industry, it is a warning to pace AI rollouts with realistic infrastructure and cost models.
