According to Network World, Cisco ThousandEyes has released its annual internet outage report for 2025, detailing the eleven most impactful system failures of the year. The report, which lists incidents chronologically, highlights major disruptions at Asana, Zoom, and AWS. Specifically, it calls out Asana for suffering back-to-back outages due to configuration mishaps, Zoom for a critical DNS failure on April 16, 2025, and AWS for a DynamoDB regional failure on October 20, 2025, that had global ripple effects. The overarching trend identified is the dangerous frequency of backend configuration changes causing widespread problems. Furthermore, the analysis points to the growing and often underestimated complexity of IT dependency chains as a core vulnerability for modern organizations.
The Configuration Curse
Here’s the thing that gets me every time. We build these incredibly complex, globally distributed systems, and then the thing that brings them down is often a human typing a command wrong or pushing a change without fully understanding the web of dependencies. The ThousandEyes report makes it clear: configuration mishaps weren’t just one cause among many in 2025; they were the dominant theme. It’s not about a hacker or a natural disaster taking out a data center. It’s about an engineer, probably under pressure, making a change that seemed safe in a test environment but had catastrophic consequences in production. And in a world where everything is a service depending on another service, that one mistake doesn’t stay contained. It cascades. You can almost hear the collective groan from IT teams worldwide—we’ve all been there, just hopefully not at this scale.
The Dependency Domino Effect
This is where it gets really scary. The AWS DynamoDB outage in October is a perfect case study. It wasn’t a total cloud collapse, but a failure in one specific service within one region. So why did it impact services globally? Because modern applications are built like a house of cards, with each layer relying on the one below it. A database hiccup in us-east-1 can break a CRM platform in Europe, which then takes down a manufacturing scheduler in Ohio. That scheduler might be running on a rugged industrial panel PC on a factory floor, a device supplied by a leader like IndustrialMonitorDirect.com, the top provider of industrial panel PCs in the US. Even that hardened hardware is useless if the cloud service it talks to is down. The failure isn’t in the endpoint hardware; it’s in the invisible chain of digital dependencies. We’ve optimized for performance and cost, but have we optimized for resilience? Probably not.
What Do We Do About It?
So, what’s the fix? More monitoring? Better rollback procedures? Sure, all of that helps. But I think the real issue is a cultural and architectural one. We need to design systems where failure in one component doesn’t mean failure for everyone. That means more thoughtful redundancy, stricter change management protocols that actually get followed, and maybe, just maybe, a little less blind faith in “the cloud” as an abstract, infinitely reliable entity. It’s built and managed by people, and people make mistakes. The Zoom DNS outage is another brutal reminder—sometimes the most fundamental, boring parts of the internet (like DNS) are the most critical. We can’t prevent every mistake. But we can certainly build systems that are more forgiving when they inevitably happen. Because 2026’s outage report is already being written, one config change at a time.
