Every time you trigger a Cursor or Claude Code agent session today, it commits code. Sometimes several commits. Sometimes dozens. GitHub COO Kyle Daigle confirmed in April the platform is processing 275 million commits a week — on track for 14 billion this year, against roughly 1 billion in all of 2025. A 14× increase. Not from more developers. From machines operating at machine speed.

Nine service incidents in May. June availability below 99.9%, failing the enterprise SLA threshold GitHub's own CTO acknowledged they missed in February and March. On June 16, Microsoft confirmed it is routing GitHub overflow traffic to Amazon Web Services. Not Azure. AWS — its direct competitor in enterprise cloud.

That sentence deserves a beat. Microsoft and Amazon compete directly for enterprise cloud contracts worth hundreds of billions annually. GitHub is Microsoft's most strategically important developer platform. Routing it through a competitor's infrastructure, however quietly, is an acknowledgment: Azure's current expansion timeline cannot absorb what AI-assisted coding has done to the platform.

What actually happened

The root cause is straightforward. AI agent pull requests went from roughly 4 million per month in September 2025 to more than 17 million in March 2026. Agents don't read documentation, don't have coffee, and don't wait for CI to finish before opening another PR. They operate at throughput levels that GitHub's infrastructure was simply not designed for.

The infrastructure team had planned for a 10× capacity expansion back in October 2025. By February 2026, they revised that to 30×. The platform they needed to build turned out to be three times larger than the platform they'd been planning to build, on a timeline that had already started. Azure's migration — GitHub was running on its own legacy infrastructure and moving to Azure — was in progress, but not fast enough. As of the May 2026 availability report, 40% of monolith traffic was on Azure, up from 8% in February. Git traffic at 30%.

So: still mostly on legacy, not yet on Azure, and the load grew faster than either migration. AWS is the pressure valve while both work gets done.

Source spread

• TechTimes — GitHub's AI Agent Crisis Forces Microsoft to Tap AWS — builder. Confirmed the June 16 routing announcement; has the capacity planning revision (10× → 30×) numbers.
• Quasa — GitHub's AI Agent Tsunami: 275 Million Commits a Week — builder. Kyle Daigle's April confirmation of the commit rate and 2026 projection.
• byteiota — GitHub Is Running on AWS Now — skeptic. Good on the structural irony; frames this as a Microsoft cloud credibility problem.
• Let's Data Science — GitHub Capacity Surge Pushes Microsoft to AWS — builder. Notes that GitHub Actions runners and Codespaces are the specific workloads routing to AWS.

Pros & cons

What's holding up:

• The AWS integration is reportedly transparent to end users. Your Actions workflows, Codespaces environments, and clone operations run wherever there's capacity — you don't select it.
• Microsoft and GitHub are making real progress on the Azure migration. Going from 8% to 40% of monolith traffic in four months is not nothing.
• The nine May incidents were service degradations, not full outages. The platform stayed functional; the SLA just missed.
• GitHub's underlying platform — auth, repo storage, the API — has remained stable. The strain is hitting compute-intensive workloads (Actions, Codespaces) first.

What's still broken:

• "Below the 99.9% enterprise SLA" is a contractual breach for thousands of paid customers. That's not a marketing problem; it's a billing problem.
• Using a competitor's infrastructure as a load balancer is not a strategy. It's a workaround. Microsoft is not building long-term competitive advantage by running GitHub's overflow compute on AWS.
• The 30× capacity figure is an estimate made in February. By June, the assumption may already be wrong again. AI agent adoption is not linear.
• Cursor's acquisition by SpaceX closes around July 12. Grok V9-Medium, trained on Cursor workflow data, is expected to land after. More agents, more commits, more load — and the infrastructure is not finished being built.

What builders need to know

• CI/CD reliability has actually dropped. Nine incidents in May; June availability below the 99.9% enterprise SLA. Build retry logic for GitHub Actions workflows as if occasional failures are expected, not exceptional.
• Actions and Codespaces are the affected workloads. These are the ones routing to AWS for burst capacity. Core GitHub operations (API, auth, clone, push) are on the more stable infrastructure.
• If you're on GitHub Enterprise, your SLA breach is contractually meaningful. Check GitHub's status page history and review your service agreement terms. Credits may apply.
• The Cursor acquisition closes ~July 12. Grok V9-Medium follows. Both events will push more AI-agent commits onto an infrastructure that is already at capacity. Watch for another reliability inflection in late July.
• Evaluate GitHub Actions alternatives for workloads where reliability is hard-constraint. Buildkite, CircleCI, and self-hosted runners on your own compute give you independence from GitHub's current capacity situation.

Further reading

• TechTimes — GitHub's AI Agent Crisis Forces Microsoft to Tap AWS as Outages Break Enterprise SLAs
• Quasa — GitHub's AI Agent Tsunami: 275 Million Commits a Week, 14 Billion Projected for 2026
• byteiota — GitHub Is Running on AWS Now. Here's Why AI Coding Broke the Platform.
• AI Weekly — Microsoft Taps AWS to Keep GitHub Running Amid AI Surge
• Let's Data Science — GitHub Capacity Surge Pushes Microsoft to AWS