
30 Data Center Cooling Energy Statistics (2026)
30 evidence-based U.S. and European statistics on data center electricity demand, cooling energy, PUE, AI rack density, and the financial case for autonomous DRL cooling optimization.
Evidence-based U.S. and European statistics on data center electricity demand, thermal efficiency, power usage effectiveness, AI rack density, and the business case for autonomous cooling optimization.
Cooling is one of the most consequential controllable non-IT energy loads in a data center. The U.S. Department of Energy (DOE) states that cooling can account for up to 40% of facility energy use. At the same time, the June 2026 Lawrence Berkeley National Laboratory (LBNL) update estimates that U.S. data centers consumed 192 TWh in 2024, while the European Commission estimates that EU data centers consumed 68 TWh in 2024. These estimates cover different geographic areas in the same year, 2024, and both underscore the importance of efficient thermal operations.
Cooling infrastructure investment is also expanding. A September 2026 market report estimates 2025 data center cooling revenue at $7.99 billion in the U.S. and $4.51 billion in Europe. Market revenue indicates spending on cooling solutions and services, rather than an amount of wasted electricity.
Synapse Thermal addresses this opportunity through a cooling-focused, autonomous deep reinforcement learning (DRL) agent, described in its positioning as one unified control model. Its distinguishing combination is closed-loop optimization, one on-premises inference model, and hardware-agnostic operation across air, chilled-water, direct-to-chip liquid, and immersion cooling. Synapse states that its approach offers potential reductions of up to 40% in cooling energy costs. That is a company-stated opportunity for site-specific cooling expenditure, rather than a fixed reduction in total facility electricity use.
Key Takeaways
- Cooling energy matters: DOE says cooling can represent up to 40% of total data center energy use at some sites. This is an upper-end figure, not a typical percentage for every facility.
- U.S. electricity demand is rising: LBNL estimates 192 TWh in 2024 and models a compounded uncertainty range of 521-843 TWh in 2030, with 649 TWh in its Reference Case. The 2030 values are projections, not observed consumption.
- Europe faces a comparable efficiency imperative: The European Commission estimates 68 TWh in 2024 and projects 114 TWh by 2030 for EU data centers.
- PUE requires careful interpretation: An incomplete EU reporting sample showed a weighted-average PUE of 1.36. PUE measures total facility overhead relative to IT energy, not cooling electricity alone.
- AI increases thermal density: A U.S. liquid cooling market report describes 60-100 kW per rack in some high-density AI installations.
- Synapse brings complementary capabilities together: Its autonomous DRL control, local inference, and support for existing air and liquid systems provide a unified cooling optimization strategy. Synapse identifies 1.03 as a potential optimal PUE, with published company-stated deployment and payback timeframes of 4-6 weeks and 8-14 months, respectively.
The U.S. and European Impact of Data Center Electricity Demand
Electricity demand provides the context for efficient cooling. The following estimates distinguish observed historical consumption, projected demand, and construction capacity.
1. U.S. data centers used 192 TWh in 2024
LBNL's June 2026 energy usage update estimates that U.S. data centers used 192 TWh of electricity in 2024, equal to 4.7% of national electricity consumption. This updated historical estimate includes servers, storage, networking, and supporting facility infrastructure, but excludes cryptocurrency mining. The study uses 2024 as its last fully historical year because some 2025 equipment-shipment data remained incomplete.
2. EU data centers are projected to consume 114 TWh by 2030
The European Commission estimates that EU data centers consumed 68 TWh in 2024 and projects 114 TWh in 2030. These are electricity consumption figures for EU data centers, with the 2030 value representing a forecast. Thermal efficiency improvements can help moderate the facility overhead accompanying additional computing demand.
3. Data centers could drive 33% of U.S. load growth from 2024 to 2030
LBNL's 2026 Reference Case, considered alongside the U.S. electricity demand outlook in the 2025 North American Electric Reliability Corporation assessment, implies data centers could account for approximately 33% of total U.S. electricity load growth between 2024 and 2030. The researchers make this estimate conditional on the two studies' assumptions about future electricity use being consistent. The figure describes modeled incremental load growth, not data centers' share of all U.S. electricity consumption.
4. Eight primary U.S. markets had 5,994.4 MW under construction at year-end 2025
CBRE reported 5,994.4 MW of data center capacity under construction at the end of 2025 across its eight primary U.S. markets, compared with 6,350.1 MW at year-end 2024. These numbers describe a defined market sample and a construction pipeline. They are not national operating power demand or measured cooling electricity consumption.
5. U.S. data center use could reach 521-843 TWh in 2030
LBNL's June 2026 update places its 2030 U.S. data center electricity Reference Case at 649 TWh, with a broader 521-843 TWh compounded uncertainty range generated by combining the extremes of several modeling assumptions. These projections cover total data center electricity, including IT equipment and facility infrastructure. The bounds are stress-test scenarios rather than measured consumption, a confidence interval, or an estimate of electricity savings achievable solely through cooling optimization.
6. U.S. data centers could use 9.5-15.3% of national electricity in 2030
LBNL's 2026 update forecasts that U.S. data centers could consume 9.5-15.3% of the country's total electricity in 2030 under its compounded uncertainty scenarios, with 11.8% in its Reference Case. For comparison, its revised historical estimate is 4.7% in 2024. These figures describe a national electricity share, not cooling energy share or individual facility PUE, and rely on projected U.S. electricity demand as the denominator.
7. EU data centers may account for 3.2% of electricity demand by 2030
In its September 2026 report, the European Commission projects data center electricity consumption equivalent to 3.2% of the EU's electricity demand by 2030. This is a regional projection, not a measured percentage for 2026. It provides a policy and infrastructure context for reducing avoidable cooling overhead throughout European data center operations.
Breaking Down Cooling Energy and Infrastructure Overhead
Cooling, power distribution, lighting, and other infrastructure contribute to non-IT electricity consumption. Distinguishing each component prevents overstatement of the savings available from cooling control alone.
8. Cooling can account for up to 40% of data center energy
DOE identifies cooling as a substantial energy load and notes that it can account for up to 40% of total data center energy use. The actual share depends on facility design, weather, IT load, equipment efficiency, and operating strategy. The figure establishes the potential significance of cooling optimization without assigning the same baseline to every U.S. or European facility.
9. A hypothetical PUE of 1.36 implies 360 W of overhead per IT kW
The reported EU sample PUE of 1.36 provides a numerical illustration. Using the DOE PUE definition, a hypothetical site operating at PUE 1.36 with 1 kW of average IT power uses 1.36 kW of total facility power. Its non-IT infrastructure overhead is therefore (1.36 - 1.00) x 1,000 W = 360 W. Those 360 W include cooling and other non-IT loads. The calculation illustrates the metric; it is not a measured cooling-load breakdown for a particular EU facility.
10. Servers represented an estimated 7% of U.S. commercial electricity in 2025
The U.S. Energy Information Administration (EIA) estimates that data center servers alone accounted for 7% of commercial-sector electricity consumption in 2025. This figure refers to server electricity, not total data center electricity or cooling energy. The distinction matters when calculating thermal infrastructure overhead in commercial buildings.
11. Server electricity could reach 22-33% of U.S. commercial use by 2050
Across the EIA's Annual Energy Outlook 2026 cases, data center servers are projected to account for 22-33% of U.S. commercial-building electricity use by 2050. This is a scenario range for server demand, not an observed outcome and not a cooling forecast. It reinforces the importance of thermal control that adapts as equipment populations and compute workloads evolve.
Power Usage Effectiveness: What PUE Does and Does Not Measure
Power Usage Effectiveness (PUE) is the ratio of total data center facility energy to IT equipment energy over a consistent reporting period. A lower PUE indicates less non-IT energy overhead per unit of IT energy. Cooling optimization can reduce that overhead, although PUE does not separately report cooling-system electricity consumption.
12. An initial EU reporting sample recorded 1.36 weighted-average PUE
An April 2026 analysis of the European Commission's initial reporting data identified a weighted-average PUE of 1.36. The submissions covered 18 EU countries and information for 2023 and 2024, with incomplete participation. Accordingly, 1.36 characterizes the reported dataset, not the average of all EU data centers.
Synapse Thermal's PUE opportunity. Synapse identifies 1.03 as a potential optimal PUE for its cooling optimization approach, not as an already achieved outcome or guaranteed facility result. Synapse's value proposition extends beyond a PUE target: its unified cooling-focused DRL agent uses local inference to coordinate thermal control across supported air, chilled-water, direct-to-chip liquid, and immersion systems. Its complimentary efficiency analysis estimates facility-specific cooling energy and cost improvement potential using passive telemetry.
Data Center Cooling Market Size and Investment in the U.S. and Europe
Cooling market revenue measures spending on equipment, systems, and related services under the research publisher's defined scope. The regional figures attributed to Global Market Insights below use its September 2026 report edition. Market revenue helps describe demand for thermal infrastructure, but it is not a direct estimate of electricity waste or achievable software savings.
13. The U.S. cooling market was estimated at $7.99 billion in 2025
A September 2026 study by Global Market Insights estimates $7.99 billion in U.S. data center cooling market revenue for 2025. This is a publisher-produced market estimate, rather than directly metered operating expenditure or an estimate of inefficiency.
14. U.S. cooling revenue could reach $23.70 billion by 2035
The same 2026 market study forecasts the U.S. cooling market at $23.70 billion in 2035. It is projected spending within the publisher's market definition, not revenue already realized. For operators, greater equipment diversity also creates opportunities for coordinated operation of existing and new thermal assets.
15. The UK cooling market could increase to $3.02 billion by 2035
Global Market Insights estimates the United Kingdom's data center cooling market at $1.08 billion in 2025 and forecasts $3.02 billion in 2035. These are UK-specific figures, separate from broader European market totals, and both values reflect the report's market boundaries.
16. EIA models up to 2.9 times greater cooling intensity in data center floorspace
In its 2026 projections, EIA assumes that space-cooling requirements in data center floorspace can be as much as 2.9 times as energy intensive as in non-data-center floorspace, on average. This is a modeling assumption about cooling intensity, not a measured ratio for every facility and not a claim that 2.9 times as much total data center electricity is used for cooling.
Why Synapse's control architecture matters. Fixed setpoints, rule-based sequencing, proportional-integral-derivative (PID) control, model predictive control, and learned control policies offer different ways to manage thermal equipment. Synapse focuses on closed-loop DRL, using operating feedback to coordinate control decisions across interacting cooling variables. Its hardware-agnostic design allows operators to pursue optimization with existing HVAC, CRAC, chilled-water, direct-to-chip liquid, and immersion infrastructure. Synapse states a 4-6 week deployment timeframe without wholesale equipment replacement, subject to project scope.
AI Cooling Optimization and the Value of Coordinated Control
AI-powered optimization connects operating measurements with changing thermal requirements. The most useful comparisons separate energy cost, cooling electricity, total facility energy, and PUE rather than treating these as interchangeable metrics.
17. Synapse reports potential cooling cost savings up to 40%
Synapse Thermal states that its autonomous DRL approach can reduce cooling energy costs by up to 40% while maintaining thermal constraints. This is a company-stated potential tied to cooling operating costs and site conditions. It is distinct from the DOE estimate that cooling itself can account for up to 40% of facility energy and does not imply an automatic 40% reduction in total data center electricity or PUE.
Synapse's cooling-focused agent is designed to combine real-time monitoring, LSTM-based thermal forecasting, multivariable DRL optimization, and ongoing learning. That combination supports adaptive control of fan speeds, chilled-water flow rates, and supported liquid cooling loops while prioritizing thermal performance.
18. Eight primary U.S. markets absorbed 2,497.6 MW in 2025
CBRE reports that the eight primary U.S. data center markets registered 2,497.6 MW of net absorption during 2025, versus 1,809.5 MW in 2024, an increase of approximately 38%. Net absorption is a real estate market measure of capacity taken up by users. It does not represent cooling energy consumption, but it indicates the scale of data center growth that thermal infrastructure must serve.
19. Vacancy reached 1.4% across primary U.S. markets in 2025
CBRE recorded a 1.4% vacancy rate at year-end 2025 across its eight primary U.S. data center markets. This capacity indicator reflects constrained space availability in the surveyed markets, not a measured percentage of facilities implementing AI cooling optimization. Efficient cooling control can support thoughtful use of installed infrastructure as computing requirements change.
Cooling Technology: Air, Chilled Water, Direct-to-Chip, and Immersion
Higher rack power densities increase the importance of matching heat-removal technologies to server configurations. Many U.S. and European facilities operate mixed cooling environments, making supervisory optimization across equipment categories especially relevant.
20. Some U.S. AI installations produce 60-100 kW per rack
A 2026 U.S. liquid cooling study describes high-density AI deployments with 60-100 kW of heat per rack. This range refers to specific demanding installations, not every AI rack or a universal cutoff between air and liquid cooling. Heat load, airflow, supply temperatures, and liquid cooling architecture determine the appropriate design.
21. The U.S. liquid cooling market reached an estimated $1.30 billion in 2025
Emergen Research estimates $1.30 billion in 2025 revenue for the U.S. liquid cooling systems market. This is a specific market segment estimate, not the share of data centers with liquid cooling installed. The growth of liquid systems makes coordination with upstream chillers, pumps, and heat rejection increasingly important.
22. EIA projects an 84 TWh commercial cooling difference between scenarios by 2050
In its Annual Energy Outlook 2026, EIA estimates that U.S. commercial-sector space-cooling electricity in the High Electricity Demand case is 84 billion kWh, or 84 TWh, higher in 2050 than in the Counterfactual Baseline case. The difference partly reflects more intensive data center operations. It is a gap between scenarios, not an available 84 TWh cooling savings opportunity or a forecast of energy saved through liquid cooling.
23. U.S. liquid cooling revenue could reach $7.33 billion by 2035
Emergen Research forecasts $7.33 billion in 2035 U.S. liquid cooling systems revenue, compared with its $1.30 billion estimate for 2025. The forecast applies specifically to the U.S. market and the publisher's liquid cooling category.
Synapse's unified cooling approach. Synapse Thermal positions its hardware-agnostic DRL agent to coordinate air cooling, chilled-water systems, direct-to-chip liquid cooling, and immersion cooling through one cooling-focused optimization strategy. Its on-premises inference keeps mission-critical model execution local and independent of a cloud connection. This combination supports integrated thermal management as facilities add high-density equipment alongside existing cooling assets.
High-Density AI Infrastructure: Cooling Targets and Market Expansion
Rack heat load, cooling research targets, data center inventory, and facility electricity use are distinct measures. Distinguishing these values helps operators relate AI infrastructure expansion to thermal energy requirements.
24. DOE targets cooling energy below 5% of IT load in high-density systems
The U.S. Department of Energy's ARPA-E COOLERCHIPS research program targets total cooling energy use below 5% of IT energy use for high-density data center systems operating in U.S. locations. Its definition of cooling energy includes the energy needed to reject heat generated by IT equipment and non-IT infrastructure to the ambient environment. The 5% figure is a research objective, not a measured national average, a completed deployment result, or an equivalent PUE of 1.05. The target illustrates the emphasis placed on efficient end-to-end heat rejection as computing density increases.
25. U.S. server electricity could reach 446-818 TWh by 2050
Across its Annual Energy Outlook 2026 cases, EIA projects 446-818 billion kWh, equivalent to 446-818 TWh, of annual U.S. data center server electricity consumption in 2050. The range excludes cooling and other infrastructure electricity from the server total. Cooling investment and optimization must therefore be evaluated against both IT power growth and facility overhead.
26. Northern Virginia inventory reached 4,039.6 MW in 2025
CBRE reported 4,039.6 MW of total data center inventory in Northern Virginia at year-end 2025, a 37% year-over-year increase. The same study records 1,102 MW of net absorption in Northern Virginia during 2025. These measures describe a specific U.S. data center market, rather than cooling system energy consumption.
For variable AI workloads, Synapse states that its LSTM neural networks forecast thermal load trajectories 15-30 minutes ahead. Its DRL controller uses this forward-looking information alongside temperatures, environmental conditions, workload signals, and energy consumption to select cooling actions within thermal constraints. The forecast horizon describes the company's published architecture, not a fixed physical cooling response time.
The Financial Case for Intelligent Cooling
Cooling optimization economics depend on baseline equipment operation, electricity prices, IT demand, and the proportion of facility overhead attributable to cooling. Facility-specific calculations provide a disciplined way to distinguish modeled savings from measured market data.
27. An EU reporting sample represented 6.4 GW of installed IT power
An April 2026 account of the European Commission's initial reporting tranche identified 6.4 GW of installed IT power among 776 reported data centers. The submissions spanned 18 EU countries and included information for 2023 and 2024. This is an incomplete reporting dataset, not a census of EU data center capacity or a measurement of cooling-only electricity.
The DOE PUE equation can illustrate operating savings. For a hypothetical U.S. facility with 50 MW of constant IT power, PUE improving from 1.40 to 1.30, 8,760 operating hours annually, and an illustrative electricity price of $0.10/kWh, the annual facility electricity cost reduction would be $4.38 million:
50,000 kW x 0.10 x 8,760 hours x $0.10/kWh = $4,380,000.
This is an illustrative reduction in total non-IT electricity overhead, not a reported customer outcome and not automatically a cooling-only saving. A cooling-specific business case separately measures baseline cooling electricity use and achievable operating changes.
28. Europe's cooling market was estimated at $4.51 billion in 2025
Global Market Insights estimates $4.51 billion in 2025 European data center cooling market revenue and forecasts an 11.3% compound annual growth rate (CAGR) during 2026-2035. These figures are specific to the European data center cooling market.
Synapse Thermal states that customers can achieve investment payback in 8-14 months through energy savings alone. This is a published company-stated potential that provides a framework for evaluating site-specific savings against project investment. Synapse also states a 4-6 week deployment timeframe, supported by its hardware-agnostic approach to existing infrastructure. Its complimentary DRL efficiency analysis uses passive facility telemetry to project potential cooling energy reductions and annual electricity cost savings.
Cooling Priorities for Enterprise, Hyperscale, and Colocation Facilities
Operator needs differ, but all three segments can benefit from optimization that jointly considers equipment operation, thermal limits, and site economics.
29. Germany's cooling market may grow at 11.9% CAGR through 2035
Global Market Insights forecasts an 11.9% CAGR from 2026 to 2035 for the German data center cooling market. This is a market revenue forecast for Germany, not a statement of verified cooling energy savings or an individual supplier's market share.
30. U.S. liquid cooling revenue may grow at 18.9% CAGR through 2035
Emergen Research projects an 18.9% revenue CAGR over 2026-2035 for U.S. liquid cooling systems. This forecast is specific to liquid cooling, rather than the entire data center cooling market.
Techstars-backed Synapse Thermal highlights its EPC channel and Southwest U.S. footprint, including New Mexico, bringing a cooling optimization perspective relevant to operators managing hot and dry climate conditions. Its cooling-first control architecture is relevant to the operating priorities of three facility types:
- Enterprise data centers: A unified, hardware-agnostic DRL strategy supports more efficient operation of existing HVAC, CRAC, chilled-water, and supported liquid cooling equipment while preserving investment in installed assets.
- Hyperscale data centers: At large IT loads, even relatively small reductions in cooling power can produce material annual cost savings. Synapse's local inference and autonomous multivariable control are suited to the challenge of changing thermal demand across complex facilities.
- Colocation data centers: Mixed rack densities and tenant requirements create a need for coordinated thermal management. Synapse's cooling-focused approach supports optimization across supported cooling types while keeping operating actions aligned with facility thermal constraints.
The Future of Cooling: European Efficiency Policy and On-Premises AI
In September 2026, the European Commission adopted a delegated regulation establishing a common sustainability rating scheme for individual data centers with capacity above 500 kW. The regulation remained subject to a two-month scrutiny period by the European Parliament and Council before entry into force, with first individual sustainability labels expected in 2027. The Commission also described a waste-heat opportunity: reusing roughly half of the waste heat generated by European data centers could theoretically meet the heating demand of 4 million households. This is potential energy reuse, not documented heat already supplied to those households.
Operational efficiency and data privacy are closely connected to how AI cooling software is deployed. Synapse Thermal states that its edge AI architecture performs inference on-premises with microsecond-scale model latency, keeping the mission-critical inference path independent of cloud connectivity. Model inference latency is distinct from the time required for sensor communications and mechanical cooling equipment to respond.
Synapse's four-stage cooling optimization approach combines real-time monitoring, thermal prediction, autonomous cooling adjustments, and continuous learning. Its closed-loop DRL policy is designed to minimize cooling energy while maintaining thermal constraints. The company's cooling-only focus brings these capabilities together across the data hall and supporting thermal equipment, rather than treating each cooling subsystem as an isolated optimization problem.
Its complimentary cooling analysis provides an initial, passive-telemetry-based view of facility-specific savings potential. Projected results from this analysis can inform the scope and economics of autonomous optimization on existing equipment.
Frequently Asked Questions
What percentage of data center energy goes to cooling?
DOE says cooling can account for up to 40% of a data center's energy use. This is an upper-end figure, with the actual percentage determined by the site's cooling architecture, IT load, climate, and operations.
How does autonomous AI improve data center cooling efficiency?
Autonomous AI cooling uses sensor data and workload information to predict thermal conditions and coordinate variables such as fan speed, chilled-water flow, and liquid loop operation. Synapse Thermal uses a cooling-focused DRL agent with continuous learning and on-premises inference to optimize supported equipment in a closed loop. Synapse states potential cooling energy cost savings of up to 40%, with the opportunity determined by site conditions.
What is Power Usage Effectiveness (PUE)?
PUE is total facility energy divided by IT equipment energy over the same reporting period. A theoretical PUE of 1.0 would mean no non-IT overhead. For example, a hypothetical reduction from 1.50 to 1.40 PUE at constant IT load represents a 20% decrease in non-IT overhead and approximately a 6.7% reduction in total facility energy. Those percentages do not isolate cooling electricity. Synapse identifies 1.03 as a potential optimal PUE, rather than an already achieved or guaranteed result.
Can DRL cooling optimization work with existing infrastructure?
Yes. Synapse states that its hardware-agnostic platform supports existing HVAC and CRAC systems, chilled-water infrastructure, direct-to-chip liquid cooling, and immersion systems. Its unified cooling-focused agent is designed to coordinate supported equipment categories without requiring wholesale infrastructure replacement. Synapse publishes a 4-6 week deployment timeframe for its approach.
What is the financial benefit of cooling optimization?
Lower cooling energy use can reduce electricity spending and contribute to lower non-IT facility overhead. The calculation depends on cooling-specific baseline consumption, local electricity prices, and the operational improvement achieved. Synapse reports potential cooling energy cost savings of up to 40% and a company-stated 8-14 month investment payback through energy savings. Its complimentary efficiency analysis develops facility-specific savings projections.
How does AI cooling respond to high-density GPU workloads?
Predictive AI cooling responds to changes in measured power, temperature, workload, and ambient conditions. Synapse says its LSTM models predict thermal trajectories 15-30 minutes ahead and supply those forecasts to its DRL policy. The policy coordinates supported cooling controls while maintaining thermal constraints, making it relevant to mixed air and liquid cooling environments serving changing AI workloads.
How is Synapse Thermal different from conventional cooling control?
Conventional PID loops regulate individual variables based on feedback error, while rule-based sequences follow predefined conditions. Synapse Thermal combines autonomous closed-loop DRL, predictive thermal modeling, local model inference, and hardware-agnostic control across multiple cooling technologies. Its cooling-focused strategy unifies supported air, chilled-water, direct-to-chip, and immersion systems within a coordinated optimization approach, with published company-stated deployment and payback timeframes.