Data center chillers: tight supply supports growth, while free cooling changes the economics of air-cooled versus water-cooled systems
AI summary card
Data center chillers: tight supply supports growth, while free cooling changes the economics of air-cooled versus water-cooled systems
Bernstein estimates the all-in data center chiller market at roughly $8 billion in 2026 and about $16.5 billion in the base case by 2030; the choice between air-cooled and water-cooled systems is not simply a matter of efficiency, but depends on electricity prices, water prices, water scarcity constraints, and the share of free cooling.
- Chillers are important infrastructure for both liquid-cooled and air-cooled data centers, providing facility water for CDU, CRAH, or broader HVAC systems.
- The report estimates market size using new GW of capacity, 1.2x power redundancy, an all-in cost of about $1,500 per ton, and chiller penetration assumptions.
- Under the base case, the all-in market is about $8 billion in 2026 and about $16.5 billion in 2030, implying roughly 20% CAGR; the bear case is about $9 billion, while the bull case is about $20 billion.
- Air-cooled chillers have lower upfront capex and use no water, but are about 30% less efficient than water-cooled chillers on a like-for-like ton basis; water-cooled chillers save more electricity but usually require substantial water use.
- When the share of free cooling is very high, compressor power declines, but water-cooled systems using cooling towers have relatively stable water costs, which can make air-cooled operating costs more attractive.
- Trane, JCI, and Carrier each have differentiated water-cooled chillers, while Vertiv has a different positioning through trim coolers; however, current supply shortages and strong demand matter more than specification differences.
Report interpretation
Overview
This report is the first of Bernstein's three-part series on data center chillers and is positioned as an industry deep dive. It focuses on the role of chillers in data center cooling, the differences between air-cooled and water-cooled technologies, the methodology for estimating market size, the impact of free cooling on operating costs, and product comparisons among major manufacturers. The report argues that chillers serve both liquid-cooled architectures and air-cooled data centers, and are key equipment for removing heat from facility water, coolant, or air systems.
Core views
The core view is: first, the data center chiller market is being driven by AI and data center expansion, with an all-in market size of about $8 billion in 2026 and about $16.5 billion by 2030 in the base case. Second, there is no absolute winner between air-cooled and water-cooled systems; water-cooled units are usually more energy-efficient, while air-cooled units typically have lower capex and use no water. In regions with a high share of free cooling, water scarcity, or high water prices, air-cooled systems may become more attractive over the long run. Third, while product differentiation among major OEMs exists, it is not extreme, and tight supply plus strong demand are likely to be more important in determining order performance. Fourth, water management and local water restrictions at data centers could become an important variable affecting technology share.
Analysis framework
The report uses a top-down market sizing framework and operating cost sensitivity analysis. Market size starts from annual new data center GW capacity, adds a 1.2x power waste and safety redundancy assumption, estimates costs using about $1,500 per ton all-in, and builds base, bear, and bull scenarios through assumptions about chiller penetration. The operating cost section uses an 800-ton, roughly 2.5 MW chiller as an example, comparing electricity and water costs for air-cooled and water-cooled systems under 0%, 50%, and 90% free-cooling scenarios, and further compares electricity and water price differences across regions such as Northern Virginia, DFW, Atlanta, and Chicago.
Methodology notes
Deriving chiller demand from new data center GW capacity
Convert new data center capacity into a heat-load baseline, add 1.2x power waste and safety redundancy, and estimate market size using an all-in cost of about $1,500 per ton and chiller penetration assumptions.
Base, bear, and bull penetration scenarios
The scenarios do not come from differences in new GW capacity, but from changes in the share of chillers within data center cooling architectures: the base case keeps share stable, the bear case assumes dry coolers gain share, and the bull case assumes chiller share rises.
Operating cost sensitivity for air-cooled versus water-cooled systems
Using an 800-ton chiller as an example, compare the impact of electricity cost, water cost, and changes in free-cooling share on total operating cost and payback period for air-cooled and water-cooled systems.
Refrigeration cycle composed of the evaporator, compressor, condenser, and expansion valve
Chillers absorb heat from facility water systems through a refrigerant cycle, with the compressor being the most energy-intensive component; the condenser rejects heat through air or water loops, and the expansion valve reduces refrigerant pressure and temperature.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- TT (Trane Technologies PLC)A major chiller OEM, rated Outperform in the report table
- Strengths
- Its water-cooled product lineup includes CenTraVac CDHH, with the largest product reaching up to 21 MW, indicating the ability to serve large data center deployments.
- Weaknesses
- No new air-cooled product announcement similar to competitors was identified, and some metrics may temporarily lag.
- Comparison
- Compared with JCI and Carrier, Trane stands out on large water-cooled unit capacity, but the report believes specification differences are being muted by tight supply and demand.
- Risks
- Escalation of pricing-manipulation litigation; CDU or broader liquid-cooling innovation being copied by more competitors; further deterioration in the residential or transportation businesses; and chiller sales and service headwinds offsetting liquid-cooling tailwinds.
- CARR (Carrier Global Corporation)A major chiller OEM, rated Market-Perform in the report table
- Strengths
- It has both air-cooled and water-cooled product offerings; on the water-cooled side, AquaEdge 19MV4 is mentioned, and restart time under 3 minutes is a selling point.
- Weaknesses
- It is rated Market-Perform, and the target price implies less relative upside than JCI and VRT.
- Comparison
- It belongs to the same strong-product group as Trane and JCI; Carrier emphasizes short restart time.
- Risks
- Escalation of pricing-manipulation litigation, with Carrier having unique exposure; a slowdown in hyperscaler data center capex; and tighter-than-expected or prolonged delays in R-410A supply tightening.
- JCI (Johnson Controls International PLC)A major chiller OEM, rated Outperform in the report table
- Strengths
- The York HT water-cooled product advertises a 110°F lift, which can be paired with dry coolers in hot climates and eliminate water use.
- Weaknesses
- Dry-cooler configurations may carry higher electricity costs because of the higher lift, and the full specification table is still pending.
- Comparison
- Compared with Carrier and Trane, JCI appears more differentiated on high lift and potentially water-free operating solutions.
- Risks
- Failure of the internal lean transformation to translate into results; headwinds in data center chillers; and operating leverage below management's 50% guidance.
- VRT (Vertiv Holdings Co)A data center thermal management and cooling equipment supplier, rated Outperform in the report table
- Strengths
- It occupies a different market position through trim coolers, which look like air-cooled chillers but operate like dry coolers most of the time, helping reduce energy use.
- Weaknesses
- The report was unable to find a standalone chiller product line on its U.S. website, so product disclosure is relatively limited.
- Comparison
- Compared with traditional chiller OEMs, Vertiv is positioned closer to overall data center thermal management and dry-cooler-like solutions.
- Risks
- Efficiency gains reducing the amount of cooling required; an unexpected slowdown in data center expansion; and a faster-than-expected shift from NVIDIA chips to custom silicon, which could reduce the advantage from Vertiv's strong customer relationships.
- Data center chiller marketThe report's core industry asset/theme
- Strengths
- AI and new data center capacity are driving demand, the 2026-2030 base case implies about 20% CAGR, and supply is currently tight.
- Weaknesses
- Market size is sensitive to assumptions about new GW capacity, chiller penetration, and all-in cost.
- Comparison
- Compared with CDU, the report believes major chiller OEM product differentiation is less pronounced.
- Risks
- Dry coolers taking share; water-use restrictions changing technology choices; changes in electricity and water prices affecting air-cooled versus water-cooled economics; and a slowdown in hyperscaler capex.
Key data
- 2026 all-in chiller market sizeAbout $8BIncludes equipment, installation, engineering, commissioning, and auxiliary equipment.
- 2030 base-case market sizeAbout $16.5BAssumes chillers continue to cool the same share of data center compute, implying roughly 20% CAGR.
- 2030 bear-case market sizeAbout $9BAssumes dry coolers take share from chillers, implying annual growth below 5%.
- 2030 bull-case market sizeAbout $20BAssumes chiller share rises, implying annual growth above 25%.
- All-in chiller cost assumptionAbout $1,500/tonMixed air-cooled and water-cooled capex, including equipment, engineering, commissioning, pumps, cooling towers, and other support equipment.
- 2026 chiller penetration assumption75%The report believes chillers remain the cooling method for most data centers.
- Water use assumption for water-cooled chillersAbout 2.3 gallons/ton-hourOf this, about 1.8 gallons/ton-hour is evaporation and about 0.5 gallons/ton-hour is blowdown.
- Air-cooled versus water-cooled efficiency gapAbout 30% worseAir-cooled units consume more electricity for the same tonnage, but do not require operating water.
- 0% free-cooling operating cost for an 800-ton chillerAir-cooled about $0.63M, water-cooled about $0.50MWhen the compressor runs 24/7, annual operating cost for air-cooled systems is about 25% higher, corresponding to a payback period of about 3 years for the cost difference.
- 90% free-cooling operating cost for an 800-ton chillerAir-cooled about $0.14M, water-cooled about $0.20MAt a high share of free cooling, water-cooled systems still have relatively stable water costs, so air-cooled systems may be cheaper.
Impact & implications
The investment implication is that the data center cooling chain remains an important source of growth for electrical equipment and multi-industry companies, especially Trane, Johnson Controls, Carrier, and Vertiv, which have chiller, cooling system, or related product exposure. In the short term, supply-demand tightness may support orders and pricing; over the medium to long term, regional water resources, water-use regulation, free-cooling conditions, and the mix of electricity and water prices will affect the share shift among air-cooled, water-cooled, dry-cooler, and trim-cooler solutions.
Risks
- The assumption for new data center GW capacity is highly uncertain and could cause market size estimates to deviate materially.
- Dry coolers or other cooling solutions may take share from chillers, pushing the 2030 market size in the bear case down to about $9 billion.
- Regional water restrictions and rising water prices could weaken the economics of water-cooled systems that use cooling towers.
- A high share of free cooling reduces compressor power consumption, which can offset the electricity savings of water-cooled systems through stable water costs.
- A slowdown in hyperscaler data center capex would affect demand for cooling equipment.
- The major product specification tables are still incomplete, and some product-performance differences will need further validation.
- Supply-chain and capacity constraints may support orders in the near term, but they could also limit the pace of revenue recognition.
What to watch
- Revisions to forecasts for new data center GW capacity.
- Changes in chiller penetration within new data center cooling architectures.
- Order share shifts among air-cooled, water-cooled, dry cooler, and trim cooler solutions in key regions.
- Changes in electricity prices, water prices, and water-use regulation in major data center regions such as Northern Virginia, DFW, Atlanta, and Chicago.
- Actual operating costs and customer choices in regions with high free-cooling shares.
- Full product specification tables and delivery capabilities for Trane, JCI, Carrier, and Vertiv.
- Energy performance of JCI's York HT dry-cooler configuration in high-temperature environments.
- Whether Carrier's short restart time, Trane's high-capacity water-cooled products, and Vertiv's trim-cooler positioning translate into order advantages.
- The next two reports: the second on 'chiller gate' and the third on chiller service economics.