Coinbase reviews the May outage incident: AWS cascading failure exposes architectural risks

Coinbase released a retrospective report on the large-scale service interruption event on May 7, 2026.

The outage lasted approximately 8 hours, with full recovery taking about 12 hours. During this time, trading, deposits, withdrawals, and most core services were unavailable or severely degraded. Coinbase stated that the outage was caused by multiple cooling units failing simultaneously in the cooling system of a data center in one availability zone (use1-az4) in the AWS us-east-1 region, triggering cabinet thermal protection shutdowns, which led to EC2 instances and EBS volumes going offline, affecting multiple internet services.

During the recovery process, the Coinbase trading matching engine lost quorum due to the cluster architecture deployed in a single AWS data center losing most nodes. It required urgent code adjustments and the reconstruction of a new node group to restore operation, gradually restarting market trading during the recovery.

Additionally, the AWS-managed Kafka (MSK) service experienced control plane failures, preventing the automatic re-election of partition leaders, further blocking quotes, fees, and some settlement and data flow systems, which expanded the overall impact.

After manual partition migration in collaboration with the AWS engineering team, the system gradually returned to normal. Coinbase stated that this incident exposed its shortcomings in cross-availability zone automatic switching capabilities and disaster recovery for managed middleware. The company will upgrade its cross-region hot backup architecture, strengthen regular failure drills, and migrate the Kafka system from dual availability zones to a three availability zone deployment, while also working with AWS to advance root cause fixes and improvements.