What is a good PUE for a data center?

The industry average PUE is approximately 1.55 to 1.60 as of 2026. A PUE below 1.40 is considered efficient, and hyperscale operators like Google and Microsoft report annual PUE values between 1.10 and 1.20. A PUE of 1.0 would mean all electricity goes directly to IT equipment with zero overhead, which is physically impossible. For Bitcoin mining facilities, PUE values of 1.03 to 1.10 are achievable because mining hardware tolerates higher ambient temperatures and can operate with minimal cooling infrastructure in favorable climates.

How do you calculate PUE?

PUE is calculated by dividing total facility energy consumption by the energy consumed by IT equipment. The formula is PUE = Total Facility Energy / IT Equipment Energy. Total facility energy includes everything the data center uses: IT loads, cooling systems, lighting, security systems, UPS losses, and power distribution losses. IT equipment energy includes only the power drawn by servers, storage, and networking hardware. For example, if a facility uses 10 MW total and 7 MW goes to IT equipment, the PUE is 10/7 = 1.43.

What is WUE and why does it matter?

Water Usage Effectiveness (WUE) measures the volume of water a data center uses per unit of IT energy consumed, expressed in liters per kilowatt-hour (L/kWh). WUE matters because evaporative cooling systems, which many data centers use to achieve low PUE values, consume significant volumes of water. A typical WUE for a facility using evaporative cooling is 1.0 to 2.0 L/kWh. Air-cooled facilities achieve WUE of 0 to 0.1 L/kWh but at the cost of higher PUE. In water-stressed regions like the Middle East and American Southwest, WUE is becoming as important as PUE in facility design and regulatory compliance.

Does liquid cooling improve PUE?

Yes, direct liquid cooling and immersion cooling can significantly improve PUE. Traditional air cooling requires computer room air handlers (CRAHs) and raised-floor air distribution that consume 30 to 40 percent of total facility energy. Direct-to-chip liquid cooling eliminates most air handling equipment, and single-phase or two-phase immersion cooling can reduce cooling energy to 2 to 5 percent of total facility energy, yielding PUE values of 1.02 to 1.06. The higher heat capacity of liquid coolants allows heat rejection at higher temperatures, enabling dry coolers instead of evaporative towers, which also reduces WUE.

How do Bitcoin mining facilities achieve such low PUE values?

Bitcoin mining facilities achieve PUE values of 1.03 to 1.10 through several design advantages. First, ASIC miners tolerate inlet air temperatures up to 35 to 40 degrees Celsius, allowing free air cooling in most climates for most of the year. Second, mining facilities use simple air-side economizers or direct outdoor air rather than precision air conditioning. Third, mining loads have minimal ancillary IT infrastructure (no storage arrays, minimal networking, no redundant servers), so nearly all power goes to productive hash computation. Fourth, many mining facilities use containerized or open-air designs that eliminate the building envelope losses associated with traditional data centers.

Data Center PUE and WUE: Energy and Water Efficiency Metrics for Sustainable Operations

Data Center Operations September 24, 2026 14 min read Rax Data & Energy

Data center cooling infrastructure with water treatment systems and efficiency monitoring equipment

Why Efficiency Metrics Matter for Data Centers

A data center that draws 10 MW from the grid does not deliver 10 MW of compute. A portion of that power goes to cooling systems, lighting, power distribution losses, UPS conversion overhead, and building systems. The ratio between what the facility draws and what the IT equipment actually consumes determines whether the operation is a lean compute engine or a building that happens to contain servers.

For colocation and hosted mining operators, efficiency metrics directly translate to cost competitiveness. A facility with a PUE of 1.60 spends 60 cents on overhead for every dollar of IT power delivered. A facility at 1.10 spends 10 cents. At scale, that difference is measured in millions of dollars per year and determines which operators can offer competitive hosting rates while maintaining margins.

For AI compute workloads, efficiency has taken on additional urgency. Training and inference clusters running NVIDIA H100 or H200 GPUs can draw 60 to 100 kW per rack, pushing cooling systems beyond what traditional air handling was designed for. The facilities that can handle these densities without proportionally increasing overhead energy are the ones winning hyperscaler and enterprise deployments.

Power Usage Effectiveness: The Industry Standard

PUE was introduced by The Green Grid in 2007 and has become the most widely cited data center efficiency metric. The formula is straightforward:

PUE = Total Facility Energy / IT Equipment Energy

A perfect PUE of 1.0 would mean every watt entering the facility reaches IT equipment, with zero overhead. This is physically impossible because power distribution equipment, cooling systems, and building infrastructure always consume some energy. In practice, PUE ranges from approximately 1.03 for purpose-built mining containers to above 2.0 for poorly designed or aging facilities.

What Counts as Total Facility Energy

Total facility energy includes everything the data center consumes at the utility meter or behind-the-meter generation point: IT equipment power, cooling systems (chillers, cooling towers, CRAHs, fans, pumps), UPS systems (conversion losses, battery charging), power distribution (transformer losses, PDU losses, switchgear), lighting and security systems, and building mechanical systems (fire suppression, humidification, elevator equipment).

What Counts as IT Equipment Energy

IT equipment energy is the power consumed by compute, storage, and network hardware at the point of delivery: servers and compute nodes (CPU, GPU, memory, fans, internal PSU losses), storage systems (disk arrays, flash storage, SAN switches), network equipment (switches, routers, load balancers), and any other IT-classified load. For Bitcoin mining facilities, IT equipment energy is essentially the total ASIC miner power draw including internal fan power.

Industry PUE Benchmarks

Facility Type	Typical PUE Range	Best-in-Class PUE	Primary Cooling Method
Hyperscale (Google, Microsoft, Meta)	1.08 - 1.20	1.06	Evaporative / free cooling
Enterprise colocation (Equinix, Digital Realty)	1.30 - 1.50	1.20	Chilled water / air-side economizer
Legacy enterprise on-premises	1.60 - 2.50	1.40	Traditional CRAH / chiller
Bitcoin mining (air-cooled containers)	1.03 - 1.10	1.02	Direct air / evaporative pre-cool
Bitcoin mining (immersion-cooled)	1.02 - 1.05	1.01	Single-phase immersion / dry cooler
AI/GPU compute (liquid-cooled)	1.05 - 1.15	1.03	Direct-to-chip liquid / rear-door heat exchanger

Water Usage Effectiveness: The Hidden Resource

While PUE captures energy efficiency, it says nothing about water consumption. WUE fills this gap by measuring the volume of water consumed per unit of IT energy:

WUE = Annual Water Usage (liters) / IT Equipment Energy (kWh)

The unit is liters per kilowatt-hour (L/kWh). A WUE of 0 means the facility uses no water for cooling. A WUE of 1.8 means the facility consumes 1.8 liters of water for every kilowatt-hour of IT energy delivered.

WUE has become increasingly important as data center construction accelerates in water-stressed regions. Facilities in the American Southwest, the Middle East, and parts of India face scrutiny over water consumption from both regulators and communities. Several jurisdictions now require water impact assessments as part of the data center permitting process.

The PUE-WUE Trade-Off

Evaporative cooling is one of the most energy-efficient methods available for rejecting heat from a data center. Cooling towers and direct evaporative systems can achieve approach temperatures within 2 to 3 degrees Celsius of the wet-bulb temperature, reducing or eliminating the need for mechanical refrigeration. This drives PUE down, sometimes dramatically.

The cost is water. A 10 MW IT load data center using evaporative cooling in a hot climate might consume 15 to 25 million liters of water per year. The same facility using air-cooled chillers or dry coolers would consume near-zero water but require significantly more energy for heat rejection, pushing PUE up by 0.10 to 0.30.

Cooling Strategy	Typical PUE Impact	Typical WUE (L/kWh)	Best For
Evaporative cooling towers	1.10 - 1.25	1.2 - 2.0	Water-abundant, hot climates
Air-side economizer	1.10 - 1.30	0 - 0.3	Cool/mild climates
Mechanical chiller (air-cooled condenser)	1.30 - 1.60	0 - 0.1	Water-restricted sites
Mechanical chiller (water-cooled condenser)	1.20 - 1.40	0.8 - 1.8	High-density, moderate water
Direct liquid cooling + dry cooler	1.03 - 1.08	0 - 0.05	High-density GPU/AI, water-restricted
Immersion cooling + dry cooler	1.02 - 1.05	0 - 0.02	Maximum density, zero-water target

Measuring PUE and WUE Accurately

The value of PUE and WUE as management tools depends entirely on measurement accuracy and consistency. The Green Grid defines three levels of PUE measurement that reflect increasing precision.

Category 1: Basic

Total facility energy is measured at the utility meter. IT energy is estimated based on nameplate ratings or UPS output readings. This is the most common measurement approach and is sufficient for trending and high-level comparison but can have measurement uncertainty of 10 to 15 percent.

Category 2: Intermediate

Total facility energy is measured at the utility meter with dedicated submeters for major non-IT loads (cooling plant, lighting, building mechanical). IT energy is measured at the UPS output or PDU input with dedicated metering. Measurement uncertainty drops to 5 to 10 percent.

Category 3: Advanced

Every significant load within the facility is individually metered. IT energy is measured at the server power supply input using intelligent PDUs with per-outlet metering. Measurements are recorded at intervals of 15 minutes or less and annualized. This level of granularity enables root-cause analysis of efficiency deviations and supports real-time optimization. Measurement uncertainty is below 5 percent.

Temporal Considerations

PUE varies with weather, IT load, and operational conditions. A facility might achieve PUE 1.10 on a cold winter night when free cooling handles the entire load and PUE 1.40 on a hot summer afternoon when mechanical chillers run at full capacity. Reporting annual average PUE provides a more honest representation than cherry-picking favorable conditions. The Uptime Institute's annual survey uses annualized PUE for this reason.

Improving PUE: Where the Overhead Goes

Reducing PUE requires understanding where overhead energy is consumed and which interventions offer the best return on investment.

Cooling System Optimization

Cooling typically represents 30 to 45 percent of non-IT energy consumption and is the largest single opportunity for PUE improvement. Strategies include raising ASHRAE-recommended supply air temperatures from the traditional 18 to 20 degrees Celsius to the allowable 27 degrees Celsius or higher (each degree reduces cooling energy by 2 to 4 percent), implementing hot/cold aisle containment to prevent mixing (reduces cooling energy 15 to 25 percent), maximizing economizer hours by using air-side or water-side free cooling whenever ambient conditions allow, and transitioning from constant-speed to variable-speed drives on fans, pumps, and compressors.

Power Distribution Efficiency

Power distribution losses (transformer, UPS, PDU) account for 10 to 20 percent of overhead energy. Improvements include deploying high-efficiency transformers (99+ percent efficient at rated load), using UPS systems with eco-mode operation (99+ percent vs 94 to 97 percent in double-conversion mode), and considering DC power distribution to eliminate multiple AC-DC-AC conversion stages.

Lighting and Mechanical

Building lighting and mechanical systems typically represent 3 to 8 percent of overhead energy. LED lighting with occupancy sensors, variable-speed drives on all motors, and envelope improvements (insulation, air sealing) offer incremental PUE gains that add up in aggregate.

PUE for Bitcoin Mining Facilities

Bitcoin mining operations achieve the lowest PUE values in the data center industry for structural reasons. ASIC miners have a single computational function (SHA-256 hashing) with no complex memory hierarchies, storage subsystems, or networking stacks. The power consumed by an ASIC miner is almost entirely converted to hash computation and heat.

Mining facility design exploits the thermal tolerance of ASIC hardware. Manufacturers like Bitmain and MicroBT rate their miners for inlet temperatures up to 40 degrees Celsius, far above the 27 degrees Celsius maximum recommended for enterprise servers by ASHRAE. This allows mining facilities in most climates to use direct outdoor air for cooling during 70 to 95 percent of annual hours, requiring evaporative assist or mechanical cooling only during extreme heat events.

Containerized mining deployments further reduce overhead by eliminating the building envelope entirely. A shipping container with intake louvers and exhaust fans adds minimal energy overhead to the IT load, achieving PUE values of 1.03 to 1.05 in temperate climates. Immersion-cooled mining setups push this even lower by eliminating miner-internal fans and rejecting heat through dry coolers at elevated coolant temperatures.

PUE and WUE for AI Compute Facilities

The rise of GPU-intensive AI workloads has complicated the PUE landscape. A single NVIDIA DGX H100 system draws 10.2 kW. A rack of four DGX units approaches 45 kW. At these densities, traditional air cooling cannot remove heat fast enough, regardless of how much air is pushed through the racks.

Direct liquid cooling (DLC) has emerged as the standard for high-density AI deployments. In DLC configurations, coolant flows through cold plates mounted directly on GPU and CPU packages, capturing 60 to 80 percent of the heat at the source. The remaining heat is handled by supplemental air cooling for memory, storage, and auxiliary components. This hybrid approach achieves PUE values of 1.05 to 1.10 while enabling rack densities of 60 to 100+ kW.

The WUE advantage of liquid cooling is significant. Because DLC raises the coolant return temperature to 45 to 55 degrees Celsius, heat can be rejected through dry coolers rather than evaporative towers in most climates. This delivers WUE values near zero while maintaining competitive PUE, resolving the traditional PUE-WUE trade-off that has constrained air-cooled designs.

PUE and WUE in the UAE and Middle East

Operating data centers in the UAE and broader Gulf region presents unique efficiency challenges. Ambient temperatures regularly exceed 45 degrees Celsius in summer, and high humidity in coastal locations limits the effectiveness of air-side economizers. At the same time, water is a scarce and valuable resource throughout the region, making high-WUE cooling strategies problematic.

Facilities in the UAE that rely on evaporative cooling face both high water costs and community pressure to minimize consumption. Treated desalinated water in the UAE costs approximately AED 10 to 15 per cubic meter, adding $0.003 to $0.005/kWh to cooling costs for evaporative systems. Several UAE municipalities have implemented or proposed water consumption limits for large commercial facilities.

This environment makes liquid and immersion cooling particularly attractive for Middle East deployments. Immersion cooling with dry coolers achieves PUE values of 1.03 to 1.06 and WUE near zero even in 50-degree-Celsius ambient conditions, because the elevated coolant temperature (55 to 65 degrees Celsius) provides sufficient temperature differential for air-cooled heat rejection without water.

Beyond PUE: Emerging Efficiency Metrics

While PUE and WUE remain the most widely used metrics, the industry is developing additional measures that capture dimensions of efficiency and sustainability that PUE alone misses.

Carbon Usage Effectiveness (CUE)

CUE measures CO2 emissions per kWh of IT energy. CUE equals the total CO2 emissions caused by the data center divided by IT equipment energy. A facility powered by renewables might have excellent PUE but zero CUE if the grid electricity is clean, or poor CUE if the renewable energy displaces other users rather than adding to the grid. CUE provides a more complete picture of environmental impact than PUE alone.

Energy Reuse Effectiveness (ERE)

ERE accounts for waste heat that is captured and reused outside the data center. ERE equals (Total Facility Energy minus Reused Energy) divided by IT Equipment Energy. A facility that captures waste heat for district heating, industrial processes, or greenhouse agriculture can achieve ERE values below 1.0, meaning the facility returns more useful energy to the surrounding community than its IT equipment consumes in overhead. This metric is gaining traction in Scandinavian countries where district heating integration is common.

Total Cost of Ownership per Useful Compute

PUE measures energy efficiency but not compute efficiency. A facility with excellent PUE running underutilized servers wastes more energy per useful computation than a higher-PUE facility with well-managed workloads. Metrics that combine infrastructure efficiency with compute utilization (such as energy per transaction, energy per inference, or joules per hash) provide a more operationally relevant efficiency picture.

Rax Approach to Efficiency

Rax designs and operates facilities with efficiency as a primary engineering constraint, not an afterthought. Our Bitcoin mining hosting facilities achieve PUE values of 1.03 to 1.08 through containerized deployments with direct air cooling and evaporative pre-cooling where climate demands. Our AI compute infrastructure uses direct liquid cooling to support rack densities above 80 kW while maintaining PUE below 1.10 and WUE near zero.

For clients evaluating hosting providers, we recommend requesting annualized PUE (not snapshot), asking about WUE (especially in water-stressed locations), understanding the measurement methodology (Category 1 vs 2 vs 3), and verifying that quoted PUE values include all facility loads including office space, security, and utility compound equipment.

Facility efficiency is a key component of competitive hosting economics. Lower PUE means lower overhead costs passed through to clients, which directly translates to better mining profitability and lower AI compute costs per GPU-hour.

Back to Knowledge Center