AWS said the problem started after temperatures rose inside a single data centre, blaming a cooling system failure while explaining that engineers brought extra cooling capacity online as recovery work continued.

The outage affected services in one Availability Zone, which is a group of connected data centres designed to operate separately within an AWS region. AWS said it redirected traffic away from the affected zone for most services to reduce disruption.

Later in the day, recovery was taking longer than expected because more cooling capacity was still needed before remaining systems could safely return online. AWS further added that it did not yet have a timeline for full recovery.

Coinbase confirmed its trading platform issues were directly linked to the AWS outage. Users reported problems accessing services and delays during trading, although the company later said all markets had been restored and trading resumed normally.

Meanwhile, CME Group reported login and latency problems on its CME Direct trading platform. In a notice to users, the exchange said it had completed “essential maintenance work” and confirmed customers could log back in. The company did not explain the cause of the disruption.

Neither AWS nor CME immediately responded to requests for additional comment outside normal business hours.

The incident again exposed how heavily large financial platforms depend on cloud providers such as AWS. Even a problem in a single data centre zone can spread quickly across trading services, apps and online platforms used worldwide.

AWS has faced similar problems before. In October last year, a major outage disrupted thousands of websites and apps, including Snapchat and Reddit. That disruption became one of the largest internet outages in recent years.

A month later, CME Group experienced another major interruption after cooling systems failed at a data centre operated by CyrusOne in the Chicago area. Trading across stocks, bonds, commodities and currencies stopped for several hours.

The latest outage also points to stress on data centres as companies expand artificial intelligence systems and use more high-density servers, which generate far more heat and demand stronger cooling systems.

AWS has advised customers to spread workloads across multiple Availability Zones and regions to reduce the risk of large-scale disruptions when outages happen.

The newspaper said the incidents resulted from errors involving Amazon’s own tool, known as Kiro. In one case in mid-December, AWS customers experienced a 13-hour interruption.

Engineers had allowed the tool to carry out certain system changes. It then decided to “delete and recreate the environment”, the report said, which led to the disruption.

AWS disputed that account.

In an emailed response to Reuters, a company spokesperson said the disruption was brief and blamed it on user error. “This brief event was the result of user error-specifically misconfigured access controls, not AI.”

The spokesperson added that the interruption was “an extremely limited event” affecting a single service in one of AWS’s two mainland China regions. It did not impact compute, storage, database, AI technologies, or any other AWS services, the company said.

The December incidents follow an outage in October that disrupted Amazon’s cloud operations globally. That earlier failure affected Amazon’s own services and several high-profile apps, including Reddit, Roblox and Snapchat.

AWS is the cloud division of Amazon and supports a large share of the internet’s infrastructure. Because of that reach, even short interruptions can affect millions of users and businesses.

Both Amazon Web Services outages in December have drawn attention because they involved automation tools that can act with limited human input.

Cloud providers have been expanding the use of such systems to manage complex infrastructure. At the same time, customers expect stability and clear accountability when problems occur.

Competitors including Microsoft Azure and Google Cloud are also developing automated tools to manage their platforms.

AWS maintains that the December disruption resulted from misconfigured access controls, not from the coding tool acting on its own.