2025: A Year of Cloud Growing Pains
2025 was a pivotal year for cloud services. AI workloads pushed infrastructure to its limits, new services scaled faster than their reliability engineering could keep up, and several high-profile outages reminded us that no service is truly invulnerable.
Here's our roundup of the most significant cloud outages of 2025 and what each one teaches us.
1. The CrowdStrike-Induced Windows Outage (July 2024 Aftermath)
While the CrowdStrike incident technically started in July 2024, its aftershocks continued well into 2025 as organizations overhauled their update processes.
What happened: A faulty CrowdStrike Falcon sensor update caused millions of Windows machines to blue-screen simultaneously.
Impact: Airlines, hospitals, banks, and government agencies worldwide were affected.
Lesson: Third-party security agents with kernel-level access are a single point of failure. Staged rollouts and canary deployments aren't just for your own code — they need to apply to every piece of software running on your infrastructure.
2. Azure Active Directory Outages
Azure AD (now Entra ID) experienced multiple authentication outages in 2025, each affecting the entire Microsoft 365 ecosystem.
What happened: Authentication infrastructure issues prevented users from signing into Teams, Outlook, SharePoint, and other Microsoft services.
Impact: Millions of enterprise users unable to access core productivity tools during business hours.
Duration: Individual incidents lasted 2-6 hours.
Lesson: Authentication is the ultimate single point of failure. When your auth provider goes down, everything behind it becomes inaccessible. Organizations should consider emergency break-glass access procedures that don't depend on their primary identity provider.
3. GitHub Actions Degradation
GitHub Actions experienced recurring periods of degraded performance throughout 2025, with queue times spiking during peak hours.
What happened: Growing demand for CI/CD compute exceeded capacity, causing jobs to queue for extended periods.
Impact: Development teams unable to merge and deploy code on schedule. Some organizations saw build times increase from minutes to hours.
Lesson: CI/CD pipelines are critical infrastructure, not just developer convenience. Have a backup plan — self-hosted runners, alternative CI systems, or the ability to deploy manually when automated pipelines are slow.
4. Cloudflare Edge Incidents
Cloudflare experienced several edge network incidents that affected thousands of websites and APIs simultaneously.
What happened: Configuration changes and software updates caused routing issues at edge locations.
Impact: Websites returning 500 errors, API calls failing, DNS resolution delays.
Lesson: CDN and edge providers are "blast radius multipliers" — when they have issues, the number of affected downstream services is enormous. If you use Cloudflare (or any CDN) for critical paths, have a DNS failover strategy.
5. OpenAI API Capacity Issues
2025 saw explosive growth in AI API usage, and OpenAI's infrastructure struggled to keep pace.
What happened: Recurring capacity constraints led to elevated error rates, slow response times, and rate limiting during peak hours.
Impact: Applications built on OpenAI APIs experienced degraded functionality or complete feature failures.
Duration: Intermittent issues spanning hours to days during peak demand periods.
Lesson: AI services are still maturing from a reliability perspective. If your application depends on an AI API, build graceful degradation — cached responses, fallback models, or the ability to function (even with reduced features) when the API is unavailable.
6. AWS us-east-1 Events
AWS's largest and oldest region continued to be a source of notable incidents in 2025.
What happened: Various service-specific issues in us-east-1 affecting EC2, S3, and Lambda.
Impact: Applications running exclusively in us-east-1 experienced downtime. Multi-region applications were largely unaffected.
Lesson: us-east-1 is the most feature-complete but also the busiest and most incident-prone AWS region. Critical workloads should be multi-region, and global services (like IAM and Route 53) that only run in us-east-1 need special attention in your architecture.
Common Themes Across 2025 Outages
1. Scale Is the Enemy of Stability
The fastest-growing services (AI APIs, identity providers) had the most incidents. Growth outpaces reliability engineering.
2. Cascading Failures Are the Norm
Most major outages in 2025 weren't isolated — they cascaded across dependent services. An auth outage becomes a productivity outage becomes a revenue outage.
3. Status Pages Remain Delayed
The gap between when users notice issues and when status pages acknowledge them remains 15-30 minutes on average. Independent monitoring is essential.
4. Communication Quality Varies Wildly
Some vendors provided excellent real-time updates and thorough post-incident reports. Others were opaque, slow, and vague.
How to Protect Yourself in 2026
Monitor your services | View current status of 600+ services