"In the era of AI, a fatal hidden danger emerges as the shutdown of the Chicago Mercantile Exchange data center reveals underlying concerns in the cooling system."

A trading interruption at the Chicago Mercantile Exchange (CME) brought the cooling issue of data centers into the public eye. On November 27, the trading platform of CME Group, the world's largest futures exchange operator, experienced a several-hour interruption, affecting contracts worth trillions of dollars spanning stocks, foreign exchange, bonds, and commodities. The direct cause of this event was a malfunction in the cooling system of its data center in Aurora, Illinois. The data center is owned by CyrusOne, an operator owned by private equity firms KKR & Co. and Global Infrastructure Partners. CyrusOne stated that a chiller in the facility malfunctioned, affecting multiple cooling units, and this "simple" physical failure caused global market turmoil. To avoid equipment overheating, capital expenditure on cooling systems can account for up to 15% of the total investment in data center projects. This event was not just an isolated technical failure. Against the backdrop of the AI wave that once propelled NVIDIA to the title of the world's most valuable company, the cooling issues of data centers have become more prominent. Where does the heat come from? Data centers are buildings filled with servers, which consist of stacks of chips working together to process and store data. Processing power, often referred to as "compute," has become a critical commodity necessary for training models for AI companies. Data centers profit by renting out computing power to other companies, meaning operators have an incentive to cram as many servers as possible into the same space to maximize capacity. All these servers consume a significant amount of electricity. Due to their high energy consumption and round-the-clock operation, a data center's energy consumption per square foot is 50 times that of a regular office building. Most of the energy they consume is eventually dissipated as waste heat. It's like how a personal laptop or phone heats up when handling complex tasks. Cooling technology and trade-offs Traditionally, servers have been cooled using air, similar to how a home air conditioning system works. Fans blow cold air towards the servers, and then hot air is expelled from the room. However, as data centers for artificial intelligence generate more heat, liquid cooling systems have become increasingly common since around 2022. There are various ways liquid cooling can be used, such as sending cold liquid through pipes to heat sinks in close contact with chips, or immersing entire servers in containers filled with cooling liquid. Some systems use low-boiling liquids, which absorb heat when in contact with high-temperature chips and then evaporate, condensing back into liquid for reuse in a cycle. Compared to air, liquid can carry more heat energy per unit volume, making it more efficient. But these systems are complex to install and costly, and once an issue arises, they can be tricky to deal with - no one wants expensive chips to be submerged in liquid. Whether using air or liquid, once heat is transferred from the chips, it eventually passes to a cooling water circulation system, and the heat is released to the external environment by cooling towers or industrial chiller units. This is why data centers consume a significant amount of water and have raised concerns about exacerbating water resource pressures in water-stressed regions. The cost of overheating Overheating in data centers can lead to data loss, damage expensive chips inside servers, and cause service interruptions for customers. The consequences are similar to recent service interruptions caused by technical failures at several digital infrastructure providers. For example, network security company Cloudflare Inc. experienced a significant network interruption in November last year, resulting in various websites, from social platform X to ChatGPT, being inaccessible. Amazon Web Services, CrowdStrike, and Microsoft have also had similar issues. Typically, data centers invest heavily in redundancy measures, including setting up backup generators, additional cooling units, or even replicating entire facilities to minimize the possibility of interruptions. But as systems become more complex, interruptions may still be hard to avoid even with redundancy measures in place. Replay of the CME event CME's trading platform is located in a data center operated by CyrusOne in a campus in Aurora, a suburb of Chicago. According to CyrusOne, on November 27, a chiller in its Aurora facility malfunctioned, affecting multiple cooling units, ultimately leading to the trading interruption. After the event, CyrusOne stated that while working to restore full cooling capacity, they had deployed temporary cooling equipment to supplement the permanent system. According to information on the company's website, the Aurora campus has "advanced cooling technologies," utilizing air-cooled chillers and natural air or water cooling when temperatures drop below 30 degrees Fahrenheit (approximately -1 degrees Celsius). According to weather reports, the local temperature in Aurora was around 28 degrees Fahrenheit by 10:40 am on November 28. It's worth noting that CyrusOne's website also claims their Aurora facility has additional cooling units to handle failures in the air-cooled chiller system. It's currently unclear whether the redundant system played the expected role in this event. This article was selected from "Wall Street See News", written by Bai Yilong; GMTEight Editor: He Yucheng.