Liquid and Immersion Cooling Market Outlook 2026–2034: Market Size, Growth Drivers, Key Players, Strategic Developments & Adoption Forecast for Direct-to-Chip Cold Plates, Coolant Distribution Units, Rear-Door and Immersion Systems, Warm-Water Operation and AI Rack Densities, A Meticulous Next™ Foresight Brief

This Meticulous Next™ brief examines how liquid and immersion cooling, from direct-to-chip cold plates and coolant distribution units to rear-door heat exchangers, single-phase and two-phase immersion and the warm-water plant behind them, becomes the standard thermal architecture for AI data centers over the next eight years. The shift begins with the chip and ends at the cooling tower, and each step in between is being redesigned.

The starting point is rack density. A GB200 NVL72 rack draws about 120 to 132 kW, against roughly 8 kW for an average enterprise rack. At that density air can no longer carry the heat away, so liquid cooling has moved from an option for a few halls to a requirement for every AI hall. That requirement is what the rest of the market now responds to.

The chip roadmap then removes the fans altogether. NVIDIA presented the Vera Rubin platform at CES on 5 January 2026 as 100% liquid-cooled, running on 45°C inlet water, with about twice the power of Grace Blackwell and 3.5 times the training performance. At 45°C, the company says, a data center needs no chiller. The claim sent HVAC shares lower the next morning, and LiquidStack qualified it, noting that halls running only Vera Rubin may still need chillers for secondary-loop temperatures. Either way, the water temperature is now set by the chip vendor, and the plant must follow.

Because the architecture is settled, capital has moved quickly. Ecolab agreed in March 2026 to buy CoolIT Systems for about USD 4.75 billion and closed on 2 July 2026, with CoolIT sales up more than 100% year to date. Trane completed its acquisition of LiquidStack on 3 March 2026 after buying Stellar Energy for USD 553.4 million in February. Schneider Electric took 75% of Motivair for about USD 850 million in February 2025, and LITEON invested about USD 176 million for roughly 25% of DCX on 3 September 2026. ZutaCore raised USD 100 million in June 2026 and Iceotope USD 26 million in May. Within eighteen months the leading cold-plate and CDU makers have passed into industrial groups, which brings scale and service but narrows the supplier base as demand doubles.

The deployments confirm the direction. Vertiv reported a direct-to-chip installation supporting 30 MW of AI workloads at the CLDC Mumbai 01 site of CapitaLand on 16 October 2026, and its second-quarter 2026 revenue reached USD 3.3 billion, up 24%. Cold-plate suppliers began shipping Vera Rubin modules in the third quarter of 2026. Liquid cooling is therefore no longer a pilot technology but a utility-scale one, and the questions that remain are about pace and plant rather than principle.

The brief maps the technology, its indicative market size and forecast, the factors behind its growth, the developments of the last 24 months, the key players operating in the space and the adoption trajectory to 2034.

It is a focused 108-page decision brief for data center operators and hyperscalers, server and rack OEMs, cooling equipment and component makers, building-services and HVAC groups, utilities and investors. It presents an indicative trajectory rather than a segmented market model. Its purpose is to identify which cooling architectures win at each density, how warm-water operation changes the plant, and who captures value as the heat moves from air to liquid.

Liquid cooling moves heat from the chip in a fluid rather than in air. Direct-to-chip cold plates sit on GPUs, CPUs and memory and carry water-glycol through a secondary loop. Coolant distribution units separate that loop from the facility water, pump it and hold its temperature. Manifolds and quick disconnects link the racks, and rear-door heat exchangers capture whatever heat still reaches the air. Together these parts form a path from silicon to plant that did not exist in the air-cooled hall.

Immersion takes the same idea further by submerging the server in dielectric fluid. Single-phase systems circulate the fluid to a heat exchanger, while two-phase systems boil and condense it inside the tank. Immersion removes nearly all the heat without cold plates, but it needs purpose-built servers and fluid handling. For that reason it has settled into high-density, edge and retrofit niches, while cold plates have taken the mainstream of rack-scale AI.

The chip decides which of these architectures is required. A GB200 NVL72 rack at 120 to 132 kW already relies on direct-to-chip cooling, and Vera Rubin extends liquid to the network cards, busbars and optical transceivers so that the rack has no fans at all. The chassis assembles in about five minutes with no cables or water pipes, against 43 cables, six pipes and about two hours for Grace Blackwell. At the volumes now being shipped, assembly time of that order is a production variable rather than a convenience, and it favours designs that arrive liquid-ready from the factory.

The water temperature in turn decides the plant. At 45°C inlet, heat can be rejected to dry coolers in most climates for most of the year, which is why NVIDIA says no chiller is needed and why LiquidStack expects wider free cooling even where chillers remain for secondary loops. The consequence is less compressor plant, less water consumption and a heat stream warm enough for district reuse, which changes what a data center buys from its HVAC suppliers.

The constraint, then, is the transition itself rather than the technology. Operators must bring water to the rack, specify coolant distribution units at the megawatt scale, where DCX now offers an 8 MW unit rated for 45°C, and manage fluids, leaks and commissioning with trades that were built for air. Consolidation into Ecolab, Trane, Schneider and Vertiv brings scale and service to that task, but it also concentrates supply at the moment demand is doubling, and that balance will shape prices and lead times through 2028.

The liquid and immersion cooling market, covering direct-to-chip cold plates and loops, coolant distribution units, manifolds and quick disconnects, rear-door heat exchangers, immersion systems and fluids, and installation and service for data centers, and excluding chillers, air handling, dry coolers and the servers themselves, is valued at USD 6.5–8.5 billion in 2026 and is projected to reach USD 45–60 billion by 2034, at a CAGR of 27–29%.

The 2026 base rests on rack-scale AI shipments that NVIDIA partners expect to double this year, supplied with cold plates and CDUs by CoolIT, Motivair, Vertiv, Boyd, Cooler Master, AVC, Auras and DCX, with immersion from LiquidStack, Submer, GRC and Iceotope, and with deployments such as the 30 MW Mumbai site. From that base, growth to 2034 follows three steps: fully liquid-cooled racks become the default for new AI capacity, 600 kW rack designs arrive with the Rubin Ultra generation, and liquid-ready specifications spread to enterprise and colocation halls, after which service revenue grows as installed loops age.

The forecast is sensitive to AI capital spending, to the pace at which warm-water plant replaces chilled water and to supplier capacity after consolidation. Regionally, the United States leads on deployments and suppliers, Taiwan on cold-plate manufacturing, Europe on heat reuse and immersion, and India, the Gulf and Southeast Asia on new liquid-first builds.

The base case assumes rack-scale AI shipments doubling in 2026 and growing through 2028, Vera Rubin and successor racks fully liquid-cooled, and warm-water plant adopted in most new AI builds by 2029. An accelerated case adds 600 kW racks at volume by 2028 and liquid-ready mandates in colocation, lifting the 2034 value to the top of the range. A delayed case assumes a slowdown in AI capital spending and supplier capacity limits after consolidation, holding growth at the bottom of the range.

Growth drivers

• Rack density: 120–132 kW GB200 NVL72 racks against 8 kW enterprise racks; 600 kW designs announced.
• Fanless, fully liquid-cooled platforms: Vera Rubin on 45°C water, extending liquid to NICs, busbars and optics.
• Capital and consolidation: Ecolab–CoolIT USD 4.75 billion; Trane–LiquidStack and Stellar Energy; Schneider–Motivair USD 850 million; LITEON–DCX.
• Scale deployments: Vertiv 30 MW Mumbai; Vertiv USD 3.3 billion quarterly revenue, up 24%.

Enablers

• Warm-water operation enabling dry-cooler free cooling and heat reuse.
• Megawatt-class CDUs (DCX 8 MW at 45°C) and standardised manifolds and quick disconnects.
• Cold-plate supply chain in Taiwan and the US (Cooler Master, AVC, Boyd, Auras) shipping Rubin modules from Q3 2026.
• Venture funding: ZutaCore USD 100 million; Iceotope USD 26 million; Accelsius Series B.

Restraints and barriers

• Retrofit burden: bringing water to racks, leak management and trades built for air.
• Supplier concentration after consolidation at the moment demand doubles.
• Chiller debate: secondary-loop temperatures may still require chillers in Rubin-only halls.
• Dependence on AI capital spending cycles.

Five converging forces will determine how fast liquid and immersion cooling scales.

(1) Rack density set by chip roadmaps. (2) Warm-water operation and plant redesign. (3) Supplier consolidation and capacity. (4) Operator readiness and skills. (5) Energy, water and heat-reuse policy. The brief assesses each force for direction, speed and confidence.

The key players operating in liquid and immersion cooling include Vertiv Holdings Co, Schneider Electric SE (Motivair), Ecolab Inc. (CoolIT Systems), Trane Technologies plc (LiquidStack, Stellar Energy), nVent Electric plc, Boyd Corporation, Cooler Master Technology Inc., Asia Vital Components Co., Ltd., Auras Technology Co., Ltd., Delta Electronics, Inc., LITEON Technology Corporation and DCX Liquid Cooling Systems, Flex Ltd. (JetCool), ZutaCore Ltd., Iceotope Technologies Limited, Accelsius, LLC, Submer Technologies S.L., Green Revolution Cooling, Inc., Asetek A/S, Chilldyne, Inc., Corintis SA, Jentech Precision Industrial Co., Ltd., Supermicro, Inc., Dell Technologies Inc., Hewlett Packard Enterprise Company, Lenovo Group Limited, Foxconn (Hon Hai Precision Industry Co., Ltd.), Wiwynn Corporation, Quanta Computer Inc., NVIDIA Corporation, Advanced Micro Devices, Inc., Johnson Controls International plc, Carrier Global Corporation, Daikin Industries, Ltd., Munters Group AB, Alfa Laval AB, Danfoss A/S, Microsoft Corporation, Alphabet Inc., Amazon.com, Inc., Meta Platforms, Inc., Oracle Corporation, Equinix, Inc., Digital Realty Trust, Inc., CapitaLand Investment Limited, CoreWeave, Inc. and ASHRAE.

The brief profiles representative players in each archetype and assesses which are positioned to own the thermal path from chip to atmosphere.

The competitive landscape is forming around six archetypes. Integrated cooling groups. Cold-plate and CDU component makers. Immersion specialists. Rack and server OEMs. Plant, HVAC and heat-reuse suppliers. Operators, chip vendors and standards bodies. Competitive intensity is high in 2026 and is expected to remain high as consolidation completes and rack-scale shipments double through 2028.

Where value migrates.

In 2026 value sits in cold plates and CDUs for rack-scale AI, in acquisitions and in first 30 MW deployments. By 2029 it moves to fully liquid-cooled racks on warm water, megawatt CDUs, chiller-light plant and heat reuse. By 2034 it settles in integrated cooling groups, rack OEMs with liquid designs, warm-water plant suppliers and service providers. Air-only designs for high-density capacity lose ground.

The signals point the same way. Density is decisive, with 120 to 132 kW racks today and 600 kW designs announced. The platform is fixed, with 45°C water and fanless racks shipping from 2026. Capital is large, with USD 4.75 billion, USD 850 million, USD 553.4 million and USD 176 million transactions and USD 126 million of venture rounds in 2026. Deployments have reached 30 MW at a single site, and revenue follows, with Vertiv at USD 3.3 billion a quarter, up 24%, and CoolIT sales doubling. Each signal confirms that the architecture is decided and that the market is now about volume and pace.

Research and patent activity reflects the next bottlenecks rather than the last ones: microchannel and two-phase cold plates, megawatt CDUs, dielectric fluids, leak detection and heat-reuse integration. The brief therefore tracks four indicators that will show how fast the pace quickens: Vera Rubin shipment volumes, 600 kW rack deployments, chiller-free plant designs in new AI builds and heat-reuse contracts.

Regionally, the United States leads on deployments and suppliers, Taiwan on cold-plate manufacturing, Europe on heat reuse and immersion, and India, the Gulf and Southeast Asia on liquid-first new builds. Hyperscale AI, GPU cloud and colocation for AI tenants are the first demand pools, and enterprise halls follow as liquid-ready specifications become standard.