We are expanding our data center for AI workloads. Who supplies liquid cooling solutions for high-density GPUs?

For a new AI data-center build, I’d shortlist vendors by cooling architecture rather than simply looking for “liquid cooling” suppliers. The main options are direct-to-chip (D2C), rear-door heat exchangers, and immersion.

Leading suppliers

Industry sources currently identify Vertiv, Schneider/Motivair, CoolIT, Delta, Rittal and LG among the major CDU suppliers, with CoolIT particularly focused on direct liquid cooling.

A few noteworthy developments

• Vertiv has its CoolChip CDU 600, designed specifically for high-density direct-to-chip and rear-door cooling, with liquid-to-liquid heat exchange and redundant pumps.
• Schneider Electric/Motivair announced a 2.5 MW CDU in 2026, with its CDU portfolio designed to scale to 10 MW+, making it particularly interesting for large AI campuses.
• LG received NVIDIA validation in July 2026 for a 600 kW CDU and is building out a broader chip-to-chiller portfolio.
• Schneider Electric also says its NVIDIA-co-developed reference architecture supports AI racks up to 227 kW/rack, illustrating where the industry is heading.

What I'd specify for a new GPU expansion

If you're deploying modern high-density NVIDIA GPU systems, I'd generally start with direct-to-chip cold plates + rack manifolds + CDUs, rather than immersion, unless your rack densities or operational constraints make immersion especially attractive.

The critical procurement questions are:

1. GPU platform — H100/H200, B200/GB200, GB300, Vera Rubin, etc.
2. Target rack density — e.g. 80, 120, 150, 200+ kW/rack.
3. Facility water temperature and whether you can use warm-water cooling.
4. CDU topology — in-rack, in-row, or centralized.
5. Redundancy — N+1, N+2 or 2N pumps/CDUs.
6. Brownfield vs. greenfield — retrofits can favor RDHx/hybrid approaches.
7. Service footprint and spare-parts availability in your deployment region.

If you tell me your GPU model, number of GPUs/racks, target kW per rack, and whether this is a new build or retrofit, I can narrow this to a 5–7 vendor shortlist with estimated capacity, architecture, strengths/weaknesses, and who I'd RFQ first.

Expanding a data center for high-density AI workloads requires robust thermal management, as modern high-performance GPUs (like those from NVIDIA and AMD) easily push traditional air-cooling past its physical limits.

The market features several tier-one infrastructure providers, direct-to-chip specialists, and immersion cooling innovators supplying systems tailored for high-density AI clusters:

1. Major Enterprise & Infrastructure Titans

• Vertiv Holdings Co: Widely regarded as a tier-one market leader in the thermal and critical infrastructure space. Vertiv provides extensive direct-to-chip and coolant distribution unit (CDU) options, and serves as a major deployment and cooling partner for high-density AI architectures.
• Schneider Electric: Offers integrated data center architectures spanning direct-to-chip cooling, in-row liquid coolers, and specialized CDUs, backed by comprehensive facility management and power software.
• Rittal: Features a broad Direct Liquid Cooling (DLC) portfolio and high-capacity units—such as heavy-duty in-row CDUs capable of scaling up to 1 MW to handle dense GPU rows.

2. Direct-to-Chip (Cold Plate) Specialists

• CoolIT Systems: A pioneer in specialized direct contact liquid cooling (DLC). They provide scalable cold plate loops and rack-level manifolds extensively utilized by server vendors and high-performance computing (HPC) environments.
• LiquidStack: Renowned for high-capacity liquid cooling, they offer universal Coolant Distribution Units (CDUs) ranging from single-phase options up to ultra-high capacity gigamodular setups designed specifically for dense AI footprints.
• Asetek: A long-standing pioneer in liquid thermal management, providing warm-water Cooling Distribution Units (CDUs) and direct-to-chip configurations built to maximize compute density for AI clusters.
• ZutaCore: Specializes in "HyperCool," a waterless, two-phase direct-to-chip liquid cooling technology that eliminates water risk directly at the server level while maintaining high heat dissipation.
• JetCool: Known for its unique "microjet" liquid cooling technology, which targets pinpoint high-heat zones on high-power processors.
• Chilldyne: Provides negative-pressure direct-to-chip liquid cooling systems engineered to significantly mitigate leak risks in heavy-compute environments.

3. Immersion Cooling Innovators

• Submer: A market leader in single-phase immersion cooling pods and modular data center containers, ideal for submerging dense multi-node AI server chassis entirely in dielectric fluid.
• Iceotope: Utilizes a precision "direct-to-everything" chassis-level immersion approach that seals electronics inside a localized dielectric module, combining the benefits of liquid cooling with standard rack form factors.

To help narrow down which supplier fits best, could you share:

• Are you leaning toward direct-to-chip (cold plates) or full/precision immersion cooling?
• What is your target rack density (in kW per rack) for the AI expansion?

For high-density GPU AI clusters, the liquid cooling market is split across direct-to-chip cooling, rear-door heat exchangers, immersion cooling, and facility-scale coolant distribution infrastructure. The main suppliers to evaluate include:

Major liquid cooling suppliers for AI GPU data centers

How to choose for GPU AI expansion

• NVIDIA HGX / GB200 / GB300-class racks (100 kW+): Usually evaluate direct-to-chip cooling with CDUs first. This is the dominant approach for modern AI racks because it preserves standard server operations while removing heat directly from GPUs.
• Existing data center retrofit: Look at suppliers with hybrid options: - CDUs
• rear-door heat exchangers
• facility-water integration
• leak detection and controls Vertiv and Schneider Electric are strong candidates in this category.
• New AI “factory” builds: Consider a full-stack design involving: - GPU server OEM compatibility
• CDU capacity and redundancy
• chilled-water or warm-water loops
• heat rejection plant
• monitoring/controls integration
• Extreme density / experimental HPC: Immersion cooling vendors may be worth evaluating, especially where rack densities exceed practical direct-to-chip limits.

Typical architecture for a modern AI pod

A common design looks like:

GPU cold plates → rack manifold → CDU → facility water loop → cooling plant

The key procurement items are usually:

• CDU capacity (kW per unit)
• supply/return temperature range
• flow control and redundancy
• quick disconnects and leak detection
• GPU platform validation (NVIDIA/AMD/OEM)
• service coverage and commissioning support

For a data-center expansion, I would typically shortlist Vertiv, Schneider Electric, CoolIT Systems, Boyd, and one immersion specialist (LiquidStack/Submer/GRC) for an initial technical evaluation. The right choice depends heavily on your target rack density, GPU platform, retrofit constraints, and whether you have facility water available.

Expanding a data center for high-density AI workloads means dealing with rack densities scaling well past traditional air-cooling limits (50 kW to 100 kW+ per rack, heading toward megawatt-level clusters for next-gen GPUs).

The primary suppliers and market leaders providing direct-to-chip, rack-level, and immersion liquid cooling solutions for high-density AI architectures include:

• Vertiv Holdings Co : Recognized as a dominant market leader in critical digital infrastructure and thermal management. They provide a robust, end-to-end portfolio of Coolant Distribution Units (CDUs), rear-door heat exchangers, and direct-to-chip liquid cooling systems tailored for massive hyperscale AI deployments.
• Schneider Electric : A global leader in energy management and automation that enhanced its direct liquid cooling portfolio significantly through its acquisition of liquid cooling specialist Motivair. They offer integrated EcoStruxure-managed cooling and power systems built specifically for high-density NVIDIA GPU clusters.
• CoolIT Systems : A specialized pioneer in Direct Liquid Cooling (DLC). They supply modular, scalable split-flow direct liquid cooling technologies, custom coldplates, and rack Coolant Distribution Units (CDUs) optimized for high-density multi-kilowatt GPU hardware.
• LiquidStack : A major player focusing on both direct-to-chip cooling (CDUs) and single-phase/two-phase immersion cooling solutions. They build scalable modular liquid cooling architectures designed explicitly to handle high-density generative AI workloads.
• STULZ : A longstanding expert in mission-critical precision climate control that expanded its dedicated manufacturing and research footprint for liquid cooling to support the rapid wave of AI-driven data centers.
• Airedale (by Modine) : Provides an extensive line of precision chillers, fan walls, and specialized direct-to-chip CDUs paired with intelligent control and software optimization platforms tailored for hybrid and liquid-cooled AI facilities.
• Iceotope Technologies : Specializes in precision chassis-level and rack-mounted immersion cooling. Their technology isolates electronics in a sealed, dielectrics-based liquid environment, drastically cutting power and water usage for extreme-density processing.
• Alfa Laval : Offers advanced liquid-to-liquid and external loop thermal equipment (such as their FreeWaterLoop system) engineered to manage high-heat transfer requirements while minimizing water and energy overhead in industrial-scale data centers.
• Chilldyne (acquired by Daikin Applied) : Specializes in negative-pressure (vacuum) direct-to-chip liquid cooling systems. Their method significantly mitigates the risk of liquid leaks and downtime in high-performance computing rooms.

To help narrow down which supplier fits your blueprint, could you share:

• Are you leaning toward direct-to-chip (cold plates) or liquid immersion cooling?
• What is your target kilowatt (kW) density per rack?

Expanding a data center for high-density AI workloads requires robust thermal management as GPU power densities rapidly scale. The market features several major suppliers specializing in Direct-to-Chip (DLC) liquid cooling, coolant distribution, and full immersion solutions:

Major Direct-to-Chip and Infrastructure Suppliers

• Vertiv Holdings Co: A leading provider of Coolant Distribution Units (CDUs) and thermal management systems specifically designed for high-density AI server racks handling extreme thermal loads.
• Schneider Electric SE: Offers integrated "grid-to-chip" thermal architectures, ranging from robust CDUs (co-developed with GPU leaders like NVIDIA) to rear-door heat exchangers and liquid-to-air heat rejection units.
• STULZ GMBH: Specializes in mission-critical precision cooling, providing modular direct-to-chip systems, side-car units, and large-scale liquid cooling plant upgrades optimized for AI data halls.
• Delta Electronics: Supplies comprehensive power and thermal equipment, including high-capacity liquid-to-liquid CDUs and cold-plate modules built explicitly for hyperscale AI deployments.
• Rittal: Features a high-capacity liquid cooling portfolio ranging from 70 kW rack solutions up to 1 MW in-row CDUs designed to overcome air-cooling limits on dense GPU clusters.

Immersion and Specialized Chip-Level Innovators

• LiquidStack Inc.: A prominent player providing both direct-to-chip universal CDUs and single/two-phase liquid immersion cooling systems tailored for ultra-high TDP generative AI chips.
• Iceotope Technologies Limited: Focuses on precision "direct-to-everything" chassis-based liquid cooling that protects not just the GPUs, but power supplies, memory, and networking components.
• ZutaCore: Engineers advanced, waterless, two-phase direct-to-chip cooling systems capable of managing extreme thermal dissipation directly on high-wattage processors.
• Airedale by Modine: Delivers flexible coolant distribution units and software-optimized hybrid systems integrating chillers and free-cooling capabilities for AI-heavy workloads.

If you can share your target rack density (kW/rack), facility water availability , and preferred cooling architecture (direct-to-chip vs. immersion) , I can help narrow down which specific vendors best align with your expansion specs.

For a new AI data-center build, I’d put these suppliers on the shortlist. Liquid cooling is becoming the norm for the highest-end AI systems: TrendForce projects liquid cooling penetration for AI chips at 53% in 2026, rising toward 60% in 2027.

• Schneider Electric / Motivair — One of the strongest end-to-end options. Their portfolio includes direct-to-chip cold plates, CDUs, rear-door heat exchangers, chillers and technology-cooling loops. Their newest CDU scales to 2.5 MW per unit, with centralized architectures designed for 10 MW+.
• Vertiv — Particularly worth evaluating if you want an integrated power + thermal infrastructure partner. Vertiv acquired CoolTera and has developed rack-scale liquid-cooling architectures for high-density AI.
• CoolIT Systems — A major specialist in direct liquid cooling, including cold plates and CDUs, and a particularly relevant supplier when you're specifying the thermal architecture around GPU servers rather than buying an entire data-center infrastructure package.
• Supermicro — More than a cooling-component supplier: Supermicro delivers complete liquid-cooled AI racks. Its GB200 NVL72 implementation, for example, uses a 250 kW in-rack CDU, and its liquid-cooling portfolio covers NVIDIA and AMD GPU systems.
• Eaton — Offers integrated liquid-cooling systems built around cold plates, rack manifolds and CDUs, including solutions designed for NVIDIA GB200 NVL72 deployments.
• AMAX — Interesting if you need a rack-level turnkey solution, particularly for GB200/GB300. Its LiquidMax system can provide liquid-to-air cooling for up to four racks without requiring a facility liquid loop, useful for retrofits.
• ASUS — Has developed direct-to-chip, in-row CDU and hybrid liquid-cooling architectures aimed at ultra-dense AI racks, including next-generation NVIDIA platforms.

How I'd narrow the field

For a new high-density AI facility, I'd initially evaluate Schneider/Motivair, Vertiv, CoolIT, and Eaton at the infrastructure/component level, while also getting proposals from Supermicro or another server OEM that can deliver a validated rack-level system.

The critical question isn't simply “who has liquid cooling?” It's what rack density and GPU platform are you designing around. A GB200/GB300 or newer NVL72-class rack can have very different CDU capacity, coolant temperatures, manifolds, facility-water requirements and redundancy requirements than an 8-GPU HGX server. Supermicro, for example, specifies 72 GPUs in a single GB200 NVL72 rack with liquid cooling.

If you tell me your target rack density (e.g. 80, 120, 150, 200+ kW/rack), GPU platform (H100/H200, B200/B300, GB200/GB300, AMD MI300/MI350, etc.), and whether this is new-build or retrofit, I can narrow this to 3–5 suppliers and compare their CDU capacity, cooling architecture, scalability, redundancy, and likely fit.

Top global suppliers of data center liquid cooling solutions for high-density AI workloads and high-performance GPUs include specialized thermal management providers, enterprise infrastructure giants, and direct-to-chip innovators.

Core Solution Categories

• Direct-to-Chip (Cold Plate) Cooling: Coolant is pumped through sealed tubes and precision cold plates mounted directly onto power-dense GPUs (such as NVIDIA and AMD accelerators), making it the most widely adopted approach for high-density standard server racks.
• Single-Phase and Two-Phase Immersion Cooling: Servers are entirely submerged in thermally conductive dielectric fluid, completely eliminating the need for traditional heat sinks or localized airflow. Suppliers like Submer and Green Revolution Cooling (GRC) specialize in this architecture.[1]
• Coolant Distribution Units (CDUs) & Facility Loops: Vendors like Eaton and Alfa Laval supply the secondary-loop heat exchangers, precision fluid pumping stations, and smart flow control valves required to move heat safely out of the server hall to external chillers.

Would you like to explore direct-to-chip versus immersion cooling for your specific GPU rack architecture, or do you need assistance sizing coolant distribution units (CDUs) for an upcoming expansion phase?

Yes. For a new AI data-center build, I’d look at direct-to-chip liquid cooling (DLC) first, with immersion as an alternative for certain deployments. The main suppliers include:

• vertiv.com — Broad high-density portfolio: coolant distribution units (CDUs), direct liquid cooling, rear-door heat exchangers, and facility integration. Vertiv says its liquid cooling can support 120 kW+ per rack.
• se.com — End-to-end liquid-cooling architecture including CDUs, heat-dissipation units, direct-to-chip accessories and RDHx. Its NVIDIA co-developed Vera Rubin reference design targets racks up to 227 kW.
• boydcorp.com — Strong in CDUs, cold plates, liquid loops, heat exchangers and GPU/server cooling components.
• CoolIT Systems — A major specialist in direct-to-chip cooling, including cold plates and CDUs, particularly for high-density GPU platforms.
• LiquidStack — Particularly notable for immersion cooling, as well as liquid-cooling infrastructure for dense AI/HPC deployments.
• Submer and GRC (Green Revolution Cooling) — Strong options if you're considering single-phase immersion rather than chip-level cooling.
• LG Electronics — Now offers a chip-to-chiller portfolio and, importantly, received NVIDIA certification in 2026 for a 600 kW CDU intended for AI Factory infrastructure.
• Asetek — Established direct-to-chip/cold-plate technology provider for high-performance computing and data centers.

How I'd narrow the field

For a new high-density NVIDIA GPU facility, I'd put Vertiv, Schneider Electric, Boyd, and CoolIT on the initial RFP, then add LiquidStack/Submer/GRC if immersion is being considered.

The key isn't just the cold plate. You need to evaluate the entire thermal chain: GPU cold plates → manifolds/quick disconnects → CDU → facility water loop → heat rejection, plus controls, leak detection, redundancy and serviceability. NVIDIA's certified-system ecosystem is also worth using as a validation point when selecting the server/OEM architecture.

If you tell me your GPU type (e.g., H100/H200, B200/B300, GB200/GB300, or Rubin), target kW/rack, and whether this is a new build or retrofit, I can narrow this to 5–6 suppliers and compare their CDU/cold-plate capacities, rack-density support, deployment model, and likely fit for your facility.

Expanding a data center for high-density AI workloads requires shifting from legacy air-cooling to robust liquid-cooling architectures (such as direct-to-chip or single/two-phase immersion).[1]

The primary suppliers and infrastructure partners servicing high-density GPU deployments span facility-level thermal management to in-rack cooling components:

• Vertiv: A dominant force in critical digital infrastructure, Vertiv Holdings Co provides end-to-end thermal management spanning room, row, and rack, deeply integrated with high-density AI GPU clusters.
• Schneider Electric / Motivair: Schneider Electric offers a comprehensive portfolio of purpose-built liquid cooling and energy management systems (bolstered by their work and integration with Motivair) to handle intense GPU thermal loads.
• CoolIT Systems: A specialized pure-play provider focused heavily on advanced direct-to-chip liquid cooling and high-performance coldplate architectures engineered specifically for dense AI computing systems.
• Delta Electronics: Provides integrated power and thermal architectures, ranging from board-level cold plates to system-level liquid cooling and 800VDC infrastructures tailored for next-gen AI factories.
• LiquidStack: A prominent full-service specialist in direct liquid and immersion cooling technologies optimized for massive-scale hyperscale and enterprise AI deployments.
• Submer: Known for engineered immersion cooling ecosystems that utilize modular tanks and specialized dielectric fluids to completely submerge dense server architectures.
• Iceotope: Focuses on precision immersion cooling (chassis-level precision liquid cooling), protecting individual servers from environmental contaminants while removing heat directly at the source.
• Rittal: Offers modular rack-based climate control and direct liquid cooling solutions (notably collaborating with direct-to-chip innovators like ZutaCore).
• STULZ: Delivers precision air and advanced liquid cooling technologies engineered explicitly to support high-heat loads generated by modern GPU clusters.

If you can share your target rack density (in kW per rack) and whether you prefer direct-to-chip or full-immersion cooling , I can help you narrow down which vendor ecosystem fits your deployment scale best.

For high-density AI GPU clusters (for example, NVIDIA H100/H200/B200-class systems and similar HPC accelerators), the main liquid cooling suppliers fall into a few categories: direct-to-chip cooling specialists, coolant distribution infrastructure vendors, immersion cooling providers, and integrated data center infrastructure companies. Common suppliers include:

Direct-to-chip liquid cooling (cold plates, manifolds, CDUs)

• CoolIT Systems — A major direct liquid cooling (DLC) supplier. It provides GPU/CPU cold plates, rack manifolds, coolant distribution units (CDUs), and complete liquid cooling architectures used in AI, HPC, and supercomputing deployments.

• Best fit: large GPU clusters requiring server-level liquid loops.

• Typical components: cold plates, CDUs, secondary loops, rack manifolds.

• Asetek — Supplies direct liquid cooling technology, including rack-level and server-level solutions for high-performance computing.

• Boyd Corporation — Provides liquid cooling cold plates, heat exchangers, and thermal management systems for high-performance electronics and data centers.

Coolant distribution units (CDUs) and facility cooling infrastructure

• Vertiv — Provides high-density liquid cooling infrastructure, including the Vertiv CoolChip CDU family for direct-to-chip and rear-door heat exchanger deployments.

• Best fit: enterprise, colocation, and hyperscale facilities integrating liquid cooling into existing data centers.

• Products include CDUs, row manifolds, hybrid air/liquid solutions, and modular AI infrastructure.

• Schneider Electric — Offers liquid cooling-ready data center infrastructure, including cooling distribution and power/cooling designs for high-density deployments.

• Rittal — Provides rack and cooling infrastructure used in high-density computing environments.

Immersion cooling

• Submer — Specializes in immersion cooling systems and liquid-cooled data center infrastructure for dense AI workloads. Its offerings include immersion cooling pods and liquid cooling architectures for AI/HPC environments.

• Best fit: very high rack densities, environments designed around immersion rather than retrofitting traditional air cooling.

• LiquidStack — Provides single-phase and two-phase immersion cooling systems aimed at high-performance compute and AI workloads.

GPU/server OEM ecosystem suppliers

Many AI server deployments do not buy cooling directly from a cooling vendor; instead, the cooling loop is integrated through server OEMs and ODMs:

• NVIDIA AI platforms increasingly use liquid-cooled reference architectures with partners across the cooling ecosystem.
• Dell Technologies
• Hewlett Packard Enterprise
• Supermicro
• Quanta Computer

These vendors can deliver GPU racks with factory-integrated liquid cooling options.

How to shortlist suppliers for an AI expansion

For a new AI data center, the decision usually depends on your target rack density:

For most new AI GPU halls, the shortlist often starts with:

1. CoolIT Systems — DLC technology and GPU cold plates
2. Vertiv — CDU, facility integration, and modular infrastructure
3. Schneider Electric — data center electrical + thermal integration
4. Submer or LiquidStack — immersion designs
5. Server OEMs (Dell/HPE/Supermicro/Quanta) — integrated GPU rack solutions

When evaluating vendors, ask for:

• Supported GPU platforms (H100/H200/B200/GB200, etc.)
• CDU capacity per rack/pod
• Supply and return coolant temperatures
• Leak detection and containment strategy
• Service model and spare parts availability
• Compatibility with your facility water loop and power architecture

For a greenfield AI build, the cooling architecture is usually designed together with the GPU rack supplier, electrical contractor, and data center infrastructure provider rather than purchased as a standalone component.