Global Data Centre Liquid Cooling Market Outlook to 2032

Executive Summary

Data-centre liquid cooling has crossed from niche to standard-of-care for AI compute. The global market doubled in 2025 to approximately US$3 billion (Dell'Oro October 2025) and grows to US$12 billion by 2032 at 22% CAGR, anchored by NVIDIA GB200 NVL72 deployment thermal density and hyperscaler GB300 NVL72 cluster scale-up.

Three forces drive the trajectory. NVIDIA's GB200 NVL72 specifies 120 kW per rack at 1.36 metric tons fully liquid-cooled; 2025 shipments reached 25,000-35,000 cabinets (revised down from initial 50,000-80,000 forecast). HPE delivered its first GB200 NVL72 in February 2025; AWS EC2 P6e-GB300 UltraServers reached general availability December 2, 2025; Microsoft Azure commissioned its first large-scale GB300 NVL72 cluster for OpenAI workloads in October 2025. And vendor consolidation accelerated: Vertiv-CoolTera (December 2023), Schneider-Motivair (October 2024), Daikin Applied-DDC Solutions (August 2025), Daikin Applied-Chilldyne (November 4, 2025).

Market Overview

Data-centre liquid cooling captures rear-door heat exchangers, in-row chilled-water systems, direct-to-chip cooling, and immersion cooling. The category is distinct from traditional CRAC and CRAH air cooling and is increasingly the standard architecture for AI compute racks above approximately 30 kW per rack.

The category sits at the intersection of three forces. NVIDIA's GB200, GB300, and Rubin generation roadmap drives thermal density growth that air cooling cannot economically address; rack power density rising from approximately 10-15 kW (legacy enterprise) to 50-120 kW (current AI) toward more than 200 kW (next-generation). Hyperscaler AI capex is the demand-side anchor: Microsoft, Amazon, Google, Meta, and Oracle deploy at scale through 2027. And manufacturer consolidation is reshaping vendor architecture, with the top-5 vendors capturing 35% share by 2025 and the M&A pace accelerating into 2026.

Geopolitically, liquid cooling demand concentrates where AI compute capacity concentrates: US (Microsoft Azure, AWS, Google Cloud, Meta data-centre regions), East Asia (Korean and Japanese hyperscaler facilities), and emerging Middle East deployments anchored by Stargate UAE and HUMAIN.

Market Size and Growth Outlook

The 2025 doubling reflects the GB200 NVL72 deployment ramp at hyperscaler scale. From 2026-28 growth moderates to 25-50% as GB300 deployments compound and Rubin-generation thermal-density requirements drive next-wave capex. Beyond 2030, growth moderates to approximately 12-16% as the AI rack-density curve plateaus and the easy retrofit demand saturates.

Market Segmentation

By Cooling Technology

Direct-to-chip cold plate at 48% is the dominant architecture for GB200 and GB300 deployments. Rear-door heat exchangers at 22% serve hybrid deployments where existing air cooling extends into liquid-augmented designs. Two-phase direct-to-chip and immersion cooling at 16% combined are the fastest-growing technologies as densities of more than 200 kW per rack arrive in 2027-28 with Rubin generation.

By Buyer Cohort

Hyperscalers at 56% dominate the buyer cohort. AI-native clouds at 18% concentrate on GB200 NVL72 and GB300 deployment with high-density direct-to-chip cooling architecture (CoreWeave's deployment uses 85% liquid, 15% air). Colocation at 14% scales as Equinix, Digital Realty, and Iron Mountain retrofit existing facilities and build new AI-ready capacity.

By Region

The US at 56% reflects hyperscaler concentration and AI-native cloud deployment scale. China and East Asia includes Korean and Japanese hyperscaler-equivalent facilities. Middle East at 6% expands rapidly through 2030 as Stargate UAE and HUMAIN scale. The fastest-growing regional segment is the Middle East at approximately 50% CAGR through 2028.

Trends and Developments

NVIDIA GB200 NVL72 thermal density drives architecture shift

NVIDIA's GB200 NVL72 specifies 120 kW per rack at 1.36 metric tons fully liquid-cooled. 2025 cabinet shipments reached 25,000-35,000 (revised down from initial 50,000-80,000 forecast on TSMC CoWoS-L and HBM3E supply constraints). HPE delivered its first GB200 NVL72 in February 2025; CoreWeave's GB200 order (US$2.3 billion for 2025 delivery) deployed with 85% liquid, 15% air. The GB300 generation (December 2025 production) extends the thermal density trajectory; the Rubin generation in 2026-27 pushes density toward more than 200 kW per rack.

Hyperscaler GB300 deployment cluster

Microsoft Azure commissioned its first large-scale GB300 NVL72 cluster for OpenAI workloads in October 2025. AWS EC2 P6e-GB300 UltraServers reached general availability December 2, 2025. Google Cloud, Meta, and Oracle Cloud scale parallel deployments. Each hyperscaler deployment locks multi-year liquid cooling demand at scaled vendor capacity, anchoring the category's structural growth.

Vendor M&A consolidation accelerates

Vertiv acquired CoolTera (closed December 2023). Schneider acquired Motivair (October 2024). Daikin Applied acquired DDC Solutions (August 2025) and Chilldyne (November 4, 2025). Top-5 vendor share reached 35% in 2025 per Dell'Oro. Concentration is structurally increasing as hyperscaler procurement favours scaled multi-modal vendors with direct-to-chip, rear-door, CDU, and immersion capability.

Other relevant developments include immersion cooling commercial scale at hyperscaler pilots, direct-to-chip second-generation cold plate designs reducing flow rate per kW, two-phase architecture entering volume deployment in 2026, and the broader AI-data-centre power-and-cooling supply chain (covered in our Sovereign AI Infrastructure outlook) compounding capacity-allocation pressure.

Competitive Landscape

The category is moderately concentrated: top-5 control 35% of 2025 share. Schneider Electric leads post the Motivair acquisition, anchored by EcoStruxure data-centre infrastructure scale. Vertiv (post-CoolTera) is the closest competitor, with stronger AI-native cloud customer base. Rittal, Stulz, and Boyd round out the top-5 with specialised cooling-equipment heritage. Daikin Applied's two 2025 acquisitions (DDC Solutions August, Chilldyne November) signal accelerating consolidation. CoolIT Systems anchors mid-market direct-to-chip deployment. Asetek, Wiwynn, Submer, GRC, and emerging immersion specialists capture niche segments.

Key Policies and Regulatory Environment

EU Energy Efficiency Directive (recast)

The recast EU Energy Efficiency Directive (in force October 2023, transposition through 2025-26) requires annual data-centre energy efficiency reporting for facilities above 100 kW in the EU. Liquid cooling supports meeting Power Usage Effectiveness (PUE) targets approaching 1.1 versus air-cooled approaching 1.4-1.5. The framework creates structural demand-side underwriting for liquid cooling in EU jurisdictions.

US DOE Better Climate Challenge and state-level frameworks

US DOE Better Climate Challenge and state-level data-centre energy frameworks (notably Virginia, Texas, Oregon, Washington) drive PUE and water-usage-efficiency targets. California's emerging data-centre legislation tightens PUE benchmarks for new facilities. The frameworks support liquid cooling adoption in markets where new data-centre capacity is planning-permission-constrained.

Other relevant frameworks include China's PUE caps for new data centres in tier-1 cities (target 1.3, tier-1 facilities capped at 1.25), Singapore's IMDA Green Data Centre Roadmap, India's Data Centre Policy 2025, and emerging Middle East frameworks tied to Stargate UAE and HUMAIN deployments.

Future Outlook

The data-centre liquid cooling market in 2032 reaches approximately US$12 billion in annual value. Direct-to-chip captures approximately 52% of category by 2032 (versus 48% in 2025). Two-phase direct-to-chip and immersion cooling expand from 16% to 26% as Rubin-generation thermal density forces architecture shift. Rear-door heat exchangers compress from 22% to 14% as hybrid retrofit demand saturates.

The competitive landscape consolidates further. We expect 2-3 additional acquisition events among smaller liquid cooling vendors by 2027 as hyperscaler procurement favours scaled multi-modal capability. Top-5 share rises from 35% (2025) to approximately 48% by 2032. Daikin Applied, Schneider, and Vertiv emerge as the integrated multi-modal leaders.

The biggest risk is an AI capex pause through 2027-28 that compresses GB300 and Rubin deployment demand. The leading indicator is hyperscaler quarterly capex disclosures and NVIDIA cabinet shipment trajectory through 2026-27.