Is Microsoft’s cloud dependence becoming a risk for AI platforms?

Users of Microsoft 365 continued to experience service disruptions early Friday. At 2:14 a.m. UTC on November 13, Microsoft’s service-health dashboard continued to report service degradation for business and consumer products. The issues were not limited to Outlook. Other services, including Microsoft 365 Copilot and the Microsoft connectors utilized in systems like ChatGPT and Claude, were also affected.

Shortly before 1 p.m. ET on Thursday, November 12, Outlook users started to complain about the loss of their connections. Later updates from Microsoft indicated that many of the affected services were dependent on Exchange Online. This explains why Teams chats and online meetings did continue to work, but scheduling meetings did not function properly at all. In addition, files that were saved in SharePoint were available.

The disruption also affected other Microsoft tools that depend on email and calendar data. Copilot Chat could fail to load or respond. Microsoft Graph requests could time out. Microsoft also warned about delayed emails and failures in Safe Links verification.

Other affected services included Defender, Intune, Purview, Planner, To Do, administrative tools, and some Power BI exports. Microsoft also reported issues with certain Education features, Universal Print functions and Windows 365 Cloud PCs.

Why ChatGPT and Claude flashed errors

ChatGPT and Claude have also been affected by the Microsoft 365 disruption through their connectors. Anthropic reported that most email, calendar, and search requests through Claude’s Microsoft 365 connector were not successful. Claude for Outlook was affected as well. OpenAI reported problems with new Microsoft connections and connector tools in ChatGPT.

Neither Anthropic nor OpenAI reported an outage of its core AI models. The failures involved access to Microsoft 365 services. Microsoft 365 Copilot faced a different problem. According to Microsoft’s service bulletin, Copilot Chat itself could fail to load or respond.

What Microsoft said went wrong

Microsoft’s updates showed that the strain came from lost server capacity, not a surge in traffic. Some servers failed automated health checks and dropped out of service. That left the remaining machines handling more requests.

Microsoft engineers found that a core process was overloaded, causing affected servers to be unresponsive. They also suspected a problem with the component handling automated health checks. However, they had not confirmed what triggered the failure.

Emergency fixes, not a permanent cure

Microsoft temporarily suspended deployments and configuration changes while investigating the disruption. Engineers redirected Outlook connections away from affected infrastructure in the United States and redistributed traffic across Microsoft’s global network. They also restarted unresponsive servers and tested a different health-check method to bring them back online, according to Microsoft’s updates.

Some servers returned to service, only to become unresponsive again when traffic resumed. At 2:14 a.m. UTC on November 13, Microsoft’s service-health dashboard still reported service degradation. Microsoft was adding infrastructure capacity as regions came back online and monitoring the results. The company had not confirmed full restoration or announced a permanent fix.

Why one vendor’s bad day matters

A disruption at one cloud provider can affect businesses worldwide. Synergy Research put third-quarter cloud infrastructure spending at $168 billion, up 55% from a year earlier. AWS held 26% of the market, Microsoft 19% and Google 16%—61% combined.

Cloud investment is also accelerating. Gartner expects AI infrastructure spending to reach $2.4 trillion in 2027, accounting for 59% of global AI spending.

The financial risks are substantial. Uptime Institute found that 57% of surveyed operators’ latest major outages cost over $100,000. One in five exceeded $1 million. Third-party providers also accounted for roughly two-thirds of publicly reported outages over nine years.

Switching cloud providers isn’t always easy. The UK’s competition watchdog flagged barriers and recommended examining Microsoft and AWS for strategic market status.

Multicloud networking could help businesses reduce their dependence on one provider. Cryptopolitan previously reported on a service connecting AWS and Google Cloud through private networks. Such connections may improve resilience, but they cannot restore an application that still depends on Microsoft 365.

If you're reading this, you’re already ahead. Stay there with our newsletter.