Microsoft Taps AWS as GitHub AI Agents Break SLAs

In an unprecedented move to stabilize its platform under a relentless deluge of AI coding agent traffic, Microsoft has quietly routed core GitHub operations through rival Amazon Web Services (AWS), following a series of crippling outages that saw availability dip below 99% in June and nine incidents in May alone. The root cause traces back to GitHub's aging 2008 Ruby on Rails monolith, where a single failure—such as February's authentication database crash—can cascade to disable Actions, Copilot, and the UI simultaneously. With Azure migration stalled at just 40% (target: May 2026), AWS provides a critical buffer—but a class-action lawsuit now alleges Microsoft misled investors about Azure capacity and Copilot adoption.

AI Agents Expose the Rails Monolith’s Breaking Point

The surge in traffic stems from AI agents like Cursor and Claude Code, which operate 24/7 via API, shattering human-paced traffic models. AI-generated pull requests hit 17 million in March—a 325% surge over prior levels. GitHub is now targeting a 30x capacity increase and architectural decoupling, including refactoring critical Ruby components into Go and implementing load-shedding mechanisms. However, the legacy monolith remains a single point of failure, as demonstrated by February's auth DB outage that simultaneously took down Actions, Copilot, and the UI.

AWS as a Crutch: Azure Migration Only 40% Complete

Microsoft’s decision to route GitHub traffic through AWS—its direct cloud competitor—signals that Azure cannot yet handle its own flagship developer platform under AI workloads. The Azure migration, originally planned to complete by May 2026, has only reached 40% completion. AWS acts as a temporary failsafe, absorbing the excess traffic that GitHub’s own infrastructure cannot yet sustain. This move underscores a deeper vulnerability: even for Microsoft, scaling AI-era traffic on Azure remains a work in progress.

Class-Action Lawsuit Alleges Misleading Azure and Copilot Claims

A class-action lawsuit now accuses Microsoft of misleading investors about both Azure capacity and Copilot adoption rates. The legal challenge arrives as GitHub’s reliability crisis becomes public, with the platform failing to meet enterprise SLAs for two consecutive months. The situation serves as a harbinger for all AI-infrastructure platforms: continuous, machine-speed traffic demands fundamental architectural redesign, not just more compute. GitHub’s 30x capacity target and language migration (Ruby→Go) are reactive fixes—but the deeper lesson is that legacy architectures cannot be patched to handle AI agents’ relentless API calls.