Vertiv vs CoolIT vs Motivair: 12-Rack Cooling Gap [2026]

Every rack of Nvidia’s GB300 NVL72 now draws up to 142 kW, according to Nvidia’s own enterprise reference architecture documentation, and HPE’s QuickSpecs sheet for the same platform lists a peak electrical design power near 155 kW per rack. Air alone cannot pull that much heat out of a server cabinet anymore. That single fact has turned three companies most gamers and consumers have never heard of – Vertiv, CoolIT Systems, and Motivair (now majority-owned by Schneider Electric) – into some of the most important hardware vendors in the AI buildout. They build coolant distribution units, or CDUs, the pumps and heat exchangers that pipe water or dielectric fluid directly onto GPU and CPU cold plates instead of relying on fans and chilled air.

This comparison breaks down what each vendor actually sells, how their coolant distribution units stack up in kilowatts and rack counts, what the direct-to-chip liquid cooling market is worth in October 2026, and which vendor fits which kind of AI data center deployment – the latest entry in our ongoing hardware coverage. Because public pricing for enterprise CDUs is rarely published, we lean on verified capacity specs, rack-support numbers, and corporate deal data pulled directly from vendor datasheets, press releases, and market research firms rather than guessing at dollar figures nobody has disclosed.

Why Air Cooling Stopped Working for AI Racks

For two decades, data centers relied on computer room air handlers (CRAHs) and raised floors to push cold air past server fans. That approach tops out somewhere around 20 to 30 kW per rack before airflow physics simply can’t move enough heat fast enough. A single Nvidia GB300 NVL72 rack packs 72 Blackwell Ultra GPUs, nine NVLink switch trays with two NVSwitch ASICs apiece, and eight power shelves each capable of delivering 33 kW, according to Nvidia’s own enterprise reference architecture documentation. That design point sits at roughly five to seven times the ceiling of air cooling.

Direct-to-chip liquid cooling solves the problem by routing coolant through cold plates mounted straight onto the GPU and CPU dies, carrying heat away before it ever reaches the room’s air. The GB300 NVL72 reference design documented by HPE uses cold plates on both CPUs and GPUs, which is why Nvidia bakes rack-level and tray-level liquid-leak detection directly into the chassis. The three vendors in this comparison build the CDUs, the box that sits between the facility’s chilled-water loop and the rack’s internal coolant loop, acting as the actual engine of a liquid-cooled AI data center.

Vertiv’s own leadership frames this shift as permanent rather than cyclical. Scott Armul, Vertiv’s chief product and technology officer, said the data center industry “is continuing to rapidly evolve how it designs, builds, operates and services data centers, in response to the density and speed of deployment demands of AI factories,” in comments carried in Vertiv’s 2026 investor outlook. That framing – cooling as a core design constraint rather than an afterthought – is exactly why three separate companies now compete head-to-head on CDU capacity, rack-count support, and Nvidia compatibility claims rather than treating liquid cooling as a niche add-on.

Data Center Liquid Cooling Market Size in 2026

Market researchers don’t fully agree on the size of the data center liquid cooling market, partly because firms define the category differently (some lump in liquid-to-air CDUs, others count only liquid-to-liquid immersion systems). Grand View Research pegs the global market at $6.7 billion in 2025, growing to roughly $8.2 billion in 2026, and projects it will reach $29.5 billion by 2033 – a 20.1% compound annual growth rate across that window. A separate analysis cited by Dell’Oro Group puts the 2026 figure closer to $6.0 billion with an 18.2% CAGR running through 2035, and credits Vertiv with holding more than 11.3% of that market as of 2025.

Those numbers vary by several billion dollars depending on methodology, but the direction is identical: triple-digit percentage growth within a decade, driven almost entirely by AI rack density rather than traditional enterprise computing. Dell’Oro Group research director Alex Cordovil’s own assessment of the inflection point, cited later in this piece, captures why Vertiv, CoolIT, and Motivair are now racing to lock in hyperscaler and colocation contracts before the next wave of Nvidia Rubin-generation racks arrives – the same growth story that has sent Trane Technologies’ stock climbing on its own AI cooling bet and pushed cooling-equipment maker Phononic to reportedly explore a $1.5 billion sale as consolidation sweeps the sector.

Vertiv CoolChip and Liebert XDU: The Incumbent’s Lineup

Vertiv is the largest and oldest name in this comparison. Emerson sold its Network Power business to Platinum Equity in 2016 in a deal valued at more than roughly $4 billion, and the unit was rebranded Vertiv. The company went public on the New York Stock Exchange under ticker VRT on February 10, 2020, through a merger with a Goldman Sachs-backed special purpose acquisition company, and is now running a reported $15 billion-plus order backlog tied to AI infrastructure, according to Forbes’ April 2026 coverage of the stock. Vertiv CEO Giordano Albertazzi’s own comments to the outlet, quoted later in this piece, help explain why the company keeps expanding its CDU range rather than consolidating it.

Vertiv’s current CDU portfolio spans two overlapping product families. The CoolChip CDU line is rated at 70 kW, 100 kW, 121 kW, 450 kW, 600 kW, 1,350 kW, and 2,300 kW depending on the model, according to Vertiv’s own product catalog. The CoolChip CDU 121, specifically, is documented as a single-rack unit designed to support Nvidia GB200 systems, running on water or a water/glycol (PG-25) fluid mix. Separately, Vertiv also markets the Liebert XDU family – XDU070, XDU350, and XDU1350 – which HPE lists as a compatible CDU option for its own GB300 NVL72 rack reference configuration. HPE’s documentation specifies that a single 1.3 MW-rated XDU1350 can support up to eight GB300 NVL72 racks, each pulling roughly 132 kW nominal, for a combined load of about 1.056 MW – leaving some headroom below the unit’s 1.3 MW ceiling.

CoolIT Systems: Betting on Rack-Scale Density

CoolIT Systems, a Calgary-founded liquid cooling specialist, has built its entire current product push around explicit Nvidia GB200 and GB300 NVL72 compatibility claims rather than general-purpose CDU ratings. The company’s CDU product line includes the AHx240, a self-contained liquid-to-air unit rated at 240 kW that needs no facility liquid supply, and which CoolIT’s own product page states can support two Nvidia GB300 NVL72 or GB200 NVL72 racks simultaneously at a 15°C approach temperature difference with N+1 pump redundancy.

At the high end, CoolIT’s CHx2000 is a row-based CDU rated at 2,000 kW that the company explicitly states can support up to 12 Nvidia GB300 NVL72 racks from a single unit – the highest documented rack count in this comparison. A smaller CHx200 model, rated at 200 kW, is built as a rack-mount unit with N+1 redundant pumps and power supplies and is marketed as supporting up to 200 servers per rack, though CoolIT’s published materials don’t specify a fixed rack count for that particular model. CoolIT doesn’t publish coolant chemistry details for the CHx family in its public datasheets, which is a meaningful gap compared with Vertiv’s explicit water/PG-25 disclosure for the CoolChip 121.

Motivair by Schneider Electric: The $850 Million Wild Card

Motivair Corporation, headquartered in Buffalo, New York, became the most-discussed name in liquid cooling after Schneider Electric announced it would acquire a 75% stake in the company for a reported $850 million, a deal announced in October 2024 and completed in February 2025, with an option for Schneider to buy the remaining stake by 2028. That price tag alone signals how seriously a traditional power-management giant is treating direct-to-chip cooling as a growth category rather than a side business.

Motivair’s branded MCDU lineup, now sold under the “Motivair by Schneider Electric” label, ranges from the compact MCDU-4U at roughly 105 kW nominal up through the MCDU-70, which Motivair’s own 2025 materials describe as delivering 2.5 MW per unit – the single largest capacity figure of any model from the three vendors in this piece. In between sit the MCDU-10 (up to 210 kW), MCDU-15 (310 kW), MCDU-25 (625 kW), and MCDU-40 (1,250 kW). Motivair explicitly markets the MCDU-10 as engineered for Nvidia’s NVL72 rack configuration supporting GB200 systems, delivering up to 210 kW of cooling from a compact 4U chassis. Schneider Electric’s own press materials describe the broader portfolio, including the newer MCDU-45 and MCDU-55 models, as capable of scaling to 10 MW and beyond for what the company calls “next-gen AI factories,” per a Schneider Electric press release.

Specs Comparison: Vertiv vs CoolIT vs Motivair

The table below pulls together the publicly documented specifications for each vendor’s direct-to-chip liquid cooling lineup as of October 2026. Figures vary with operating conditions (approach temperature difference, coolant mix, and ambient site conditions), so treat capacity numbers as vendor-rated ceilings rather than guaranteed real-world output at every site.

Spec	Vertiv	CoolIT Systems	Motivair (Schneider Electric)
Flagship product family	CoolChip CDU / Liebert XDU	CHx / AHx series	MCDU series
Smallest rated model	CoolChip CDU 70 (70 kW)	AHx240 (240 kW, liquid-to-air)	MCDU-4U (105 kW)
Largest rated model	CoolChip CDU 2300 (2,300 kW)	CHx2000 (2,000 kW)	MCDU-70 (2,500 kW)
Max racks per unit (documented)	1 rack (CoolChip 121)	Up to 12 GB300 NVL72 racks (CHx2000)	Not specified per-unit; portfolio scales to 10 MW+
Nvidia GB200/GB300 NVL72 compatibility claim	Yes (CoolChip 121; XDU1350 listed in HPE’s GB300 reference design)	Yes, explicit (AHx240, CHx2000)	Yes, explicit (MCDU-10)
Coolant type disclosed	Water or PG-25 water/glycol mix	Not publicly specified for CHx line	Not publicly specified in retrieved materials
Facility liquid required	Yes	No for AHx240 (self-contained, liquid-to-air)	Yes
Parent/independent status	Independent, publicly traded (NYSE: VRT)	Independent (private)	75% owned by Schneider Electric since Feb. 2025
Headquarters	Westerville, Ohio	Calgary, Canada	Buffalo, New York
Reported 2025-2026 market share (liquid cooling, Dell’Oro estimate)	Over 11.3% (2025)	Not publicly disclosed	Not publicly disclosed
Named 2026 financial signal	$15B+ AI-driven order backlog (Forbes, Apr. 2026)	Not publicly disclosed	$850M stake deal (Schneider Electric, Feb. 2025)
Redundancy features disclosed	Not detailed in retrieved CoolChip datasheets	N+1 redundant pumps (AHx240, CHx200)	Not detailed in retrieved MCDU brochure excerpts

Benchmarks: How Much Cooling a GB300 NVL72 Rack Actually Needs

Rather than comparing synthetic lab benchmarks, which these vendors rarely publish in a directly comparable form, the most useful real-world benchmark here is rack power density itself, cross-checked against three independent sources. Nvidia’s own enterprise reference architecture documentation states a GB300 NVL72 rack can draw up to 142 kW and houses 72 Blackwell Ultra GPUs delivering 130 TB/s of aggregate NVLink bandwidth across the domain, alongside 800 Gb/s ConnectX-8 networking per compute tray. HPE’s QuickSpecs sheet for its own GB300 NVL72 hardware configuration lists a slightly different but consistent figure: 132 kW nominal thermal design power and roughly 155 kW peak electrical design power per rack, built from eight power shelves each capable of 33 kW, with each shelf containing six 5.5 kW power supplies.

Put those two primary-source numbers against the CDU capacity table above, and the practical math becomes clear. A single Vertiv Liebert XDU1350, rated at 1.3 MW, can realistically support about eight racks running at HPE’s 132 kW nominal rating – which checks out almost exactly against HPE’s own published guidance of eight racks per CDU. CoolIT’s CHx2000, rated at 2,000 kW, claims support for up to 12 GB300 NVL72 racks, which at 142 kW per rack would total roughly 1.7 MW of actual IT load against a 2 MW-rated unit – a similar headroom ratio to Vertiv’s configuration. Motivair’s flagship MCDU-70, at a claimed 2.5 MW, would need to support a comparable 14 to 17 racks to hit similar utilization, though Motivair’s public materials don’t state an explicit per-unit rack count the way CoolIT does.

The third benchmark data point comes from the market-research side rather than engineering specs: Grand View Research’s market model assumes liquid cooling adoption scales in lockstep with rack density increases, which is why its growth curve (20.1% CAGR from 2026 through 2033) tracks almost identically to Nvidia’s own generational rack-power roadmap – from roughly 40 kW racks in the Hopper era to over 140 kW for GB300, with Rubin-generation racks widely expected to push past 200 kW within the next product cycle.

Nvidia’s Next Rack: What Rubin Means for Cooling Vendors

None of the three CDU vendors in this comparison can afford to design solely around today’s GB300 NVL72 generation, because Nvidia’s roadmap doesn’t stop there. Early planning figures for Nvidia’s next-generation Vera Rubin NVL144 platform put per-rack power consumption at approximately 190 kW, with some infrastructure analysts citing a broader 190 kW to 230 kW range depending on configuration, and a distinct higher-density CPX variant reportedly reaching around 370 kW per rack. Nvidia has not published an official NVL144 rack-power specification as of this comparison, so these figures should be treated as industry planning estimates rather than confirmed datasheet numbers – but even the low end of that range sits roughly 35% above the GB300 NVL72’s documented 142 kW ceiling.

That trajectory matters directly for the CDU comparison above. Vertiv’s CoolChip CDU 2300, CoolIT’s CHx2000, and Motivair’s MCDU-70 are all already rated well above what a single Rubin-generation rack will need, which means the real competitive question for the next hardware cycle isn’t raw per-unit capacity – it’s how many Rubin-class racks each CDU can serve once rack-level power jumps by a third or more. A CoolIT CHx2000 that comfortably supports 12 GB300 NVL72 racks today, for instance, would likely need to drop to somewhere closer to eight or nine Rubin NVL144 racks to stay within the same utilization safety margin, unless CoolIT revises the product’s rated capacity. Expect all three vendors to announce updated rack-count guidance once Nvidia locks in official Rubin power specifications.

PUE and the Efficiency Case for Liquid Cooling

Power usage effectiveness, or PUE, measures how much total facility energy a data center consumes for every unit of energy that actually reaches IT equipment; a PUE of 1.0 would be a theoretical facility with zero cooling or power-distribution overhead. Data center efficiency tracking site mGrid reported in February 2026 that direct-to-chip liquid cooling installations commonly land in a 1.10 to 1.20 PUE range, while two-phase immersion cooling – a more intensive but less widely deployed approach – can push PUE down to roughly 1.02 to 1.07. The same analysis characterized direct-to-chip cooling as roughly 30% to 40% more energy-efficient than a comparable traditional air-cooled facility, though it stopped short of citing a single universal air-cooled PUE baseline, since that number swings heavily with climate, chiller type, and economizer use.

None of Vertiv, CoolIT, or Motivair publishes a guaranteed site-level PUE figure tied to their specific CDU models, since actual efficiency depends on the full facility design around the unit, not the CDU in isolation. That’s worth remembering when a sales deck cites an efficiency number: the CDU is one component in a chain that includes chillers, cooling towers or dry coolers, pumps, and the building’s electrical distribution path. Still, the broad 1.10-to-1.20 PUE range for direct-to-chip systems lines up with why operators are willing to pay a premium for retrofits – at gigawatt-campus scale, even a 20-to-30-percentage-point PUE improvement translates into tens of megawatts of electricity that can instead go toward GPU compute. It’s also why some operators are exploring unconventional water sources for cooling loops altogether, including the industry’s emerging interest in produced water storage for data center cooling as freshwater access tightens near major AI campuses.

Data Center Liquid Cooling Pricing: What’s Actually Public

Unlike GPUs or gaming hardware, coolant distribution units don’t carry public list prices. None of the three vendors in this comparison publishes per-unit CDU pricing, and none discloses standalone product revenue broken out from their parent companies’ broader thermal-management or electrical-equipment segments. That absence of pricing transparency is itself a data point worth flagging for anyone budgeting a liquid-cooling retrofit: expect a formal RFP and site-specific quote process rather than a catalog price.

Vendor	Public pricing signal available	What it actually tells buyers
Vertiv	No per-unit CDU pricing published; $15B+ order backlog reported by Forbes (Apr. 2026)	Deep enterprise sales pipeline; pricing will be quote-based and scale-dependent
CoolIT Systems	No per-unit CDU pricing or company revenue published	Privately held; pricing negotiated directly through hyperscaler/OEM channels
Motivair (Schneider Electric)	$850M for a 75% ownership stake (Feb. 2025 close)	Values the whole company, not a single CDU – but signals Schneider’s confidence in near-term AI cooling demand
Market-wide context	$6.7B-$8.2B global market in 2025-2026 (Grand View Research, Dell’Oro-cited estimates)	A market this size, split across a handful of vendors, implies individual large AI-campus contracts routinely run into tens of millions of dollars

The practical takeaway: budgeting for direct-to-chip cooling at AI-data-center scale isn’t a “data center liquid cooling price per kW” lookup the way buying a gaming PC component is. It’s closer to buying a building system, where the final number depends on rack count, redundancy tier, facility retrofit scope, and length of service contract – all negotiated per project.

Coolant chemistry is one of the quieter line items buried in that negotiation, and it’s worth understanding before signing anything. Vertiv’s CoolChip CDU 121 documentation explicitly names water or a PG-25 (25% propylene glycol) mix as supported fluids, a combination chosen for its low toxicity and resistance to freezing in facilities that experience cold shutdown periods. Neither CoolIT nor Motivair publishes equivalent fluid-chemistry detail for their CHx or MCDU lines in the materials reviewed for this comparison, which means buyers should request that information directly during procurement rather than assuming compatibility across vendors. Mixing an unapproved fluid into a cold-plate loop risks galvanic corrosion between dissimilar metals, a failure mode that can take months to show up as degraded thermal performance and is far more expensive to fix after a rack is already in production than before.

Real-World Deployments and Use Cases

Five concrete, documented examples illustrate where each vendor’s hardware actually lands in 2026:

HPE’s GB300 NVL72 reference configuration lists Vertiv’s Liebert XDU1350 as a compatible CDU option, supporting up to eight racks per unit with either top-feed/overhead or bottom-feed/underfloor secondary-loop plumbing, and a documented 44U rack liquid manifold – a blueprint multiple OEM partners can replicate for enterprise AI-factory builds.
CoolIT’s CHx2000 row-based deployment model is explicitly sized around clusters of up to 12 Nvidia GB300 NVL72 racks from a single CDU, a density aimed squarely at AI cloud providers building out large GPU clusters rather than mixed-workload enterprise rooms.
Motivair’s MCDU-10 is marketed specifically around Nvidia’s NVL72 rack configuration for GB200 systems, positioning it as the go-to retrofit unit for operators moving an existing GB200 deployment from air-assisted to full direct-to-chip cooling without redesigning the whole room.
Schneider Electric’s newer MCDU models, following the Motivair deal, are pitched by the company as scaling to 10 MW and beyond for what Schneider calls next-generation AI factories – language aimed at gigawatt-scale campuses rather than single rooms.
Vertiv’s own $15 billion-plus order backlog, reported by Forbes in April 2026, reflects demand pulled largely from hyperscale and colocation operators retrofitting or building new facilities around GB200/GB300-class racks, reinforcing why Vertiv’s CEO describes the AI buildout as still early-stage, as detailed further below.

What Industry Voices Are Saying About the Shift to Liquid Cooling

Industry leaders tracking this transition consistently frame it as structural rather than a temporary spike tied to one GPU generation. Dell’Oro Group research director Alex Cordovil put it most succinctly in Vertiv’s 2026 Frontiers report, saying “liquid cooling is moving from niche to necessary.” That assessment lines up with how Vertiv’s own executives describe the market: Scott Armul, the company’s chief product and technology officer, said the industry sees “cross-technology forces, including extreme densification, driving transformative trends such as higher voltage DC power architectures and advanced liquid cooling that are important to deliver the gigawatt scaling that is critical for AI innovation,” in comments published through Vertiv’s investor relations channel.

Vertiv CEO Giordano Albertazzi, speaking to Forbes in April 2026 about the company’s AI-driven stock performance, reinforced that the growth curve isn’t close to flattening, noting “we are in the early stages of the data center buildout” and separately that “use of artificial intelligence is just scratching the surface.” Those comments matter for buyers evaluating CoolIT and Motivair too, since all three vendors are effectively competing for a share of demand that none of them currently believe has peaked.

Use-Case Recommendations: Which Vendor Fits Which Deployment

Because none of these three vendors competes on price transparency, the real differentiation is architectural fit. Here’s how the specs translate into practical buying decisions:

Hyperscaler gigawatt campuses: Motivair’s newer MCDU models, explicitly marketed by Schneider Electric as scaling to 10 MW and beyond, target this tier directly – large, purpose-built AI factories where a handful of massive CDUs serve entire buildings.
Dense GPU-cluster cloud providers: CoolIT’s CHx2000, with its documented 12-rack-per-unit support for GB300 NVL72, is built for AI cloud operators running uniform, high-density GPU rows rather than mixed workloads.
Enterprise pilots retrofitting existing GB200 racks: Motivair’s compact MCDU-10 (210 kW, 4U form factor) is explicitly positioned for exactly this scenario – adding direct-to-chip cooling to an existing NVL72 deployment without a full room redesign.
Facilities without a chilled-water loop already in place: CoolIT’s self-contained AHx240, which needs no facility liquid supply, is the only option among the three with a documented liquid-to-air design that sidesteps a water-loop retrofit entirely, at the cost of lower absolute capacity (240 kW, two racks).
OEM-validated reference builds (HPE, Dell, Lenovo server partners): Vertiv’s Liebert XDU1350 is the CDU explicitly named in HPE’s own GB300 NVL72 QuickSpecs documentation, making it the lowest-friction choice for buyers who want a pre-validated, vendor-blessed configuration rather than an independent integration project.

Migration Guide: Moving From Air-Cooled to Direct-to-Chip Cooling

Retrofitting an existing air-cooled room for direct-to-chip liquid cooling is a facilities project as much as an IT project, and most of the real work happens before a single CDU ships. Teams that treat this as a pure procurement decision – pick a vendor, order the hardware, bolt it in – routinely discover mid-project that the building’s electrical service, floor loading, or plumbing runs can’t support the new rack density without separate upgrades. That’s the same reason grid operators like PJM are now moving to formalize how AI data centers get added to the power system, as seen in FERC’s recent approval of PJM’s data center registry, and why rack-level power architecture – the kind behind AMD’s 32,768-core Helion rack for AI agents – increasingly gets designed hand-in-hand with the cooling loop rather than after it. Based on the reference architectures Nvidia, HPE, and the three CDU vendors publish, the realistic sequence looks like this:

Audit current rack power density. Compare existing per-rack draw against the target rack’s rated power – a GB300 NVL72 rack needs up to 142 kW, roughly five to seven times what a legacy air-cooled rack handles.
Decide on secondary-loop architecture. HPE’s GB300 reference design supports both top-feed/overhead and bottom-feed/underfloor plumbing – the choice depends on whether the facility already has underfloor infrastructure or overhead busway space available.
Size the CDU to the rack count, not just total kW. A CDU rated well above your total IT load on paper can still be the wrong fit if its documented max-rack-count (eight for Vertiv’s XDU1350, up to 12 for CoolIT’s CHx2000) doesn’t match your row layout.
Confirm coolant chemistry compatibility. Vertiv’s CoolChip 121 explicitly supports water or a PG-25 water/glycol mix; always confirm fluid compatibility with cold-plate manufacturer specs before procurement, since CoolIT and Motivair don’t disclose coolant chemistry as clearly in public materials.
Specify leak detection at both the rack and tray level. Nvidia’s own NVL72 reference architecture bakes in rack-level and tray-level liquid-leak detection; any retrofit should match or exceed that standard.
Plan N+1 redundancy for pumps and power. CoolIT’s AHx240 and CHx200 both ship with N+1 redundant pumps by design – confirm equivalent redundancy is specified for whichever vendor you choose, since not every published datasheet details this by default.
Stage a pilot rack before full rollout. Motivair’s compact MCDU-10 and CoolIT’s AHx240 are explicitly positioned as pilot-friendly, lower-footprint units for exactly this kind of staged migration.
Integrate monitoring before go-live, not after. A simple capacity-headroom check, run continuously against live telemetry, catches density creep before it becomes a thermal incident:

# Simple CDU headroom check (pseudocode)
# Flags when a CDU's live IT load creeps above a safe utilization ceiling

cdu_rated_kw = 1300          # e.g. Vertiv Liebert XDU1350
racks_connected = 8          # per HPE's documented GB300 NVL72 config
rack_draw_kw = 132           # HPE nominal TDP per rack

total_load_kw = racks_connected * rack_draw_kw
utilization = total_load_kw / cdu_rated_kw

if utilization > 0.85:
    alert("CDU utilization above 85% — plan capacity expansion")
else:
    log(f"CDU load: {total_load_kw}kW of {cdu_rated_kw}kW rated ({utilization:.0%})")

Negotiate service and support terms up front. None of the three vendors publishes standard warranty terms publicly, so get support-response SLAs, spare-parts lead times, and coolant-maintenance schedules written into the contract rather than assumed.
Budget for facility-side work, not just the CDU. Piping, manifolds (HPE specifies a 44U rack liquid manifold for its GB300 reference build), and electrical upgrades to feed higher-density power shelves typically cost as much as the CDU hardware itself.

Pros and Cons: Vertiv

Pro: Broadest published capacity range of the three, from 70 kW to 2,300 kW across the CoolChip family
Pro: Explicitly named as a compatible CDU in HPE’s own GB300 NVL72 reference architecture, reducing integration risk
Pro: Backed by a reported $15B+ AI-driven order backlog, suggesting strong long-term parts and support availability
Con: Public datasheets don’t detail redundancy architecture as explicitly as CoolIT’s N+1 pump disclosures
Con: Largest-capacity CoolChip models require a facility chilled-water loop already in place

Pros and Cons: CoolIT Systems

Pro: Only vendor in this comparison offering a documented liquid-to-air, facility-water-free option (AHx240)
Pro: Highest documented per-unit rack count – up to 12 GB300 NVL72 racks from a single CHx2000
Pro: N+1 redundant pumps and power supplies explicitly built into both the AHx240 and CHx200
Con: Privately held, with no public revenue or market-share figures to benchmark financial stability
Con: Coolant chemistry for the CHx line isn’t disclosed in public product materials

Pros and Cons: Motivair (Schneider Electric)

Pro: Largest single-unit capacity in this comparison – the MCDU-70 is rated at 2.5 MW per unit
Pro: Backed by Schneider Electric’s $850 million stake, bringing global supply-chain and support scale to a previously independent company
Pro: Compact MCDU-10 is purpose-built for retrofitting existing Nvidia GB200 NVL72 racks
Con: Per-unit rack-count support isn’t published the way CoolIT states it, making apples-to-apples row-planning harder
Con: Still mid-transition as Schneider integrates the brand – some materials still carry the “Motivair” name only, others “Motivair by Schneider Electric”

The Verdict: Which Liquid Cooling Vendor Should You Bet On

There’s no single winner here because the three vendors aren’t actually chasing identical customers. If the deciding factor is lowest integration risk on a validated Nvidia GB300 NVL72 build, Vertiv’s Liebert XDU1350 wins by default – it’s the CDU HPE names directly in its own reference documentation, and Vertiv’s $15 billion-plus backlog signals it isn’t going anywhere. If the priority is maximum rack density per CDU for a uniform AI cloud cluster, CoolIT’s CHx2000 is the clear leader on paper, with a documented 12-rack-per-unit figure that neither Vertiv nor Motivair matches in public materials. And if the deployment is a gigawatt-scale AI factory where a handful of enormous CDUs need to cover an entire building, Motivair’s MCDU-70 at 2.5 MW per unit, now backed by Schneider Electric’s balance sheet, is the heaviest hitter by raw capacity.

What’s consistent across all three is the scale of the shift underway. With the data center liquid cooling market estimated between $6 billion and $8.2 billion in 2026 and projected to grow at 18-20% annually through the early 2030s, and with Nvidia’s own roadmap pushing rack power well past 140 kW, every one of these vendors is selling into a market that Vertiv’s own CEO describes as still in its “early stages.” Buyers choosing between them should weight documented rack-support numbers and GB200/GB300 compatibility claims more heavily than brand recognition, since all three now have verifiable, Nvidia-aligned reference designs in the field.

Frequently Asked Questions

What is a coolant distribution unit (CDU) in a data center?
A CDU is the hardware that sits between a facility’s primary chilled-water loop and a server rack’s internal secondary coolant loop. It pumps and regulates coolant temperature and flow to cold plates mounted directly on GPUs and CPUs, which is why this approach is called direct-to-chip liquid cooling.

Why can’t air cooling handle Nvidia’s newest AI racks?
A Nvidia GB300 NVL72 rack can draw up to 142 kW according to Nvidia’s enterprise reference architecture documentation, and HPE’s QuickSpecs list roughly 155 kW peak electrical design power per rack. Traditional air cooling tops out well below that, which is why direct-to-chip liquid cooling has become standard for GB200/GB300-class deployments.

Is Motivair owned by Johnson Controls or Schneider Electric?
Schneider Electric. The company announced a deal to acquire a 75% stake in Motivair Corporation in October 2024, and the transaction closed in February 2025 for a reported $850 million, with an option for Schneider to acquire the remaining stake by 2028.

Which CDU supports the most Nvidia GB300 NVL72 racks per unit?
Based on publicly available vendor documentation, CoolIT Systems’ CHx2000, rated at 2,000 kW, is explicitly stated to support up to 12 GB300 NVL72 racks from a single unit – the highest documented figure among Vertiv, CoolIT, and Motivair.

How big is the data center liquid cooling market in 2026?
Estimates vary by research firm. Grand View Research puts the global market at roughly $8.2 billion in 2026, growing to $29.5 billion by 2033 at a 20.1% CAGR. A separate estimate cited through Dell’Oro Group puts 2026 at closer to $6.0 billion with an 18.2% CAGR through 2035.

Do these vendors publish pricing for their CDUs?
No. None of the three vendors covered here – Vertiv, CoolIT Systems, or Motivair – publishes per-unit list pricing for its CDU lineup. Pricing is negotiated through direct sales or OEM channels and depends heavily on project scale, redundancy requirements, and service terms.

Can a data center add direct-to-chip cooling without a full room redesign?
Yes, in some cases. Motivair’s compact MCDU-10 and CoolIT’s self-contained AHx240 are both explicitly positioned as lower-footprint options suited to retrofitting an existing Nvidia GB200 NVL72 deployment rather than requiring a ground-up facility redesign.

What’s the difference between liquid-to-air and liquid-to-liquid CDUs?
A liquid-to-liquid CDU, like most Vertiv CoolChip and Motivair MCDU models, exchanges heat between the rack’s internal coolant loop and the facility’s chilled-water supply, requiring that water infrastructure to already exist. A liquid-to-air CDU, like CoolIT’s AHx240, rejects heat directly to room air instead, which means it needs no facility liquid supply but typically covers fewer racks per unit.

How much more power will Nvidia’s Rubin-generation racks need compared to GB300?
Early 2026 planning estimates put Nvidia’s Vera Rubin NVL144 rack power at roughly 190 kW, with some estimates ranging up to 230 kW, against the GB300 NVL72’s documented 142 kW ceiling – though Nvidia has not published an official NVL144 power specification yet. That jump is expected to push CDU vendors to revise their per-unit rack-count guidance for the next hardware generation.