Microsoft is offloading GitHub traffic to AWS. Its own competitor. Because it cannot keep up.
AI coding agents generated such a surge of traffic that GitHub stopped holding its SLA — 9 degraded-service incidents in May alone. The owner of Azure renting capacity from AWS is a rare public admission of strain.
Same month: Azure OpenAI went down for ~8 hours — an innocuous change → a retry storm → a shared load balancer giving out → a cascade across Europe and Australia.
The verdict for team leads: if your CI/CD hangs off github.com, treat degradation as the baseline, not the exception. Retries with jitter, circuit breakers, a fallback to self-hosted runners. Right now even the hyperscalers are not made of granite.