A cooling failure at a data center in Aurora shut down CME trading for hours

When a cooling failure inside a data center in Aurora, Illinois, stopped futures and options trading at the Chicago Mercantile Exchange, the disruption exposed how one technical fault can hit global markets at the same time.

The outage began on November 27 inside a facility run by CyrusOne, whose cooling equipment broke down and pushed temperatures high enough to shut servers handling contracts tied to equities, foreign exchange, bonds, and commodities.

The stoppage lasted hours and froze activity linked to trillions of dollars, according to Bloomberg. It showed how markets that run nonstop can still be knocked offline by overheating machines inside a single building.

The facility that supports the CME platform is part of a campus that KKR & Co. and Global Infrastructure Partners bought in 2022. CyrusOne said “a chiller plant failure” hit several cooling units at the same time, which caused the servers to power down for safety.

Once the shutdown began, CyrusOne brought in temporary cooling equipment to stabilize temperatures while its staff worked on getting the main systems online again. The company says its Aurora site uses air-cooled chillers and cold outdoor air when temperatures fall below 30°F, and weather logs showed 28°F in the area that morning.

CyrusOne states on its website that the facility includes extra cooling units designed to protect against these failures, but it is not clear whether any of those backups did anything during this incident.

Cooling breakdowns trigger outages across crypto-linked infrastructure

The CME shutdown came at a moment when data centers are becoming more important to markets, crypto exchanges, AI developers, and cloud providers.

These buildings hold rows of servers packed with chips that store and process huge amounts of information. Their compute power runs everything from market feeds to blockchain analytics to cloud wallets.

These servers use so much electricity that a data center can burn 50 times more energy per square foot than an office building, and most of that power turns into waste heat that must be removed to avoid failures.

Operators try to fit as many servers as possible into each room so they can lease more compute to customers, but that also creates more heat.

Cooling these sites has turned into a large part of the total cost. Some projects spend up to 15% of their budget on cooling alone. Older sites relied on cold air. Fans pushed chilled air across the racks, and warm exhaust was pushed outside.

But once AI workloads grew heavier around 2022, companies started shifting to liquid cooling. Cold liquid runs through plates under the chips or fills tanks where full servers are submerged. Some systems use special liquids with low boiling points that evaporate when they touch the hot chips, then condense back into fluid.

Liquid can absorb more heat than air, but it adds risk. A leak can damage expensive hardware and take systems offline.

Once the heat is collected, the liquid or air moves it to a chilled water loop that carries it to a cooling tower or industrial chiller. These towers also use large amounts of water because a portion evaporates during the cooling process. Many communities have raised concerns about heavy water use, especially in regions already dealing with shortages.

Overheating events shut down platforms and stall trading systems

When a data center overheats, chips can become damaged, data can be lost, and services can fail. Outages in November hit Cloudflare, knocking out sites like X and ChatGPT.

Separate issues at Amazon Web Services, Microsoft, and CrowdStrike shut parts of the internet down for hours at a time. These events show how much of daily life depends on a few companies running always-on infrastructure.

Operators build in redundancy with backup generators, extra cooling units, and even full duplicate campuses. But the CME outage shows that even these layers can fail when several parts break at once.

One malfunctioning chiller pushed a major market operator offline long enough to affect contracts worldwide, and even with extra cooling equipment on-site, it is still unknown whether any redundancy helped at all.

Join a premium crypto trading community free for 30 days - normally $100/mo.