13 Nov 2026
Why Are Data Centers Moving to Liquid Cooling?
Data centers are adopting liquid cooling because the newest AI racks produce more heat than fans and air conditioning can remove efficiently. Liquids carry heat far better than air, so cold plates on chips or tanks of non-conductive fluid can cool racks drawing 100 kilowatts or more. The shift is concentrated in AI and high-performance computing, while most conventional enterprise servers still run on air.
Why does AI create a cooling problem?
Nearly all the electricity a server consumes becomes heat, and that heat must be removed to keep chips within safe operating temperatures. For decades, data centers handled this with air: fans push cool air across servers, and computer room air handlers or chillers remove the warm air.
That approach works well at moderate densities. The Uptime Institute’s 2026 survey found the most common rack density was about 11 kilowatts, up from 9 kilowatts a year earlier.1 Uptime describes perimeter air cooling as typical up to about 20 to 25 kilowatts per rack, close-coupled air systems up to about 50 kilowatts, and liquid cooling as the usual approach above 50 kilowatts.2
AI racks have moved well past those levels. Nvidia’s reference architecture for one current AI system calls for up to 142 kilowatts per rack, and Schneider Electric has noted that a newer platform can push rack requirements up to 246 kilowatts.13
How does liquid cooling work?
Direct-to-chip (cold plate) cooling. Metal plates with internal channels sit directly on processors and memory. Coolant, usually a water-glycol mix, flows through the plates, absorbs heat and carries it to a coolant distribution unit (CDU), which transfers the heat to the building’s water loop. Fans still cool the remaining components. Schneider Electric estimated that direct-to-chip systems held about 55% of the liquid cooling market in 2026.1
Rear-door heat exchangers. Radiator-like doors on the back of racks remove heat from exhaust air. They extend air cooling to higher densities without bringing liquid to the chip.
Single-phase immersion. Servers sit in tanks of dielectric fluid that does not conduct electricity. The fluid absorbs heat and is pumped through a heat exchanger.
Two-phase cooling. A fluid boils at a low temperature on contact with hot components, and the vapor condenses on a cooler surface. This method moves heat efficiently but uses specialized fluids, some of which face environmental regulation.
What equipment does liquid cooling require?
Liquid cooling adds a new set of components to the data center bill of materials: cold plates, manifolds, quick-disconnect fittings, CDUs, pumps, piping, leak detection systems, heat exchangers, dielectric fluids and immersion tanks. It also changes the building itself, adding floor loading, water treatment and redundant plumbing. On the heat-rejection side, sites still need chillers, dry coolers or cooling towers. Our article on the AI data center bill of materials shows where cooling fits.
Is liquid cooling taking over all data centers?
Not yet. Survey data varies. Uptime’s 2026 survey put average density, excluding purpose-built high-density sites, at only 7.8 kilowatts per rack, while an AFCOM report put the 2026 average at 27 kilowatts, reflecting differences in who was surveyed.4 An Uptime analyst described liquid cooling as standard practice at the top of the density curve and still niche elsewhere.4 Uptime has also written that liquid cooling is likely to remain concentrated in high-density applications for some time.5
The practical picture: new AI training and large inference clusters increasingly specify liquid cooling from the start, while many existing enterprise data centers continue with air or add liquid in a few rows.
What are the benefits and trade-offs?
Benefits. Liquid cooling supports much higher rack densities, which lets more computing fit in less floor space. It can reduce the electricity used for cooling, freeing power for computing in sites where electricity is the binding constraint. It can also allow higher coolant temperatures, which may reduce chiller use.
Trade-offs. Systems cost more up front, require new skills to operate and maintain, and add the risk of leaks. Retrofitting existing buildings can be difficult. Standards for components and fluids are still developing, and supply of specialized parts such as CDUs and fittings has been tight at times.
How does cooling connect to power?
Cooling and power are linked constraints. Every kilowatt spent on fans and chillers is a kilowatt not available for computing, so sites limited by grid capacity have a strong reason to cool more efficiently. Liquid systems also allow denser racks, which shortens cable runs and can reduce networking costs.
What does this mean for investors?
Cooling is a part of AI infrastructure that grows with power density rather than with chip count alone. The supplier base includes large electrical and thermal equipment manufacturers, specialized cooling firms, component makers and fluid suppliers. Some of these companies also serve industrial, commercial and residential heating and cooling markets, which can dilute or diversify their AI exposure. Risks include technology shifts between cooling methods, customer concentration and competition as more firms enter. Our articles on how data centers work and AI power and electrification cover related layers.
Frequently asked questions
At what rack density is liquid cooling needed?
There is no single cutoff. The Uptime Institute describes air cooling as typical up to roughly 20 to 50 kilowatts per rack, depending on the design, with liquid cooling typically used above about 50 kilowatts. Some operators introduce liquid at lower densities to improve efficiency. Current AI racks often exceed 100 kilowatts, well beyond what air alone can handle.
Is liquid cooling safe for electronics?
Direct-to-chip systems keep coolant inside sealed plates and pipes, so it does not touch circuits under normal operation. Designs include leak detection and quick-disconnect fittings that close when unplugged. Immersion systems use dielectric fluids that do not conduct electricity, so electronics can sit directly in them. Leaks remain an operational risk that operators manage with monitoring and maintenance.
Does liquid cooling use more water?
Not necessarily. The liquid loop that touches the chips is usually closed and recirculated. Water use depends mainly on how a site rejects heat to the outdoors. Evaporative cooling towers consume water, while dry coolers use little or none. Some liquid-cooled designs allow higher coolant temperatures, which can reduce the need for evaporative cooling in certain climates.
What is a coolant distribution unit?
A coolant distribution unit, or CDU, is equipment that circulates coolant through the server loop at the right flow rate, pressure and temperature, and transfers heat to the building’s facility water. It typically contains pumps, heat exchangers, filters and controls. CDUs can sit inside a rack, at the end of a row or in a separate room.
Can older data centers add liquid cooling?
Some can, often in limited areas. Rear-door heat exchangers and in-row sidecar units let operators add higher-density racks without rebuilding the facility. Full direct-to-chip deployment may require new piping, structural reinforcement and changes to power distribution. Cost and complexity lead many operators to reserve liquid cooling for new buildings or dedicated halls.
How do thematic ETFs get exposure to cooling?
Some AI infrastructure, data center or electrification ETFs hold thermal management and electrical equipment manufacturers. In many funds, cooling suppliers make up a modest share alongside chip, networking and power holdings. Because several cooling companies also serve non-data-center markets, reviewing their revenue mix helps investors understand how much AI exposure a fund actually carries.
- Data Center Knowledge, “AI Rack Density’s Real Limits: Power, Cooling, Failure Risk,” 2026, citing Uptime Institute’s 16th Annual Global Data Center Survey, Nvidia’s GB300 NVL72 reference architecture and Schneider Electric.
- Uptime Institute, “AI and Cooling: Methods and Capacities,” Intelligence Update, February 4, 2025.
- Schneider Electric, “Data Center Power Density: Planning Liquid-Cooled AI Data Centers Around Grid and Power Constraints,” blog post, July 28, 2026.
- Capacity Media, reporting on Uptime Institute’s 2026 Global Data Center Survey and AFCOM’s 2026 State of the Data Center report, 2026.
- Uptime Institute, Intelligence Update on the outlook for liquid cooling beyond high-density applications, January 21, 2026.
Cite this page: Adam Patti, "Why Are Data Centers Moving to Liquid Cooling?," VistaShares, updated November 13, 2026, https://www.vistashares.com/data-center-liquid-cooling/. Plain-text version.
This material is provided for educational purposes only. It is not investment, legal or tax advice, and it is not a recommendation or an offer to buy or sell any security. Information is drawn from sources believed to be reliable as of the date of publication, is not guaranteed as to accuracy or completeness, and may change without notice. Figures attributed to third parties are their estimates. Projections and forecasts are not guarantees of future results. Investors should consult their own financial, legal and tax professionals before making investment decisions. Investments involve risk, including possible loss of principal. Companies in a single industry or theme may be affected by common economic, regulatory or technological developments and may be more volatile than the broad market. References to companies, governments or programs are for illustration only.