Liquid cooling in data centers is key to meeting AI-driven heat demands
June 24, 2026
Select your region to visit a Stantec regional site, or visit our Locations Hub on our Global site where you can learn more about Stantec in your local area.
", "languageSelectorImage": { "imageLarge": "\/content\/dam\/stantec\/images\/Icons\/Map.svg" }, "globalRegion": { "name": "Global", "regionTag": "localization:global", "description": "", "languages": [ { "langCode": "en", "name": "English", "pathLink": { "url": "en" }, "active": "true" }, { "langCode": "fr", "name": "Français", "pathLink": { "url": "fr" }, "active": "false" } ] }, "regionLanguages": [ { "regionTag": "localization:china", "name": "China", "description": "", "languages": [ { "langCode": "en", "name": "English", "pathLink": { "url": "cn" }, "active": "false" }, { "langCode": "en", "name": "", "pathLink": { "url": "en" }, "active": "false" } ] }, { "regionTag": "localization:italy", "name": "Italy", "description": "", "languages": [ { "langCode": "it", "name": "Italiano", "pathLink": { "url": "it" }, "active": "false" }, { "langCode": "en", "name": "", "pathLink": { "url": "en" }, "active": "false" } ] }, { "regionTag": "localization:latin\u002Damerica", "name": "Latin America", "description": "", "languages": [ { "langCode": "es", "name": "Español", "pathLink": { "url": "es" }, "active": "false" }, { "langCode": "en", "name": "", "pathLink": { "url": "en" }, "active": "false" } ] }, { "regionTag": "localization:netherlands", "name": "Netherlands", "description": "", "languages": [ { "langCode": "nl", "name": "Nederlands", "pathLink": { "url": "nl" }, "active": "false" }, { "langCode": "en", "name": "", "pathLink": { "url": "en" }, "active": "false" } ] }, { "regionTag": "localization:taiwan", "name": "Taiwan", "description": "", "languages": [ { "langCode": "zh_tw", "name": "繁體中文", "pathLink": { "url": "zh_tw" }, "active": "false" }, { "langCode": "en", "name": "", "pathLink": { "url": "en" }, "active": "false" } ] }, { "regionTag": "localization:turkey", "name": "Turkiye", "description": "", "languages": [ { "langCode": "tr", "name": "Türkçe", "pathLink": { "url": "tr" }, "active": "false" }, { "langCode": "en", "name": "", "pathLink": { "url": "en" }, "active": "false" } ] } ], "originPath": "/content/stantec/en/ideas/content/blog/2026/liquid-cooling-data-centers-key-to-meeting-ai-driven-heat-demands.html" }, "locationSelector": { "locationInformationText": "We\x27ve detected you are located in", "locationDropdownText": "Please select your location", "changeLocationText": " Change location", "locationSaveText": " Save location", "learnMoreText": "Learn how we use this information", "regions": [ { "tag": { "tagID": "localization:africa" }, "title": "Africa" }, { "tag": { "tagID": "localization:australia" }, "title": "Australia" }, { "tag": { "tagID": "localization:belgium" }, "title": "Belgium" }, { "tag": { "tagID": "localization:caribbean" }, "title": "Caribbean" }, { "tag": { "tagID": "localization:china" }, "title": "China" }, { "tag": { "tagID": "localization:czech\u002Drepublic" }, "title": "Czech Republic" }, { "tag": { "tagID": "localization:germany" }, "title": "Germany" }, { "tag": { "tagID": "localization:global" }, "title": "Global" }, { "tag": { "tagID": "localization:india" }, "title": "India" }, { "tag": { "tagID": "localization:italy" }, "title": "Italy" }, { "tag": { "tagID": "localization:latin\u002Damerica" }, "title": "Latin America" }, { "tag": { "tagID": "localization:middle\u002Deast" }, "title": "Middle East" }, { "tag": { "tagID": "localization:netherlands" }, "title": "Netherlands" }, { "tag": { "tagID": "localization:new\u002Dzealand" }, "title": "New Zealand" }, { "tag": { "tagID": "localization:north\u002Damerica" }, "title": "North America" }, { "tag": { "tagID": "localization:slovakia" }, "title": "Slovakia" }, { "tag": { "tagID": "localization:taiwan" }, "title": "Taiwan" }, { "tag": { "tagID": "localization:turkey" }, "title": "Turkiye" }, { "tag": { "tagID": "localization:united\u002Dkingdom" }, "title": "United Kingdom" } ], "currentRegion": "Global", "learnMoreLink": { "url": "\/en\/copyright.html", "target": "" } }, "searchBar": { "searchPageLink": { "url": "\/en\/search.html" }, "labels": { "searchInputLabel": "Search Input", "searchBarPlaceholder": "What are you looking for?", "closeButton": "Close Button", "suggestionsFooter": "Search for results", "noSuggestions": "No suggestions found" }, "searchBarSuggestions": { "endpoint": "/bin/suggestions.json", "currentPath": "\/content\/stantec\/en\/ideas\/content\/blog\/2026\/liquid\u002Dcooling\u002Ddata\u002Dcenters\u002Dkey\u002Dto\u002Dmeeting\u002Dai\u002Ddriven\u002Dheat\u002Ddemands" } }, "languageSelectorModal": { "languageSelectorTitle": "Language \/ Region", "languageSelectorDescription": "Select your region to visit a Stantec regional site, or visit our Locations Hub on our Global site where you can learn more about Stantec in your local area.
", "languageSelectorImage": { "imageLarge": "\/content\/dam\/stantec\/images\/Icons\/Map.svg", "altText": " World map" }, "globalRegion": { "name": "Global", "regionTag": "localization:global", "description": "", "languages": [ { "langCode": "en", "name": "English", "pathLink": { "url": "en" }, "active": "true" }, { "langCode": "fr", "name": "Français", "pathLink": { "url": "fr" }, "active": "false" } ] }, "regionLanguages": [ { "regionTag": "localization:china", "name": "China", "description": "", "languages": [ { "langCode": "en", "name": "English", "pathLink": { "url": "cn" }, "active": "false" }, { "langCode": "en", "name": "", "pathLink": { "url": "en" }, "active": "false" } ] }, { "regionTag": "localization:italy", "name": "Italy", "description": "", "languages": [ { "langCode": "it", "name": "Italiano", "pathLink": { "url": "it" }, "active": "false" }, { "langCode": "en", "name": "", "pathLink": { "url": "en" }, "active": "false" } ] }, { "regionTag": "localization:latin\u002Damerica", "name": "Latin America", "description": "", "languages": [ { "langCode": "es", "name": "Español", "pathLink": { "url": "es" }, "active": "false" }, { "langCode": "en", "name": "", "pathLink": { "url": "en" }, "active": "false" } ] }, { "regionTag": "localization:netherlands", "name": "Netherlands", "description": "", "languages": [ { "langCode": "nl", "name": "Nederlands", "pathLink": { "url": "nl" }, "active": "false" }, { "langCode": "en", "name": "", "pathLink": { "url": "en" }, "active": "false" } ] }, { "regionTag": "localization:taiwan", "name": "Taiwan", "description": "", "languages": [ { "langCode": "zh_tw", "name": "繁體中文", "pathLink": { "url": "zh_tw" }, "active": "false" }, { "langCode": "en", "name": "", "pathLink": { "url": "en" }, "active": "false" } ] }, { "regionTag": "localization:turkey", "name": "Turkiye", "description": "", "languages": [ { "langCode": "tr", "name": "Türkçe", "pathLink": { "url": "tr" }, "active": "false" }, { "langCode": "en", "name": "", "pathLink": { "url": "en" }, "active": "false" } ] } ], "originPath": "/content/stantec/en/ideas/content/blog/2026/liquid-cooling-data-centers-key-to-meeting-ai-driven-heat-demands.html" } }June 24, 2026
Liquid cooling innovation is essential for scaling data centers for AI. It’s more efficient. And it can work with LEED.
A version of this blog first appeared as “Liquid cooling can help data centers meet their AI compute thermal challenges” in Design Quarterly, Issue 28.
Data centers are heating up. But there’s a better way to keep them cool. What is it? Liquid cooling.
Running an artificial intelligence (AI) chatbot like ChatGPT or Copilot requires modern accelerators like graphics processing units or neural processing units. These are the chips that handle the advanced processing that makes AI possible. Performing trillions of operations per second generates heat—a lot of it. And these AI apps have millions of users.
Advanced, modern processing requires similar solutions: Advanced, modern cooling.
Data centers already use a huge amount of energy. A 2024 Report from the Lawrence Berkeley National Laboratory showed that data centers used 4.4 percent of all US electricity in 2023. And AI and other compute-intensive uses—training large language models, navigation for autonomous vehicles, virtual or augmented reality, cryptocurrency—will only increase that appetite. By 2028, they could use up to 12 percent of US electricity.
" }Most liquid-cooled systems use a coolant distribution unit for direct-to-chip cooling. A CDU distributes coolant to a closed-loop system. This schematic is simplified to convey the facility water/technical water system interface.
AI workloads have increased the power density of data centers. Higher power density means each piece of information technology equipment (ITE) consumes more energy and produces more heat per square foot. In the 2010s, cloud computing rack densities climbed to 8 to 20 kilowatts (kW). Today, it’s common to have ITE loads over 100 kW per cabinet. Soon, they are expected to surpass 500 kW.
So, where are we today? And why are we transitioning to liquid cooling?
Without cooling, data center servers would overheat and fail. So, data centers need systems to remove heat from their servers, storage, and networking equipment.
Air cooling—where cool air is circulated through racks of equipment, dissipating heat—has been the standard approach to cooling ITE for decades. With high-density data centers, air cooling is no longer the best option. We are trending toward rack densities that can’t be cooled with air alone. We must find a way to move more heat efficiently.
Liquids can absorb more heat than air by volume, and they’re more thermally conductive.
With liquids, you get more value for your investment from a heat-transfer perspective. The specific heat-transfer ability of a liquid is far superior to that of air; you can move much more heat with a liquid than you can with the same volume of air. Because liquids transfer heat so well, liquid cooling can handle data center cabinets with loads of 100 kW or more.
As we’ve said, heat load densities are increasing. That makes liquid cooling in data centers a popular solution. McKinsey projects the global data center cooling market will hit $40 billion by 2030. And it expects that liquid cooling will account for at least $15 billion of that.
Today, most liquid-cooled systems use a coolant distribution unit (CDU) for direct-to-chip (D2C) cooling. A CDU distributes coolants (commonly 25 percent propylene glycol) to a closed-loop system, which circulates through cold plates mounted on ITE. The CDU keeps the temperature of the coolant supply steady.
The technical water system is the coolant loop that flows between the CDU and the ITE. It transfers heat gained from the ITE to the facility water loop via the CDU heat exchanger. A liquid-to-liquid heat exchanger in the CDU transfers ITE waste heat from the technical water system to the facility water system, which removes heat through the building’s mechanical systems.
The CDU keeps its cooling fluids separate from the building system. This is critical for ensuring coolant quality for ITE. A liquid system should maintain the manufacturer’s guidelines on coolant quality—things such as pH, hardness, conductivity, and dissolved solids—to help reduce equipment failure and improve performance.
Liquid cooling in data centers is more efficient than air cooling for data centers. It’s true, we can improve the effectiveness of air-cooling systems with aisle containment, cabinet chimneys, and in-row cooling units. But pushing all this air means data center cooling systems consume roughly 30 to 40 percent of total data center power. Liquid coolants are more efficient than air in these high-temperature racks.
Higher ITE loads demand a higher required flow rate (for air or liquid) to cool them. But the liquid cooling system’s pumps use less energy than air fans to transfer the same amount of heat. When we look at the power usage effectiveness for data centers with liquid-cooled ITE, they spend less of their total energy on cooling and more on running IT equipment.
Higher temps mean higher efficiency. Because of liquid cooling’s heat-transfer performance, the coolant need not be “cold.” These systems can use warmer coolants, even up to 100°F (37 °C). A dedicated liquid-cooling system designed for higher-temperature operations can achieve optimum efficiency. And use less water.
Liquid cooling in data centers also takes up less space in the data hall than air-cooling systems. This allows more space for the racks and higher-density computing.
And ASHRAE reports that data centers that use hybrid (air/liquid) systems can reduce their total cost of ownership.
" }A new hybrid cooling system at the NTT Data Center Campus in Garland, Texas, combines direct chip cooling with air-assisted cooling.
One option is to combine liquid cooling and air cooling in data centers. It’s a hybrid approach.
Air-cooled data centers—with direct evaporative air-handling units or air‑cooled computer room air-conditioning (CRACs)—can add liquid cooling using CDUs with liquid-to-air heat exchangers. This is a common approach to retrofitting older facilities, which might otherwise become obsolete. A facility water system makes it easier to cool the ITE with liquid, but it’s not required.
Some manufacturers offer modular split-system chillers. These can replace existing CRACs and connect to existing air-cooled condensers on the roof to remove heat. We can install CDUs in mechanical galleries next to air-handling units and air-conditioned computer rooms or within white space near cabinets, depending on project needs.
Often, CDU-based liquid cooling in data centers can’t capture all the heat by itself. So, some lower-capacity air cooling is still needed. Greenfield data centers built for liquid-cooled systems typically use liquid cooling for 70 to 95 percent of the ITE load, with air cooling covering the rest.
There are diverse options for liquid cooling. A rear door heat exchanger cools the air leaving the server rack but never touches the ITE directly. Chilled water systems run cold water into the environment, pull the heat out, and send it to the building water system.
D2C systems circulate coolant through a cold plate in contact with server electronics, transferring heat away in the process. Immersion cooling might be the most sci-fi looking approach. Entire servers and data racks are immersed in nonconductive fluid.
Some data companies want their large-scale digital projects to be certified as sustainable. But liquid cooling involves new technology that LEED guidelines have yet to address.
Since LEED had no established methodology for modeling or certifying liquid-cooled systems, we created a process to support LEED approval.
Working within the Optimize Energy Performance credit, we developed the energy modeling framework and documentation standards from scratch. It compares hybrid liquid-cooled systems against traditional air-cooled baselines in a format LEED reviewers could evaluate and approve.
Here are four steps we’re taking to make LEED certification possible for data centers cooling their ITE with liquid systems.
" }Since LEED had no established methodology for modeling or certifying liquid-cooled systems, we created a process to support LEED approval.
Step 1: Our team worked on two hyperscale data center projects for the same confidential client. Our models showed that liquid cooling offered:
The results were impressive. The model showed gains far beyond LEED minimums for energy savings and power usage effectiveness.
And our models showed possible savings of 8 to 12 percent on IT equipment energy use thanks to better thermal management. (Project results are based on modeling and project-specific conditions; they may not be representative of all projects.)
Step 2: Liquid-cooled systems are new and not widely understood. So, provided LEED reviewers with the information to help them. It included:
Step 3: We used the LEED framework to measure and prove that a liquid-cooled system saved water versus a baseline air-cooled system. Our calculations show that the closed-loop system could save about 92,000 gallons of water per megawatt per year, when compared to evaporative systems.
Step 4: Next, we showed that liquid cooling could meet the International Energy Conservation Code requirements. We compared the total annual energy cost for the liquid-cooling design against a baseline building design. We demonstrated, through comparative analysis, that the liquid cooling design was capable of using less energy than code minimum air-cooled systems.
We regularly work with hyperscalers on cooling systems for data centers. We have found that educating operations personnel on the fundamentals of liquid cooling, like those above, goes a long way in solving some of the challenges to deployment.
It helps to be on the same page. Material compatibility between the coolant, liquid-cooled ITE, and distribution system (piping, CDU, and accessories) can also pose design challenges. But despite these challenges, liquid cooling is here. And we need it.
" }Hybrid cooling systems, like this one at the NTT Data Center Campus in Garland, Texas, combine direct chip cooling with air-assisted cooling, which enhances efficiency at the chip level.
There are several steps to get the liquid-cooling process started. They include a site audit, technical specifications, a review of space needs, and a discussion of sustainability.
Audit: Before you commit to converting your data center to hybrid cooling, specialists will need to conduct a site audit. This will likely include:
Specifications: Consider the technical requirements for the ITE you are planning to liquid cool. This includes fluid requirements, temperatures, and flows.
Space: Understand the space requirements for liquid cooling. The liquid cooling system will need space to add additional piping for the fluid network, CDUs, and treatment to support technical water and monitoring.
Sustainability: If you are seeking LEED certification on a liquid-cooled data center, you should consider selecting equipment with modularity built in. This will help your system meet energy requirements mandated by USGBC.
Here’s something else to consider: If you are looking for an ASHRAE Standard 90.4 code compliance certification on a liquid-cooled data center, you will likely need to assess the part-load energy of all of your equipment. That looks at the energy consumed when systems are operating below capacity.
As demand for AI, machine learning, and advanced processing continues to grow, so does the heat in our data racks. Liquid cooling technology isn’t just critical to tomorrow’s data infrastructure.
It’s indispensable.
" }