Liquid cooling solutions are projected to reach a $23.24 billion market by 2035, driven by gigawatt-scale AI data center deployment.
Artificial intelligence is rewriting the data center cooling playbook. As rack densities surge toward unprecedented levels, liquid cooling is becoming critical to performance, efficiency, and scalability.
AI infrastructure is moving rapidly toward larger campuses, denser GPU clusters, and more demanding thermal loads. The shift is changing cooling from a facility support function into a central design decision that influences compute performance, power efficiency, water use and deployment speed. According to DataM Intelligence, the data center liquid cooling market is estimated at USD 3.39 billion in 2025 and is projected to reach USD 23.24 billion by 2035, growing at a CAGR of 21.21% from 2026 to 2035. North America held about 35.6% of the market, while direct-to-chip cooling represented about 42.9% of technology demand.
The scale of the thermal challenge is increasing as AI racks move beyond the limits of conventional air cooling. Rack power densities above 30 kW to 100 kW are making air-based systems less practical for modern AI workloads. NVIDIA's reference architecture for Rubin systems shows rack power near 225 kW, while power suppliers are developing architectures for 1 MW racks and larger configurations. At campus scale, thousands of racks can create a concentrated heat rejection requirement that must be planned alongside substations, backup power, and water infrastructure.
This transition is creating a competitive market across direct-to-chip cooling, single-phase immersion, rear door heat exchangers, and spray or jet cooling. Each technology serves a different stage of the deployment curve. Direct-to-chip systems offer a practical route for new AI halls and phased retrofits. Rear door heat exchangers can extend the useful life of air-cooled facilities. Immersion cooling can support very high-density environments where operators can redesign server service models. Spray and jet approaches are emerging for localized hot spot management and future chip packages.
The move toward gigawatt-scale AI campuses broadens the addressable market well beyond cold plates and coolant distribution units. A campus that supports several hundred megawatts of IT load requires a coordinated thermal system covering rack loops, facility water loops, heat exchangers and heat rejection equipment. The design must also include pumps, valves, manifolds, sensors and leak detection systems that can operate across thousands of nodes with consistent reliability.
DataM Intelligence estimates that solutions account for about 74.5% of market revenue, reflecting demand for integrated systems rather than individual components. Hyperscale facilities represent about 38.9% of revenue, and AI or machine learning workloads account for about 34.8% of demand. Less than 30% of global data centers currently use liquid cooling, which leaves substantial conversion potential as new server platforms make liquid readiness a standard procurement requirement.
The largest near-term opportunity is likely to sit in direct-to-chip systems because they can remove heat at the processor while preserving familiar rack layouts and service practices. Cold plates, coolant distribution units, and warm water loops can also reduce dependence on mechanical chillers in suitable climates. Research published in 2026 showed that optimized cold plate channel designs reduced maximum chip temperature by more than 35 degrees Celsius compared with a baseline design, demonstrating how material engineering and fluid path optimization can improve thermal performance.
Immersion cooling is gaining attention for dense AI and high-performance computing environments because it surrounds electronic components with dielectric fluid. The model can reduce fan energy and enable compact server configurations. Adoption still depends on server warranties, maintenance workflows, fluid compatibility, and standardization. Rear door heat exchangers remain important for operators that need a transitional architecture, while hybrid air and liquid systems can support mixed halls with legacy servers and new GPU racks.
Cooling technology selection is increasingly shaped by local constraints. In the United States, data centers are expected to account for about half of electricity demand growth through 2030, according to the International Energy Agency. Large campuses are being planned in regions with grid access, land and gas infrastructure, yet water availability and interconnection delays are influencing facility design. Closed-loop liquid systems, dry coolers and warm water operation can reduce water dependence and support faster permitting in constrained markets.
Europe is placing greater emphasis on energy efficiency, heat recovery and urban integration. Liquid cooling can produce higher quality waste heat than conventional air systems, improving the economics of district heating and industrial heat reuse where local networks exist. The opportunity extends to heat pumps, thermal storage and digital control platforms that match cooling output with external heat demand. Operators that can monetize recovered heat may improve both sustainability performance and site economics.
The competitive landscape is expanding across several layers of the cooling value chain. Cold plate and CDU manufacturers are competing on thermal resistance, flow control and serviceability. Fluid suppliers are developing dielectric liquids, corrosion protection packages and long-life coolant formulations. Pump and valve suppliers are adapting products for higher reliability and lower leakage risk. Digital platforms are becoming important because gigawatt campuses require continuous monitoring of temperature, pressure, flow and coolant quality.
Digital twins and adaptive control can create another efficiency layer. A 2026 study based on the Frontier exascale supercomputer found that coordinated optimization of flow rate and supply temperature could deliver energy savings of about 27.8% under practical operating constraints. This suggests that the future cooling market will include software that optimizes thermal systems across changing AI workloads, rather than relying on static operating settings.
The next investment cycle will be shaped by the need to support compute at campus scale. Direct-to-chip systems are likely to lead near-term deployments because they align with current server roadmaps. Immersion cooling can gain share in specialized high-density facilities. Rear door systems can support retrofit demand. Spray and jet cooling can become more relevant as chip-level heat flux increases and packaging evolves.