Executive Summary

Between 2025 and 2035 the Liquid Cooling for AI Data Centers Market is projected to expand from 3.4 USD Billion to 23.2 USD Billion, a CAGR of 21.21%.

Two mechanisms sit behind that trajectory: accelerator power density has outrun air-cooled rack limits, and the EU Cyber Resilience Act (Regulation (EU) 2024/2847) now pulls cooling-distribution-unit controllers into secure-by-design compliance scope.

North America held 40.0% of the market in 2025, ahead of Asia Pacific at 28.0% and Europe at 20.0%. Direct-to-Chip Cooling leads the technology axis; Cooling Distribution Units lead by component.

Integration cost against legacy air-cooled shells is the principal constraint on retrofit pace. Vendors span component specialists, CDU makers, and systems integrators bidding the same hyperscale accounts, rather than one supplier controlling deployment.

Key Takeaways

• USD 23.23 Billion by 2035, up from USD 3.39 Billion in 2025, is a 21.21% compound rate.
• Direct-to-Chip Cooling holds the largest position on cooling technology.
• Cooling Distribution Units holds the largest position on component.
• 40.0% of 2025 revenue was earned in North America.
• 10 suppliers are profiled.

Market Definition and Scope

The Liquid Cooling for AI Data Centers Market covers direct-to-chip cold plates, single- and two-phase immersion tanks, hybrid liquid-air systems, rear-door heat exchangers, and in-row units, together with the cooling distribution units, manifolds, pumps, heat exchangers, and installation, maintenance, and design services that deploy them inside data centers built or retrofitted for AI training and inference workloads.

Excluded are conventional air-cooled HVAC systems for standard enterprise racks and general industrial process cooling, both of which sit outside AI-specific thermal management.

Market Trends

Accelerator Power Density Is Pushing Racks Past Air Cooling’s Practical Ceiling

Current AI training and inference accelerators concentrate more heat per socket than a chilled-air stream can remove at hyperscale rack density, a constraint hyperscalers now disclose through their own capital-expenditure guidance on AI infrastructure builds. Direct-to-chip cold plates and immersion tanks address the same watt-per-rack problem from opposite ends: one intercepts heat at the die, the other floods the chassis. Demand concentrates on new-build AI halls sized for dense, latency-bound workloads first, then flows into retrofit projects as existing air-cooled floors reach their power ceiling.

Cooling Distribution Units Are Becoming the Standard Retrofit Bridge for Existing Halls

Operators adding AI racks into facilities designed for air cooling are inserting cooling distribution units as the interface between facility chilled water and server-level coolant loops, avoiding a full mechanical rebuild. Hyperscale cloud providers building new AI-dedicated regions are specifying CDUs and hybrid liquid-air configurations at the design stage, often to meet sovereign-cloud and data-residency commitments under the EU Data Act and China’s Personal Information Protection Law that tie a given tenancy to a specific jurisdiction. US federal deployments carry a parallel constraint: facilities seeking FedRAMP authorization must qualify cooling hardware alongside compute, which slows retrofit timelines relative to commercial colocation. CDU vendors are also layering consumption-based monitoring subscriptions onto hardware sales, shifting part of their revenue from one-time equipment orders toward ARR-generating service contracts.

Embedded Controllers in Cooling Hardware Are Entering Cybersecurity Compliance Scope

Cooling distribution units and pump controllers increasingly carry network-connected firmware for remote monitoring, which brings them inside the EU Cyber Resilience Act’s (Regulation (EU) 2024/2847) secure-by-design and vulnerability-reporting requirements for products with digital elements. The EU’s NIS2 Directive compounds this by classifying data centre operators as essential or important entities, extending incident-reporting and supply-chain security obligations to the cooling vendors those operators depend on. Colocation facilities hosting regulated financial-sector workloads face a further layer under DORA, which requires EU banks and insurers to test the operational resilience of ICT third parties, including the facilities and cooling systems underpinning their compute. Vendors selling into the European Union now carry technical documentation and incident-reporting obligations alongside thermal performance specifications, which raises the qualification bar for component suppliers and favors those already maintaining security documentation for other networked data center equipment.

Growth Drivers and Restraints

Rising Accelerator Power Density Is Forcing a Shift From Air to Direct Liquid Contact

Successive generations of AI training and inference accelerators draw more power per socket than a chilled-air stream can remove within a standard rack footprint. Hyperscale cloud providers now disclose AI-infrastructure capital expenditure growth on quarterly earnings calls, and GPU vendors’ accelerator roadmaps assume liquid contact at the die from the next generation onward. Cooling suppliers are converting that demand into consumption-based service contracts rather than one-time equipment sales, and several now report the resulting annual recurring revenue (ARR) alongside unit shipments. The effect concentrates in the Direct-to-Chip Cooling segment, which the technology axis already leads, as training clusters convert first and inference fleets follow.

Facility Power Ceilings Are Rewarding Higher Cooling Efficiency Per Rack

Data center campuses face fixed grid interconnection capacity, so operators extract more compute per megawatt delivered rather than wait on new substations. Liquid cooling lowers the power a facility spends moving air and running chillers, freeing headroom for additional racks within the same interconnection agreement. Colocation operators bidding for hyperscale anchor tenants increasingly specify liquid-ready shells at the design stage, pulling cooling distribution units and heat exchangers into new-build specifications rather than treating them as a later retrofit.

EU and National Security Rules Are Naming Data Centre Operators as Regulated Entities

The NIS2 Directive (EU 2022/2555) classifies data centre and cloud service providers as essential or important entities, which brings incident reporting and supply-chain security review to networked equipment such as cooling distribution units shipping with remote-monitoring firmware. Financial institutions carry a parallel obligation under the Digital Operational Resilience Act (DORA), which tests ICT third-party risk down to facility level; a coolant-loop failure counts as an ICT incident, pushing banks toward N+1 liquid-cooling redundancy rather than single-loop designs. Outside Europe, China’s Personal Information Protection Law (PIPL) restricts cross-border data transfer and is driving liquid-cooled capacity build-out inside China instead of reliance on offshore hyperscale regions, while FedRAMP authorization requirements for U.S. federal cloud workloads add a qualification step for cooling vendors entering government data halls.

Retrofitting Legacy Air-Cooled Shells Carries a High Integration Cost

Facilities designed around computer room air handlers often lack the piping runs, leak detection, and water treatment that direct-to-chip and immersion systems require, so operators face structural modification cost on top of the cooling hardware itself. The burden falls hardest on older colocation halls competing for AI tenants against purpose-built new construction, slowing the pace at which existing capacity converts even as new-build facilities specify liquid cooling from day one.

A Shortage of Liquid-Cooling Installation and Maintenance Skills Slows Deployment

Direct-to-chip, immersion, and hybrid systems each demand different fluid-handling, leak-testing, and maintenance procedures, and the operator base trained on air-cooled facilities has not scaled at the same pace as rack deployment. Systems integrators and services providers absorb the resulting project delays most directly, since installation and commissioning work depends on specialist labor that remains concentrated among a small number of vendors, rather than on cooling technology or component supply.

Regional Analysis

North America

Revenue of USD 1.36 Billion in 2025 makes this the largest regional market, on 40.0% of the total.

Asia Pacific

At 28.0% in 2025, this is the second-largest regional market, worth USD 0.95 Billion.

Europe

Revenue of USD 0.68 Billion in 2025 makes this the third-largest regional market, on 20.0% of the total.

Segment Analysis

By Cooling Technology

• Direct-to-Chip Cooling (largest) – A liquid cooling method that circulates coolant through cold plates mounted directly on CPUs, GPUs, or other high-heat server components to remove heat at the source

• Single-Phase Direct-to-Chip Cooling
• Two-Phase Direct-to-Chip Cooling
• Immersion Cooling – A cooling approach in which entire servers or components are submerged in a dielectric fluid that absorbs heat directly from all surfaces without causing electrical damage

• Single-Phase Immersion Cooling
• Two-Phase Immersion Cooling

• Hybrid Liquid Cooling – A combined approach that pairs liquid cooling for high-heat components like processors with air cooling for the remaining rack or chassis, often used to retrofit existing facilities
• Other Liquid Cooling Technologies – A category covering additional liquid-based heat removal methods, such as rear-door heat exchangers and in-row liquid cooling units, that do not fit the direct-to-chip, immersion, or hybrid categories

• Rear Door Heat Exchanger (RDHx)
• In-Row/In-Rack Liquid Cooling
• Liquid-to-Air Cooling

Direct-to-chip cooling is the leading technology in 2025, ahead of immersion, hybrid, and other liquid-cooling formats. It leads because cold plates mounted on CPUs and GPUs pull heat directly from the hottest components while leaving the rest of the rack air-cooled, letting operators retrofit existing halls without replacing chassis or handling dielectric fluid. The approach also tracks server OEM reference designs, shortening qualification cycles for hyperscaler buyers. Siting now follows data-residency law as much as power availability: GDPR pushes EU cloud regions toward local capacity, and NIS2 classifies large data centers as essential entities, putting cooling redundancy under the same audit scrutiny as network uptime. DORA adds a further requirement for EU financial-sector tenants, who specify resilient, liquid-cooled capacity for workloads subject to operational-resilience testing. Immersion cooling is growing fastest. GPU thermal design power is rising faster than cold plates alone can extract, and two-phase immersion tanks submerge the entire server, removing heat from memory, power delivery, and networking components as well as the processor package. That shift toward full-system heat capture is what is pulling new training-cluster deployments away from air and toward immersion.

By Component

• Cooling Distribution Units (largest) – A pumped-liquid unit that interfaces facility water with server-level coolant loops, regulating temperature, flow, and pressure to IT racks

• Liquid-to-Liquid CDUs
• Liquid-to-Air CDUs
• Cold Plates & Manifolds – Metal plates mounted directly on chips that transfer heat into a coolant, connected by manifolds that route liquid to and from each server
• Cold Plates

• Copper Cold Plates
• Aluminum Cold Plates
• Manifolds
• Pumps & Heat Exchangers – Mechanical devices that circulate coolant through the loop and transfer captured heat to a secondary fluid or ambient air for rejection
• Pumps

• Centrifugal Pumps
• Positive Displacement Pumps
• Heat Exchangers
• Immersion Tanks – Sealed enclosures in which entire servers are submerged in a dielectric fluid that absorbs heat directly from components

• Single-Phase Immersion Tanks
• Two-Phase Immersion Tanks
• Services – Design, installation, integration, and maintenance work performed by vendors to deploy and operate liquid cooling systems in a data center

• Installation & Deployment
• Maintenance & Support
• Consulting & Design

Cooling distribution units lead the component segmentation in 2025, ahead of cold plates and manifolds, pumps and heat exchangers, immersion tanks, and services. The CDU interfaces the facility water loop with server-level coolant loops and regulates flow, temperature, and pressure, making it the one component every liquid-cooling architecture requires regardless of whether the rack runs direct-to-chip, immersion, or hybrid cooling. CDU firmware now shows up in CVE listings alongside other networked facility gear, and operators running FedRAMP-authorized federal workloads have converged on standardized, pre-certified CDU models to keep the authorization boundary intact. Services is growing fastest, as operators without in-house thermal engineering staff turn to vendors for design, installation, and maintenance when retrofitting air-cooled halls for liquid loops, much of it now sold as multi-year, ARR-style contracts rather than one-off installation fees. China’s PIPL localization rules add a further pull, since data kept in-country is driving new domestic builds that need commissioning support rather than imported turnkey systems. That dependence on external design and commissioning work is what is pulling services revenue up alongside each new deployment, not only alongside hardware refresh cycles.

Competitive Landscape

The liquid cooling for AI data centers market is led by a group of established thermal-management and infrastructure vendors rather than a small set of dominant suppliers. Competition centers on integration surface: cold plate and CDU designs must qualify against server OEM reference architectures, so vendors compete on thermal performance per watt, compatibility with existing rack power and networking, and the speed at which a design can be validated for hyperscale deployment. Serviceability and total cost of ownership, including coolant maintenance and leak risk, weigh alongside raw heat-rejection capacity. Vendor lock-in also plays a role, since coolant chemistry and manifold fittings chosen at initial build-out shape which supplier services the rack through its refresh cycle. Named players active in this market include Vertiv Holdings Co., Schneider Electric SE, Rittal GmbH & Co. KG, CoolIT Systems, Asetek A/S, Green Revolution Cooling, Inc., Submer Technologies S.L., LiquidStack Holding B.V., Alfa Laval AB, and Boyd Corporation, spanning direct-to-chip, immersion, and rack-level cooling infrastructure.

Strategic Outlook

The clearest whitespace is in retrofitting existing air-cooled halls in North America and Europe, where operators adding GPU racks need a cooling path that avoids a full facility rebuild. Immersion and hybrid architectures stand to benefit most, provided facility water loops and floor loading can absorb liquid infrastructure without a multi-year construction cycle.

By 2035, cooling distribution units and cold plates are likely to become standard infrastructure rather than optional upgrades, as GPU density makes air cooling the exception. Buyers are expected to weigh serviceability and coolant compatibility as heavily as raw heat-rejection capacity when selecting a vendor.

Liquid Cooling for AI Data Centers Market Report Scope