Global data center liquid cooling revenue is projected to rise from USD 3.9-5.7 billion in 2026 to USD 18.6-29.2 billion by 2033, equating to a 25.1-31.5% CAGR across the forecast window, with cold-plate direct-to-chip holding over 55% share in 2026 and immersion cooling the fastest-growing sub-segment [S3][S4][S5].
AI training clusters and hyperscale GPU fleets are the load-bearing demand driver: rack densities above 30 kW, and in some AI halls approaching 100 kW, are pushing operators off air cooling and onto liquid loops as chip TDPs cross 700W per device and power-use targets tighten [S1][S5].
Forecast Range, 2026-2030: Three Models, Same Direction
Three independently published 2026 forecasts bracket the demand curve for data center liquid cooling. Persistence Market Research sizes the market at USD 5.7 billion in 2026, scaling to USD 29.2 billion by 2033 at a 26.4% CAGR, with cold-plate systems at USD 3.1 billion in 2026 and large data centers over 10,000 sq. ft. capturing more than 64% share [S3]. MarketsandMarkets projects a steeper curve, from USD 4.07 billion in 2026 to USD 27.65 billion by 2033 at a 31.5% CAGR, citing AI, big data, and cloud workloads as the load-bearing factors [S4]. Stats N Data is the most conservative, starting at USD 3.9 billion in 2026 and reaching USD 18.6 billion by 2033 at a 25.1% CAGR, with direct-to-chip as the largest sub-segment and immersion gaining the fastest momentum in selected high-density environments [S5]. The narrower worldwide liquid coolers market, which bundles AIO PC units with data center hardware, is forecast at USD 8.14 billion in 2026 and USD 13.82 billion in 2030 at a 15.5% CAGR through 2032 [S1].
Side-by-side, the three data-center-only forecasts share a load-bearing assumption: AI training capacity additions in 2026-2028 force a step-change in cooling architecture, and air cannot reach the PUE band hyperscalers target. Cold-plate (direct-to-chip) absorbs the bulk of 2026 spending; immersion is the secondary wave that accelerates after 2027 as standalone AI campuses come online [S3][S5].
Technology Mix and the 2030 Component Sizing
Cold-plate direct-to-chip holds more than 55% of 2026 revenue at over USD 3.1 billion, driven by modular integration with existing server chassis and lower retrofit risk on operating aisles [S3]. Immersion cooling is the fastest-growing sub-segment, with cited 80% higher energy efficiency versus reference air setups and PUE of 1.02-1.03 in single-phase dielectric deployments, which fits defense, financial, and AI-training workloads where compactness and thermal headroom dominate [S3]. Spray liquid cooling rounds out the three, sized for niche high-density retrofits but materially smaller in 2026 revenue [S3].
On the component side, Guohai Securities (research note dated August 2026) projects the liquid-cooled optical module market growing from USD 2 billion to USD 8 billion at a 42% CAGR, and the CDU liquid cooling pump market reaching USD 6.6 billion by 2030, which is a useful order-of-magnitude check on how pump, manifold, and Coolant Distribution Unit (CDU) spend tracks the data center cooling total [S2]. Services attached to the installed base (installation, maintenance, optimization) are growing at a 27.8% CAGR, faster than the hardware side, which signals a 2027-2030 shift from greenfield build to lifecycle support contracts [S3].
Regional Split: North America Leads, China Accelerates, Germany Sets the Engineering Bar
North America leads with more than 36% share in 2026 at over USD 2.1 billion, on track to exceed USD 8.7 billion by 2033, concentrated in Virginia, Texas, Arizona, and Ohio where power availability and network density support AI-ready hyperscale halls [S3][S5]. The US liquid cooling market alone sits near USD 1.1 billion in 2026 and is forecast above USD 4.5 billion by 2033, with direct-to-chip dominating cloud campuses and immersion gaining traction in defense and financial verticals [S5].
China is the second engine, with 2026 revenue near USD 680 million and a path toward USD 2.8 billion by 2033, supported by domestic AI infrastructure, coastal cluster buildouts, and procurement localization that favors Chinese engineering firms on competitive bids [S5]. Germany, while smaller in absolute terms at roughly USD 240 million in 2026 and USD 930 million by 2033, is the European reference for closed-loop and rear-door heat exchanger retrofits driven by Frankfurt, Munich, and Berlin colocation, with operators prioritizing water use and waste heat recovery to meet sustainability mandates [S5].
What It Is For, and Where It Stops Paying Off
Liquid cooling is for operators running racks above 30 kW, training clusters where chip TDP exceeds 700W, and any site targeting PUE under 1.1 without paying the air-side economizer premium in hot or humid climates [S1][S3][S5]. It is the practical operating requirement once densities approach 100 kW per rack, which now occurs in flagship AI campuses, not a speculative upgrade [S5]. The technology stops paying off where rack power stays under 15 kW, where facility footprint is unconstrained, where water is restricted for direct use, or where the existing chiller plant already meets a 1.3 PUE target cheaply.
Decision rule: if the 2026-2028 build plan adds more than 5 MW of GPU capacity per hall and the local utility caps incremental power, the air-cooled option is no longer competitive on a $/MW basis once chilling, fans, and floor space are fully loaded. If a site is retrofitting a legacy raised-floor hall below 10 kW per rack, rear-door heat exchangers and partial cold-plate loops cover the gap at a fraction of full immersion conversion cost [S5].
Comparison of Main Liquid Cooling Options on 2026 Procurement Criteria
Three options dominate 2026 procurement. Cold-plate direct-to-chip: over 55% of 2026 share at USD 3.1 billion, modular, lowest retrofit risk, PUE 1.1-1.2 typical, fastest to deploy on existing aisles [S3]. Immersion (single-phase dielectric): fastest-growing sub-segment, cited 80% energy efficiency improvement over reference air setups, PUE 1.02-1.03, highest facility rework cost and longest payback on existing builds [S3]. Spray liquid cooling: niche, fits high-density retrofits where neither cold-plate nor full immersion fits the rack form factor, smaller 2026 revenue, used where space constraints rule out dielectric baths [S3].
On decision criteria, cold-plate wins on time-to-deploy and $/kW removed; immersion wins on density per square foot and steady-state PUE; spray wins only where the other two are mechanically blocked. Services (27.8% CAGR) are the layer most procurement teams underweight in 2026, but the lifecycle support overhang is the revenue line that scales fastest through 2030 [S3].
Standards, Limits, and Sourcing Discipline
Liquid cooling spend is increasingly tied to power efficiency reporting, with operators targeting PUE 1.02-1.2 in cold-plate and immersion deployments versus legacy air baselines [S3]. Liquid cooling is cited as cutting cooling energy consumption by approximately 30% or more in high-density AI data centers compared with air-cooled setups, a figure that scales with rack density and ambient wet-bulb [S1]. Rack power density above 30 kW, and approaching 100 kW in flagship AI installations, is the operational threshold above which liquid becomes a requirement rather than an option [S5].
Sourcing discipline for 2026-2030 means treating CDU pumps, manifolds, and cold plates as separate lines, since the pump and CDU market alone is sized toward USD 6.6 billion by 2030 and the optical module side toward USD 8 billion, and pricing power diverges by component [S2]. A related sourcing map for liquid cooling tiers, spec gates, and manufacturer segmentation is in the liquid cooling market 2026 tier breakdown, which gives a working procurement view of who supplies cold plates, CDUs, and immersion fluids. For the underlying compute load driving the cooling curve, the AI accelerator 2026 spec gate maps H100, MI300X, and Gaudi 3 thermal envelopes against the rack power figures cited above.
What the 2026-2030 Curve Does Not Resolve
None of the three 2026 forecasts resolves the split between single-phase and two-phase immersion at the deployment level, and the wider liquid coolers total is partially obscured by bundling AIO PC coolers (48.5% of 2025 share) with data-center hardware, which inflates the headline number relative to a facility-grade line item [S1]. Standards coverage is implicit rather than named in the cited research, so any safety-critical or flammability claim tied to a specific dielectric fluid should be checked against the supplier's UL/IEC certification rather than the market sizing report. The 30% cooling-energy reduction figure is an industry-analysis range, not a guaranteed site result, and actual savings depend on ambient conditions, heat-recovery utilization, and chiller plant efficiency [S1].
Two signals to track through 2027: the CDU pump and cold-plate pricing trend, since component-level CAGRs outpace the hardware total and will define the bill of materials on every greenfield AI hall [S2]; and the services-to-hardware revenue ratio, since the 27.8% services CAGR is the clearest leading indicator that liquid cooling is shifting from a build phase to a lifecycle phase [S3].
Spec-level background on the components involved: pressure transmitter, flow meter, and industrial valve.