On May 7 at 4:20 PM PDT, a single data center in AWS’s us-east-1 region experienced a thermal event. Cooling capacity dropped. Temperatures spiked. Servers tripped offline to protect hardware. By evening, EC2 instances and EBS volumes in the use1-az4 Availability Zone were impaired. Services that depend on them—load balancers, Kubernetes, Redis, Redshift, OpenSearch, Kafka—all saw elevated error rates and latency. Some workflows stopped entirely.
AWS began shifting traffic away from the affected zone by 5:06 PM PDT, limiting blast radius for most services. But dependent systems still took hits. Coinbase lost trading for seven hours. FanDuel’s sports betting froze during NBA playoffs. KoboToolbox, a humanitarian data platform, went dark globally while its European instance stayed up. The common thread: they all relied on resources inside one physical room in Northern Virginia.
The recovery was slow. AWS brought cooling capacity back incrementally, restoring racks one by one to avoid reigniting heat buildup. Full recovery took into May 8. The company’s status page offered transparent updates, even when progress lagged. “Slower than originally anticipated,” one note read. By early afternoon, most resources were back. A few lingered.
This isn’t new. US-EAST-1 has a track record. Outages in 2021, 2023, and 2025 all traced to the same geography. Each time, engineers promise better redundancy. Each time, failover architectures reveal hidden assumptions. Multi-AZ setups protected many customers this week. Those who tied critical paths too tightly to a single zone paid the price. Coinbase’s CEO called the outage unacceptable and signaled a review of tradeoffs between speed and resilience.
The physical reality is often overlooked. Servers don’t fail gracefully when ambient temperature exceeds design thresholds. They trip offline to protect silicon. Restoration requires careful sequencing. Bring one rack up too soon and heat builds again. The process demands patience.
Cloud buyers already pay premiums for multi-region architectures. After this week, those premiums may feel less like insurance and more like table stakes. The internet runs on a handful of physical locations, no matter how virtual the marketing makes it sound.
Source: Webpronews