On May 7, Coinbase placed its markets into Cancel Only mode because an AWS availability zone in Northern Virginia overheated. EC2 instances degraded, EBS volumes impaired, and a crypto exchange whose entire value proposition is 24/7 trustless markets couldn't let customers trade for roughly six hours. Trading didn't fully resume until 00:49 PDT on May 8.
The AWS outage was real. AWS confirmed rising temperatures knocked out hardware in use1-az4, and CME got caught in the same blast radius. But the interesting question isn't whether AWS has cooling problems. It's why Coinbase, at this scale, with this uptime promise, was single-region dependent in a way that made one overheated Virginia data center the difference between open markets and an auction queue.
This landed days after Coinbase reported Q1 2026 revenue of $1.41 billion against a $1.52 billion expectation, a net loss of $394 million, and transaction revenue $50 million short of estimates. Gergely Orosz noted the particular irony: CEO Brian Armstrong had made public comments about non-technical teams shipping to production. The ops story has a way of arriving right on time. The number that matters isn't in the headline. It's in the incident timeline.