On Monday, October 20th, I couldn't open my VDI for work. It only lasted about 20 minutes, but at the same time, I noticed that Starbucks mobile ordering and Toast — a popular restaurant ordering platform — were both down for several hours.
Amazon Web Services (AWS) has roughly 40 percent of the cloud infrastructure market, so when a major outage happens, it naturally cascades across hundreds of companies and millions of users. But it made me wonder: whose fault is it when something like this happens — AWS or the developers who build on top of it?
The outage itself was isolated to AWS's US-EAST-1 region (Northern Virginia), yet so many applications across the country went offline. Shouldn't there be failovers or backup systems to handle that? What actually failed was AWS's internal DNS system, which connects core services like DynamoDB, EC2, and CloudWatch. When those DNS lookups broke, any application trying to reach those services simply couldn't — causing a ripple effect that spread far beyond one data center.
The reality is that while multi-region redundancy and backups are possible, they're also expensive. Running mirrored infrastructure in multiple data centers can easily increase monthly cloud costs by 50–200%, depending on the service and amount of data replication. For a smaller SaaS company spending $10,000 a month on AWS, that could mean $15,000–$25,000 or more — often not worth it for protection against a once-a-year outage.
For most businesses, the cost of full redundancy isn't worth the potential risk of a few hours of downtime once or twice a year. Instead, outages like these become a built-in part of the risk model — an accepted trade-off between uptime and operating cost.
My own VDI likely runs on a high-compute, multi-region setup, possibly through a provider like Citrix, which distributes workloads across multiple clouds or geographic zones. When routing failed for those 20 minutes, it probably just switched over to another data center automatically.
That's the difference between my VDI and many consumer apps: Citrix and similar enterprise systems are designed with multi-region active-active deployments, meaning they can keep running even if one region fails. But smaller companies can still hedge with cheaper strategies like multi-AZ deployments, caching, queued writes, or static fallback pages — ways to stay partially functional without doubling their infrastructure bill.
AWS outages like this aren't new. The 2017 S3 outage, the 2021 Kinesis failure, and the 2023 EC2 disruption all originated in the same US-EAST-1 region. Each time, it shows how much of the internet depends on a single point of failure.
Sometimes, it's not that companies don't know how to prevent downtime — it's that they simply can't justify the cost of being invincible.