How AI-Driven Coding Broke GitHub’s Forecasts
GitHub’s infrastructure crisis refers to the wave of outages and performance problems caused when rapid adoption of AI-powered coding agents drove traffic far beyond Microsoft’s original capacity plans for the platform. AI assistants and agentic workflows have turned GitHub from a code repository into a high-intensity compute and automation hub, and the existing architecture could not keep up. GitHub’s own data shows how dramatic the change has been: commits were on pace to reach 14 billion in 2026, up from 1 billion in 2025, a step change that shattered demand forecasts and forced emergency scaling measures. What had been a planned, gradual migration of GitHub’s systems onto Azure suddenly collided with real-time AI-driven growth, creating a gap between Microsoft’s ambitions and the infrastructure it had ready on the ground.
Downtime, GitHub Actions, and the Strain of Agentic Workflows
The immediate impact of the surge showed up as a GitHub infrastructure outage pattern: degraded performance, stuck pull requests, and slow CI pipelines. GitHub’s May 2026 Availability Report listed nine separate incidents that degraded services, following a similar pattern in April. At the same time, GitHub pushed deeper into AI automation with Agentic Workflows in GitHub Actions, which allow coding agents to triage issues, analyze CI failures, update documentation, and maintain repositories from within workflows. These agents execute multi-step, reasoning-heavy tasks that trigger more commits, more builds, and more repository operations than traditional human-only development. Because they run as GitHub Actions, they compound load on an architecture originally designed around human-paced interactions. The combination of agentic coding, Copilot usage, and growing hosted runner images pushed monolith traffic and Git operations toward limits, despite GitHub doubling effective capacity and starting to isolate critical database components.
Why Microsoft Called Its Cloud Rival for Help
The capacity squeeze grew so severe that Microsoft turned to Amazon Web Services for overflow compute to stabilize GitHub, even while publicly championing Azure as its strategic cloud. According to reporting on the plans, GitHub, which historically ran its own data centers, was supposed to be fully on Azure by 2027. But the spike in AI activity made that timetable unrealistic. Instead, Microsoft is adding extra capacity via AWS while still accelerating the Azure migration, turning GitHub into a multi-cloud operation in practice. A Microsoft spokesperson confirmed that GitHub is using multiple cloud providers, though did not name Amazon; Amazon declined to comment on specific customers. For Microsoft, this is more than an awkward Microsoft AWS partnership moment—it is a sign that AI capacity planning assumptions failed, and that even hyperscalers sometimes need help when AI workloads surge faster than their own cloud build-out.
AI Capacity Planning Lessons for the Cloud Era
GitHub’s experience highlights the new difficulty of AI capacity planning. The platform initially expected a 10x capacity increase to be enough, but by February 2026 it saw that a 30x expansion was needed to absorb the flood of pull requests, commits, and new repositories. GitHub’s SVP of software engineering, Jakub Oleksy, said the company is “making structural changes that permanently remove failure modes,” including isolating the primary database from user, authentication, and authorization domains and moving more monolith traffic and Git traffic to Azure. Still, outages persisted, emphasizing how agentic development workflows can outpace even aggressive infrastructure upgrades. The broader lesson is that coding agent adoption and AI-powered automation do not scale in linear fashion, and cloud providers may need multi-cloud strategies, stricter AI usage controls, or more conservative rollout schedules to keep platforms reliable as AI agents do more of the work.
