Data center cooling is a USD 31.4 billion market in 2026, up from USD 26.3 billion in 2025, and is projected to reach USD 128.3 billion by 2033 at a 22.3% CAGR, a fourfold expansion that is redrawing the supplier map [S8]. The driver is unambiguous: AI training racks are crossing 70 kW per rack and pushing past 100 kW in the largest 2026 hyperscale builds, well above the 10–25 kW per rack that legacy CRAC and CRAH architectures were designed to handle [S1][S5].
Liquid cooling is the structural winner in that shift. Direct liquid cooling is growing at a 24.7% CAGR through 2035, while air cooling's 5.2% CAGR significantly lags the overall market rate of 12.3% [S4]. Major cloud providers have standardized direct liquid cooling across next-generation AI data center designs, with Google TPU v5 clusters, Microsoft Azure AI infrastructure, and Meta AI racks all specifying cold-plate or single-phase loops at the rack level [S4]. Carrier reported Q1 2026 data center orders up more than 500% year-over-year with backlog fully covering its USD 1.5 billion full-year target; Trane reported Americas commercial applied equipment bookings up more than 160% in Q1; and Ecolab paid USD 4.75 billion to acquire CoolIT Systems, a direct-to-chip liquid cooling specialist [S2].
Air, Liquid, and Immersion: The Density Decision Grid
Air cooling (CRAC, CRAH, in-row, and in-rack doors) remains the baseline for legacy raised-floor white space and typically handles 10–25 kW per rack, while rear-door heat exchangers extend air-side infrastructure to 40–60 kW per rack by pairing DX or chilled water with a rack-mounted coil [S1]. Single-phase and two-phase immersion baths unlock 200–400 kW per rack densities and are the dominant architecture for new AI training pods, with the trade-off that dielectric fluids, pump skids, and condensers raise the per-MW mechanical footprint substantially compared with chilled-water plants [S1].
Cooling systems consume 30 to 40 percent of a facility's total electricity, meaning a 100 MW campus burns 30 to 40 MW just to reject heat, which makes the air-to-liquid transition a pure OpEx lever [S3]. Power Usage Effectiveness remains the primary efficiency metric: the Uptime Institute's 2025 Global Data Center Survey reported a weighted average PUE of 1.54, unchanged for the sixth consecutive year, while the best hyperscale campuses operate between 1.03 and 1.10 and Google reports a fleet average of 1.10 [S3].
Vendor Map: US OEMs, US Reps, and Chinese System Houses
The supplier base is splitting into three models. SVL Data Center Cooling is a North American rep and integrator offering design, installation, and lifetime field service for everything from single racks to server farms, a service-heavy model common among North American independents that resell multiple brands rather than manufacturing their own coils and compressors [S1]. QuantaCool sells a patented no-pump, no-water heat-rejection module out of the US, with the explicit differentiator that the loop is pump-free and water-free, targeting edge and HPC deployments where leaks and pumped loops are unacceptable [S1].
Chinese system builders such as Soeteck Power (Shenzhen) now ship precision cooling, UPS, telecom power, AIDC racks, and energy storage as a single package, reflecting the Asian bid to export full-stack white-space solutions and mapping directly onto the AIDC reference designs being tendered across Asia in 2026 [S1]. Mainland-China sourcing directories list more than 100 vendors tagged under data center environment monitoring and adjacent cooling categories, with manufacturers that also produce temperature/humidity, LoRa, and wireless sensors for DCIM integration [S1]. A practical sourcing lens for buyers is the industrial valve and pressure transmitter supply chain behind every cold-plate loop, where the same vendors also serve chemical, HVAC, and process-plant bids.
Standards and Specifications Buyers Should Anchor To
Specifiers writing 2026 cooling RFPs are consolidating around four reference frameworks [S6]. ASHRAE TC 9.9 remains the operating-envelope bible, with a recommended inlet air temperature range for Class A1 through A4 equipment of 18 to 27 degrees Celsius (64.4 to 80.6 degrees Fahrenheit), an allowable range that extends higher, and A4 environments tolerating inlet temperatures up to 45 degrees Celsius [S3]. The Uptime Institute Tier I–IV classification defines availability targets from 99.671% (Tier I) to 99.995% (Tier IV) annually and drives UPS topology, generator count, and concurrently maintainable design [S5].
EU operators are now forced by the EU Energy Efficiency Directive (EED) recast, in force from 2024, to report PUE and to reuse waste heat above thresholds above 1 MW IT load, with national transpositions tightening reporting windows through 2026 [S5]. The facility build itself follows ISO/IEC 22237 for the physical plant and TIA-942 for telecommunications infrastructure, while the softwarestack is delivered through OpenShift or OpenStack control planes on enterprise Linux [S5]. For a process engineer evaluating a 2026 cold-plate skid, the flow meter spec on the secondary loop and the pressure sensor calibration on the dielectric circuit are the two measurements that decide whether the rack will survive a 100 kW transient.
Investment and M&A Signals: Where the Capital Is Going
Private equity is paying attention. Capstone Partners noted in their April 2026 HVAC Equipment M&A update that they currently have two HVAC liquid cooling component manufacturing deals in market, and EBITDA multiples in the HVAC equipment sector averaged 10.9x in 2025, up nearly two turns from 9.0x in 2024, with the premium being paid for companies with data center exposure [S2]. The Ecolab-CoolIT deal is the clearest evidence that the largest HVAC-adjacent companies see liquid cooling as a strategic necessity rather than an optional add-on [S2].
For HVAC contractors, the data center cooling boom is largely indirect: data center construction creates commercial project opportunities in markets where hyperscale facilities are being built, with Northern Virginia, Phoenix, Dallas, Columbus, and Chicago metropolitan areas seeing the highest concentration [S2]. The practical split between chip-level heat rejection (cold-plate, immersion) and facility-level cooling (chillers, cooling towers, precision air handling, BMS) is what keeps Trane, Carrier, Johnson Controls, and Stulz relevant even as the rack-level architecture shifts to liquid [S2][S4]. For broader context on how this procurement wave compares with other industrial sourcing booms, see the sourcing smart meters from China spec-first guide, which uses a similar supplier-tiering framework for a different equipment class.
Regional and Operator Split: Where the Buildouts Land
Asia Pacific captured 36.9% of 2025 revenue, the largest regional share, driven by hyperscale build-out across China, India, and Southeast Asia, while North America leads at 38.6% of 2026 revenue on a different methodology and Asia-Pacific is the fastest-growing region at 15.8% CAGR [S8][S4]. On the demand side, the top 10 data center operators (AWS, Microsoft Azure, Google Cloud, Meta, Equinix, Digital Realty, and others) anchor the order book, with hyperscale commanding 44.2% of end-use cooling revenue and edge data centers the fastest-growing data center type at 19.3% CAGR [S4][S8].
Cloud service providers hold 36.8% of end-user vertical revenue, the single largest customer segment, and that concentration is the reason Ecolab, Carrier, and Trane are all repositioning their commercial HVAC portfolios around the data center opportunity [S4][S2]. A side-by-side comparison of the three primary architectures: Air (CRAC/CRAH/RDHx) covers 10–60 kW per rack at 5.2% CAGR growth and lowest mechanical footprint per MW; Direct liquid cooling (cold-plate) covers 70–150 kW per rack at 24.7% CAGR and adds coolant distribution units but reuses chiller plant; Immersion (single-phase, two-phase) covers 200–400 kW per rack at 32.4% CAGR but introduces dielectric fluid aging, pump seal leaks, and condenser fouling that few legacy service contractors are trained on [S1][S4]. A related deep-dive on the demand side is the data center cooling demand 2026-2030 liquid surge and hyperscale capex piece, which extends the same CAGR cuts to a five-year window.
Selection Criteria: Who Air, Liquid, and Immersion Are For
Air cooling and RDHx are the right call for general-purpose enterprise workloads, retrofits into legacy raised-floor white space, and any buyer whose rack density will stay below 40 kW per rack over the asset's life; the supplier pool is the deepest here and service contractor coverage is universal [S1][S3]. Direct-to-chip liquid cooling is the right call for AI training pods in the 70–150 kW per rack band, for any new build with a confirmed 5-year AI workload roadmap, and for EU sites that must meet EED heat-reuse reporting above 1 MW IT load [S4][S5].
Immersion is for the 200–400 kW per rack frontier, typically hyperscale AI training clusters, and is the wrong call for any operator without an in-house dielectric-fluid management program or a screened supplier with immersion-specific field service [S1]. The cooling architecture for any 2026 bid should be selected against three gates: workload mix (AI training pushes density, traditional enterprise pushes VM consolidation), availability tier (Tier III vs Tier IV changes UPS topology and concurrently maintainable design), and energy and water footprint (PUE, WUE, and CUE reporting plus the EED heat-reuse threshold) [S5]. The wider industrial-control lens, including how a PLC and data logger layer tie the chiller plant, CDUs, and rack-level manifolds into a single control plane, is what separates a spec sheet from a working system.
Trackable signals for the next 90 days: Uptime Institute's mid-year PUE update against the 1.54 weighted average, any new HVAC-adjacent M&A in liquid cooling components (Capstone flagged two deals in market in April 2026), and EU member-state EED national transposition deadlines through the end of 2026.