35 PUE Statistics That Define Data Center Energy Efficiency in 2026
All postsIndustry Insights

35 PUE Statistics That Define Data Center Energy Efficiency in 2026

Synapse Thermal·October 15, 2026

35 U.S. and European statistics on data center PUE, electricity demand, and cooling overhead, framing AI-driven DRL optimization as a site-specific efficiency opportunity.

U.S. and European data showing why power usage effectiveness optimization through AI-driven cooling is an important opportunity for data center operators

Lawrence Berkeley National Laboratory's June 2026 report estimates that modeled U.S. national power usage effectiveness (PUE) improved from 1.55 in 2018 to 1.45 in 2024. This improvement demonstrates real progress, while leaving scope to reduce the electricity used by cooling, power distribution, and other facility infrastructure. U.S. data centers consumed an estimated 192 TWh, or 4.7% of national electricity, in 2024. In the European Union, electricity use was estimated at 70 TWh in 2024. At the other end of the efficiency spectrum, Google's Lancaster, Ohio campus reported a trailing-12-month PUE of 1.04 in Q2 2026. These figures describe different facility populations and should not be treated as a controlled comparison or a universally achievable savings target. They do, however, show why operators should examine site-specific overhead loads. Synapse Thermal addresses controllable cooling overhead with a single cooling-focused, closed-loop deep reinforcement learning (DRL) agent. Its on-premise inference and hardware-agnostic approach connect existing air, chilled-water, direct-to-chip liquid and immersion cooling, with no cloud dependency in the mission-critical control loop.

Key Takeaways

  • U.S. PUE continues to improve, but overhead remains material: Lawrence Berkeley National Laboratory (LBNL) modeled national average PUE improving from 1.55 in 2018 to 1.45 in 2024. The evidence does not establish that all conventional cooling methods have reached a hard limit.
  • PUE results differ across facility populations: Google's Fredericia, Denmark campus reported 1.06 trailing-12-month PUE in Q2 2026, while LBNL's U.S. average is a model covering a much broader facility mix. A single campus result is not a hyperscale industry average.
  • Cooling optimization deserves attention: LBNL estimates that facility infrastructure, including cooling and electrical losses, used 31% of U.S. data center electricity in 2024. Synapse's predictive DRL controls target the controllable cooling portion, not that entire 31%.
  • Electricity demand is accelerating: LBNL's 2030 scenarios place U.S. data center demand at 9.5%-15.3% of national electricity, with 11.8% in its reference case. The European Commission, citing IEA estimates, reports EU consumption could reach 115 TWh by 2030.
  • Compliance requirements differ by jurisdiction: Germany sets defined PUE limits for qualifying older and newer data centers. EU rules require reporting by data centers with at least 500 kW of installed IT power demand, but had not established a general EU-wide PUE limit of 1.1 as of October 2026.
  • AI workloads intensify thermal-management demands: LBNL projects that AI servers could use 55% of all U.S. data center electricity in 2030 under its reference case. Increasing power density makes coordinated, constraint-aware cooling more important.
  • Synapse combines autonomous cooling capabilities: Its cooling-first DRL agent brings closed-loop optimization, on-premise inference and integration with existing air, chilled-water, direct-to-chip and immersion systems together. Synapse identifies 1.03 as a potential optimal PUE, rather than a measured or guaranteed result.

Understanding Power Usage Effectiveness (PUE) in Data Centers

What is PUE?

Power Usage Effectiveness is total data center facility energy divided by IT equipment energy, measured over a consistent interval and system boundary. A PUE of 1.0 is an idealized limit in which no energy is consumed by non-IT infrastructure. At LBNL's modeled U.S. average of 1.45 for 2024, non-IT overhead equals 45% of IT energy, or approximately 31% of total facility energy. That overhead is not exclusively cooling.

1. U.S. national average PUE was modeled at 1.45 in 2024

LBNL's June 2026 analysis estimates a 1.45 national average PUE for 2024. This is a modeled national aggregate, not an average measured directly at every U.S. facility. It offers a geographically relevant baseline for assessing potential reductions in non-IT overhead.

2. Modeled U.S. average PUE improved from 1.55 in 2018 to 1.45 in 2024

The 2018-to-2024 improvement reflects changes in the U.S. data center mix, including greater use of highly efficient facilities serving AI equipment. It is evidence of progress, not proof that any single cooling intervention caused the entire change.

3. U.S. aggregate PUE is projected to reach 1.36 in 2030

LBNL's reference-case model projects a U.S. aggregate PUE of 1.36 by 2030. This is a forecast dependent on the modeled facility mix and infrastructure assumptions, not a guaranteed outcome. Further site-level improvement may require coordinated optimization of existing cooling and power systems.

Why PUE Matters for Data Center Operations

PUE links facility overhead to the energy delivered to IT equipment. For an illustrative data center with a constant 20 MW IT load, 8,760 operating hours per year, and an electricity price of $0.10/kWh, reducing PUE by 0.10 avoids 17.52 GWh of annual facility electricity use and saves $1.752 million annually. Under those same assumptions, moving from a PUE of 1.45 to 1.10 saves 61.32 GWh, or $6.132 million per year, in total facility overhead electricity, not necessarily cooling electricity alone. These are arithmetic scenarios using the PUE energy ratio; actual savings and CO₂ effects depend on the IT load, control changes, local tariff, operating hours, and electricity emissions factors.

Current Data Center Energy Consumption Statistics and PUE Averages

U.S. and European Data Center Energy Footprint

4. EU data centers used an estimated 70 TWh in 2024

The European Commission's November 2025 overview, citing an IEA estimate, places EU data center electricity consumption at 70 TWh in 2024. This estimate is for the European Union, not all European countries, and reflects the limitations of available reporting data.

5. U.S. data centers used an estimated 192 TWh, or 4.7% of electricity, in 2024

LBNL's updated historical model reports 192 TWh and 4.7% for 2024. The updated estimate provides a more current, U.S.-specific indication of the sector's electricity footprint.

6. U.S. data centers could use 9.5%-15.3% of national electricity in 2030

LBNL's June 2026 work models 521-843 TWh in compounded-uncertainty scenarios, versus 649 TWh and 11.8% in its reference case. These are forecasts, not observed 2030 consumption or a statistical confidence interval. Actual outcomes depend on AI equipment deployment, usage and other assumptions.

7. A study of 403 U.S. hyperscale facilities estimates 68-99 TWh during May 2024 to April 2025

A June 2026 research preprint estimates that 403 identified hyperscale sites in the contiguous United States, rather than the complete U.S. hyperscale fleet, consumed 68-99 TWh in the specified 12-month period. Its central modeled scenario is approximately 82 TWh. The range changes with facility utilization assumptions and is not a directly metered national total.

The Cost of Inefficiency

8. The studied U.S. hyperscale sample had estimated carbon intensity about 48% above the grid benchmark

For its 403-facility sample, the preprint estimates an electricity-weighted attributed carbon intensity of approximately 545 gCO₂/kWh, compared with an approximately 370 gCO₂/kWh U.S. grid benchmark based on 2023 data. This is a model-based comparison using the research team's attribution framework, not a measured emissions intensity for every U.S. hyperscale facility.

9. The sampled U.S. hyperscale facilities were associated with an estimated 37-54 million metric tons of CO₂

For May 2024 to April 2025, the study models 37-54 million metric tons of CO₂ attributable to electricity used by the 403 sampled facilities, with roughly 45 million metric tons in its central scenario. These are attributed electricity emissions, not direct on-site emissions or the total for all U.S. data centers.

10. Fossil fuels supplied an estimated 53.9% of electricity attributed to the studied hyperscale sites

The 2026 U.S. hyperscale preprint attributes 53.9% of the study sample's electricity to fossil-fuel generation using balancing-authority generation mixes. This estimate is sensitive to the attribution method. It also demonstrates why emissions reductions from better cooling must be calculated using a defensible regional electricity factor rather than a generic global assumption.

How AI-Powered Systems Drive Data Center Efficiency and PUE Improvement

The Limitations of Traditional Cooling

Fixed cooling setpoints and non-predictive rule-based controls may over-provision cooling when workloads and ambient conditions vary. Basic PID loops react to control error, although modern conventional systems can also employ supervisory control, predictive features, scheduling and adaptive tuning. The relevant comparison for autonomous optimization is the actual site's current control strategy, not an assumption that every conventional system is incapable of adaptation.

11. Facility infrastructure represented 31% of modeled U.S. data center electricity use in 2024

LBNL includes cooling systems and power distribution losses within this 31% facility-infrastructure category. It does not attribute the entire category to cooling or claim that AI optimization can eliminate it. The figure identifies why measuring baseline cooling loads is essential before estimating savings.

Mechanisms of DRL for Thermal Management

Synapse Thermal describes a cooling-focused DRL agent that observes server temperatures, ambient conditions, workload metrics and energy use, applies cooling adjustments and learns from operational feedback. Synapse states that it uses LSTM neural networks to forecast thermal loads 15-30 minutes ahead, informing cooling actions while maintaining thermal constraints. Unlike a monitoring-only application, this is a closed-loop optimization approach that can write configured cooling setpoints.

The platform combines on-premise inference, with no cloud dependency in the mission-critical control loop, and hardware-agnostic integration with existing air-cooling equipment, chilled-water infrastructure, direct-to-chip liquid cooling and immersion systems. Synapse states that its DRL algorithm is 35-40% more energy efficient than traditional PID and rule-based controllers. It also advertises deployment in approximately 4-6 weeks without wholesale equipment replacement. That controller comparison and the company's separately stated potential cooling-cost savings are different measures from a facility-wide PUE reduction.

12. McKinsey projected annual U.S. data center capacity-demand growth of 20% to 25% through 2030

In an August 2025 analysis of U.S. data center development, McKinsey estimated that demand for U.S. data center capacity could grow by 20% to 25% per year through 2030. This is a forecast of capacity demand, not measured electricity consumption or a quantified effect of cooling optimization. The potential scale of new installations increases the value of site-specific efficiency analysis and coordinated cooling controls.

Achieving Significant Reductions in Data Center Cooling Energy Consumption

Performance Leaders Set the Benchmark

13. Google's Columbus, Ohio campus reported 1.06 trailing-12-month PUE in Q2 2026

The Columbus campus result of 1.06 is a published, location-specific efficiency benchmark. It does not represent Google's entire fleet or establish that one particular cooling algorithm produced that performance.

14. Google's Lancaster, Ohio campus reported 1.04 trailing-12-month PUE in Q2 2026

Google's Lancaster campus recorded 1.04 for the specified reporting period, approaching the theoretical PUE floor of 1.0. This illustrates a highly efficient individual U.S. campus, not a universally achievable result for existing facilities.

15. Google's Fredericia, Denmark campus reported 1.06 trailing-12-month PUE in Q2 2026

The Fredericia result supplies a European benchmark with an identified location and reporting period. It illustrates low facility overhead at an individual hyperscale campus, rather than establishing a market-wide average or an achievable target for every site.

16. Google's Eemshaven, Netherlands campus reported 1.07 trailing-12-month PUE in Q2 2026

Google's Eemshaven result of 1.07 provides another verifiable European example. Individual campus performance cannot establish the proportion of U.S. or European data centers operating below any PUE threshold.

17. U.S. facilities serving AI equipment had a modeled average PUE of 1.145 in 2024

LBNL estimates 1.145 PUE for the U.S. facility category serving AI equipment. This is a technology-specific model result, not an average for all recently commissioned data centers or a measured figure for each liquid-cooled installation.

18. U.S. facilities serving AI equipment are modeled to reach 1.136 PUE in 2030

LBNL projects 1.136 PUE in 2030 for facilities serving AI equipment. This differs from its 1.36 projected aggregate across the broader U.S. data center fleet; neither should be described as an observed 2030 outcome or as a universal benchmark for large facilities.

Types of Data Center Cooling Systems and AI Integration

Air Cooling Technologies

Computer Room Air Conditioning (CRAC) units, airflow management and other HVAC infrastructure remain important in many facilities. Air-cooled designs can be efficient, but a facility's PUE depends on climate, heat rejection, control strategy, IT utilization and measurement boundaries. As the National Laboratory of the Rockies' PUE methodology illustrates, comparisons should account for the full facility energy boundary rather than assume a single PUE for all air-cooled designs.

Liquid Cooling Innovations

19. Warm-water liquid cooling can support rack densities of 60 kW or more, according to a U.S. national laboratory

The National Laboratory of the Rockies (NLR) describes 60 kW per rack or higher as an achievable density with warm-water liquid cooling. This is a stated engineering capability, not an independently measured average across liquid-cooled data centers. Direct-to-chip cold plates can extract heat from processors into a liquid loop, but the resulting cooling-energy savings depend on the existing plant, captured heat fraction, coolant conditions and heat-rejection design.

20. Water in NLR's liquid-cooled HPC installation is heated to around 100°F by computing equipment

At NLR's U.S. facility, cooling water reaches approximately 100°F after collecting waste heat, enabling reuse for laboratory and office heating. Immersion cooling offers another approach, using dielectric liquid around compatible server components. Its actual cooling-energy impact must be assessed against a defined baseline and include the complete liquid circulation and heat-rejection system.

21. Facility infrastructure's share of modeled U.S. data center electricity fell from 36% in 2018 to 31% in 2024

LBNL attributes the declining infrastructure share primarily to a shift of server electricity use toward more efficient facilities, including liquid-cooled AI infrastructure. This five-percentage-point change reflects shifts in the national equipment and facility mix, not an independently measured savings rate from liquid cooling alone.

Seamless Integration with Existing Cooling Systems

Synapse Thermal says its hardware-agnostic platform supports integration with existing HVAC and CRAC equipment, chilled-water plants, direct-to-chip liquid-cooling infrastructure and immersion systems. Its single cooling-focused DRL agent is designed to coordinate these four cooling categories through on-premise, closed-loop inference rather than separating the optimization task into unrelated cooling models. Synapse describes a 4-6-week deployment on existing equipment without wholesale equipment replacement. For operators prioritizing existing assets, predictive DRL, local control and the complimentary baseline assessment form a distinctive retrofit-first proposition. Synapse is Techstars-backed and describes EPC channel relationships that support its commercial approach. Its U.S. Southwest and New Mexico footprint adds regional relevance for operators managing cooling in hot, dry climates.

Optimizing PUE for High-Heat GPU Workloads in AI Data Centers

The Challenge of AI-Generated Heat

22. AI servers could use 84% of U.S. data-center server electricity in 2030

Under its reference case, LBNL projects AI servers accounting for 84% of U.S. data-center server electricity use in 2030. This is a modeled electricity share, not the percentage of data center owners currently conducting AI training or inference.

GPU-dense equipment can impose high rack heat loads and shifting cooling demands as training and inference activity changes. The cooling design must account for server power, load distribution, coolant temperatures, airflow and redundancy. Actual rack heat load depends on accelerator model, equipment utilization and rack configuration; a single heat multiplier does not characterize all U.S. and European AI deployments.

23. AI servers could account for 55% of total U.S. data center electricity in 2030

LBNL's 55% reference-case projection uses all facility electricity as the denominator, unlike the 84% server-only share above. The forecast shows that gains in chip-level computation per watt do not automatically prevent growth in total power demand.

Where Fixed-Rule Cooling Can Struggle with GPU Clusters

Fixed-rule cooling systems can struggle to follow rapid or uneven thermal-load changes without conservative setpoints. Advanced conventional controls may already address some of these variations. Synapse describes forecasting loads 15-30 minutes ahead to support proactive control of configured cooling resources while respecting thermal limits. Its on-premise DRL agent combines predictive actions across air, chilled-water and liquid-cooling loops, making the existing control strategy the practical baseline for evaluating results.

The Role of Data Center Cooling Towers in PUE Reduction Strategies

Cooling Tower Fundamentals in Data Centers

Where a facility uses evaporative heat rejection, cooling towers dissipate heat from the cooling-water loop. Fan speeds, water flow and operating approach temperatures affect energy use, but alternative plant designs may use dry coolers or other heat-rejection equipment. Cooling tower optimization must therefore follow the facility's actual topology.

24. Google's Hamina, Finland campus reported 1.09 trailing-12-month PUE in Q2 2026

Google's Hamina campus result gives a named, European example of low overhead energy use. Favorable outdoor conditions can support efficient heat rejection and economizer operation, but this one campus value does not show that latitude alone caused the result or that all high-latitude sites achieve similar PUE.

Integrating Cooling Towers into an AI-driven Strategy

Synapse Thermal's four-stage intelligent cooling system combines real-time monitoring, predictive analytics, dynamic optimization and continuous learning. Its published control approach includes automated adjustment of fan speeds, chilled-water flow rates and liquid-cooling loops. For facilities using cooling towers or other heat-rejection equipment, fan, flow and water-temperature conditions inform the cooling optimization strategy. Within its configured control scope, Synapse's on-premise DRL agent coordinates cooling actions while maintaining thermal constraints.

Pathways to Green Data Centers Through PUE Optimization

Sustainability Imperatives for Data Centers

25. Modeled U.S. data center electricity consumption increased 14% from 2023 to 2024

LBNL's June 2026 conclusion reports a 14% increase in U.S. data center electricity use between 2023 and 2024. This regional trend reinforces the need to track facility loads, but it does not measure operators' attitudes toward sustainability reporting.

26. EU data center reporting applies at an installed IT power demand of at least 500 kW

Delegated Regulation (EU) 2024/1364 requires operators of covered data centers with installed information technology power demand of at least 500 kW to submit information and key performance indicators annually to the European database. The Commission publishes aggregated indicators through its reporting dashboard. The 500 kW threshold refers to installed IT power demand, not total facility power or cooling capacity. It does not indicate what share of operators voluntarily tracks PUE.

27. LBNL's U.S. reference case models 22% electricity-use growth from 2024 to 2025

The 22% reference-case increase is from LBNL's 2026 report. The study treats 2024 as its last historical year and includes projections for 2025, so this figure is not a final observed annual growth rate.

PUE as a Key Sustainability Metric

28. LBNL's U.S. reference case projects 29% electricity-use growth from 2025 to 2026

The 29% increase projected between 2025 and 2026 is a forecast, not a measured 2026 result. It highlights why PUE should be reported alongside absolute electricity consumption: a lower PUE can coexist with higher facility energy demand as IT loads expand.

Regulatory Requirements Driving Action

29. Germany requires PUE of 1.5 from July 2027 and 1.3 from July 2030 for qualifying older data centers

Under Section 11(1) of Germany's Energy Efficiency Act, data centers that began operating before July 1, 2026 must achieve annual-average PUE no higher than 1.5 from July 1, 2027 and no higher than 1.3 from July 1, 2030, within the statute's applicable scope. These are German requirements, not blanket limits for all European facilities.

30. Germany sets a PUE limit of 1.2 for qualifying facilities entering operation from July 2026

Section 11(2) requires data centers beginning operation on or after July 1, 2026 to achieve annual-average PUE no higher than 1.2, with the applicable requirements reached within two years after commissioning. The section also specifies heat-reuse provisions and conditions. By contrast, the European Commission was consulting on possible EU-wide minimum performance standards in September 2026; no general EU-wide PUE mandate of 1.1 had been adopted as of the article's October 2026 date.

Estimating and Achieving PUE Optimization with Complimentary Efficiency Analysis

The Value of a Pre-Deployment Assessment

A facility-specific baseline makes cooling optimization proposals measurable in electrical terms. Synapse describes its Complimentary DRL Efficiency Analysis as a passive assessment of plant topology and telemetry samples that projects potential cooling energy reductions without interrupting production operations. The result provides a quantitative starting point for an optimization business case.

How Efficiency Analysis Works: A Zero-Disruption Approach

Synapse describes its complimentary analysis process as a zero-disruption assessment using passive telemetry, plant-topology information and a mutual NDA. It is designed to estimate:

  • Avoided cooling energy relative to an identified baseline controller
  • Annual electricity cost savings using site-specific operating hours and electricity rates
  • Optimization potential for the actual installed cooling configuration

The analysis distinguishes projected performance from realized outcomes. Realized savings are determined by comparing appropriately metered operating periods on an equivalent basis.

Distinguishing Projections from Published Efficiency Benchmarks

31. Google's Omaha, Nebraska campus reported 1.05 trailing-12-month PUE in Q2 2026

The Omaha campus figure is a company-reported, location-specific U.S. operator benchmark. It does not demonstrate that any particular pre-deployment assessment or DRL platform caused the outcome. A facility's savings case should instead compare its own measured baseline and post-implementation energy use.

32. NLR's U.S. ESIF data center reports annualized PUE of 1.036

The National Laboratory of the Rockies reports that its Energy Systems Integration Facility data center achieved 1.036 annualized PUE. This unusually efficient individual installation demonstrates the potential of carefully engineered thermal infrastructure, not an attainable or guaranteed PUE for every existing data center.

33. EU data center electricity consumption could reach 115 TWh in 2030

The European Commission's November 2025 overview, citing IEA estimates, describes growth from approximately 70 TWh in 2024 toward 115 TWh by 2030. These are EU-wide demand estimates, not savings that have been demonstrated through cooling optimization.

34. LBNL projects 649 TWh of U.S. data center electricity demand in 2030 in its reference case

LBNL's June 2026 report projects 464 TWh in 2028 and 649 TWh in 2030 in its U.S. reference case. The 2030 level is modeled, not measured, and the report supplies additional scenarios to reflect uncertainty. Reducing non-IT overhead can help offset part of that demand, although PUE improvements alone cannot determine total electricity growth.

35. Google's St. Ghislain, Belgium campus reported 1.08 trailing-12-month PUE in Q2 2026

Google's St. Ghislain campus result of 1.08 supplies a further company-reported European facility benchmark. A single site does not establish typical regional performance. For Synapse-optimized facilities, meaningful comparisons with this campus account for cooling architecture, workload, climate and PUE measurement boundaries.

Synapse Thermal advertises an 8-14-month payback period through energy savings and an approximately 4-6-week deployment window with existing equipment. Its homepage identifies 1.03 as a potential optimal PUE with its optimization, not a PUE already achieved at a customer facility or a guaranteed outcome. These company-stated expectations, combined with one on-premise, cooling-focused DRL agent and a complimentary pre-deployment analysis, create a quantitative, site-specific framework for assessing a retrofit-first cooling project.

Frequently Asked Questions

What is a good PUE for a modern data center?

A PUE in the 1.2-1.3 range can be a demanding operational target for many facilities, but suitability depends on cooling architecture, climate and workload rather than a universal percentile threshold. For context, LBNL models U.S. aggregate PUE at 1.45 in 2024, while Google's Lancaster, Ohio campus reported 1.04 trailing-12-month PUE in Q2 2026. For applicable German facilities, the relevant targets are the statutory limits differentiated by commissioning date, not a presumed EU-wide limit of 1.1.

How does PUE impact the operating costs of a data center?

With 20 MW of constant IT load, 8,760 annual operating hours and $0.10/kWh electricity, improving from a PUE of 1.45 to 1.20 avoids 43.8 GWh and saves approximately $4.38 million per year in facility overhead electricity. This calculation applies the total-facility-energy-to-IT-energy PUE relationship. It does not establish that all savings come from cooling, nor that equipment life will increase simply because PUE falls. Actual costs depend on metered loads, tariffs, operating conditions and control changes.

Can AI solutions be integrated with existing data center cooling infrastructure?

Yes. Synapse Thermal describes hardware-agnostic integration for its cooling-focused DRL agent across existing HVAC, CRAC, chilled-water, direct-to-chip liquid and immersion-cooling infrastructure without wholesale replacement. On-premise inference drives configured closed-loop actions while respecting thermal operating limits. Synapse publishes a 4-6-week deployment timeframe, with site integration mapping the available sensors, controllers and safety interlocks.

What is the typical ROI for investing in advanced cooling optimization systems?

Synapse advertises an 8-14-month payback window through energy savings. The company's 35-40% efficiency advantage over traditional PID or rule-based controls is a company-stated comparison of DRL controller performance. These published metrics represent different quantities: the controller comparison is not the same calculation as whole-facility PUE or total cooling-bill savings. A project's return on investment depends on controllable cooling electricity, implementation investment, operating hours and local tariffs. Synapse's complimentary analysis projects a site-specific energy and cost case ahead of deployment.

How do high-heat GPU workloads affect data center PUE?

High-density GPU deployments concentrate thermal loads and can require more sophisticated liquid-loop, airflow and heat-rejection control. PUE depends on the actual overhead electricity relative to the IT load, so more heat does not automatically imply a specific PUE change. LBNL forecasts that AI servers could represent 55% of U.S. data center electricity in 2030. Synapse's stated 15-30-minute predictive approach provides a cooling-focused, on-premise DRL method for proactively adjusting configured cooling resources as workloads vary. Its 1.03 potential optimal PUE is a stated possibility, not a measured or guaranteed site outcome.

What are green data centers and how does PUE contribute to them?

Green data centers seek to reduce environmental impacts through efficient IT and cooling, responsible electricity procurement, water management, heat reuse and waste reduction. PUE is the ratio of total facility energy to IT equipment energy, rather than the proportion going to IT; it does not by itself measure carbon intensity or water use. Germany's Energy Efficiency Act sets separate PUE thresholds for qualifying pre-July-2026 and newer data centers, while EU rules require reporting for data centers with at least 500 kW of installed IT power demand and the Commission was consulting on EU-wide minimum performance standards in September 2026. Intelligent cooling optimization can contribute to these goals when savings are measured against a credible site-specific baseline.

Synapse Thermal
Privacy Policy

© 2026 Synapse Thermal. All rights reserved.