Two Azure outages hit 18 regions in 40 hours, including six hours of Azure OpenAI failures

Microsoft's cloud had a rough week. Two separate Azure incidents across a roughly 40-hour window from September 29 to October 1 affected 18 regions. The most visible damage for AI teams was nearly six hours of Azure OpenAI API failures in Sweden, a region many European customers use to keep data in the EU.

The sources give few root-cause details. However, the scale of 18 regions across two incidents points to shared control-plane or networking dependencies rather than isolated hardware faults. For AI workloads the effect compounds. Applications pinned to a single region's model deployment had no fallback, and agent pipelines that depend on continuous API calls stalled.

The outages arrive at a sensitive regulatory moment. Bloomberg reports that EU regulators plan Digital Markets Act decisions in November that could designate Azure and AWS as gatekeepers. That would impose interoperability and anti-self-preferencing obligations. Visible reliability failures strengthen the case for portability rules that make switching providers easier.

Engineers' takeaway in community threads was blunt: single-region AI deployments are a liability, and multi-region failover for Azure OpenAI should be standard. Separately, Microsoft enabled consumption billing by default for Copilot Business, which has IT admins worried about runaway costs. Combined with outages, that adds operational risk for Microsoft customers. Things to watch:

- Microsoft's post-incident reviews

- whether any service credits are issued

- the scope of the November DMA decisions