AWS experiences an overheating outage in northern Virginia, causing disruptions for Coinbase.

      A single data centre's cooling system fell short, leading AWS to redirect traffic from the impacted zone and caution that the full restoration of remaining services would take longer than anticipated.

      On Thursday, Amazon Web Services announced that one of its data centres located in northern Virginia was overheating, which affected customer workloads. Engineers were still in the process of bringing the site back online while most users were asleep.

      The issue was straightforward: elevated temperatures within a single data centre, caused by a shortfall in the cooling system, compelled AWS to throttle and partially reroute traffic from the affected Availability Zone. According to the company, additional cooling capacity began to be implemented a few hours after the initial impact reports, and “early signs of recovery” surfaced shortly thereafter.

      However, a later update was less reassuring: the time required to implement sufficient additional cooling for safely restarting the remaining systems was longer than anticipated, and AWS did not commit to a timeline for full restoration.

      Coinbase confirmed that its trading platform issues were a result of the AWS incident. After several hours of market degradation, the exchange stated that all markets had been reactivated, and trading returned to normal.

      CME Group, the largest derivatives marketplace in the world, also encountered problems with its CME Direct platform during the same timeframe, although it only referred to the cause as “essential maintenance” without clarifying if the AWS incident was involved. Both companies declined to comment further outside of business hours.

      In AWS terminology, the northern Virginia cluster, known as US-East-1, is the company’s oldest, busiest, and most densely populated region. An Availability Zone in that area comprises one or more physical data centres designed to operate independently. AWS's official advice during the recovery process was the standard recommendation: customers operating in the affected zone should switch to another zone. This is effective for engineering teams that have prepared for it, but less so for those that have not.

      This pattern is becoming increasingly familiar. AWS experienced a much larger outage last October when a DNS resolution failure in DynamoDB cascaded across over a hundred services, taking platforms such as Snapchat, Reddit, United Airlines, and Coinbase offline. This incident lasted roughly fourteen hours and was the most significant internet-wide disruption since the CrowdStrike software malfunction of 2024.

      About a month later, CME dealt with one of its longest trading outages in years, which was linked to a cooling failure at a CyrusOne data centre in the Chicago area.

      The recurrence of these issues is significant. Cooling failures, configuration errors, and DNS misfires are different technical problems, but they lead to the same result: a single physical or logical site becomes the bottleneck for a disproportionate share of public-facing traffic. The northern Virginia region bears that burden more due to historical circumstances than through design.

      AWS launched the region in 2006, and US-East-1 has since accumulated workloads, regulatory dependencies, and customer inertia. The hyperscalers are investing tens of billions to expand other regions, but customer concentration in US-East-1 is unlikely to shift quickly.

      Coinbase’s reliance on the cloud is part of a larger trend. The Cloudflare-driven outage that affected Coinbase and other exchanges in 2019 involved a different type of failure but conveyed a similar lesson, prompting crypto exchanges to invest years in building for multi-region failover.

      Thursday’s incident illustrates that, despite these efforts, a single cooling system failure can still impact a market meant to operate continuously.

      CME’s situation is more complex. Derivatives markets rely on intricate margin and clearing processes that do not degrade gracefully; an outage during peak Asian hours, such as Thursday's, disrupts clearing-cycle deadlines that involve transferring funds the following morning.

      Whether the CME issue was directly linked to the AWS incident will influence regulatory discussions about trading resilience.

      AWS has not provided an estimate regarding the number of affected workloads, and Amazon has yet to clarify what caused the cooling system to fail, whether it was due to equipment, environmental conditions, or a combination of factors.

      Over the past year, the northern Virginia region has been adapting to a surge in new AI-training and inference capacity, which tends to run hotter and more densely than traditional cloud workloads; whether this is incidentally relevant to Thursday’s failure or a key factor is a question the post-incident report will need to address.

      For most customers, the solution is what AWS recommended in its initial update: avoid operating everything within a single Availability Zone in a single region. This recommendation has been on AWS’s architecture best-practice page for years. Each failure of this nature increases the cost of neglecting that advice.