What Is Power Usage Effectiveness (PUE) In Data Centers?

In 2024, data centers consumed around 415 terawatt hours (TWh), or about 1.5% of global electricity, says the International Energy Agency.

As the backbone of the digital economy, data centers are estimated to consume 2-3% of the world’s electricity by 2030.

To manage this consumption, industry leaders have relied on a single metric: Power usage effectiveness (PUE).

What is power usage effectiveness (PUE) in data centers?

PUE is a ratio that describes how efficiently a data center uses its power, developed by The Green Grid in 2007. It specifically measures how much energy is used by the  T equipment (like servers and storage) compared to the overhead required to keep those machines running (like cooling and lighting).

The formula is straightforward:

• Total facility power: Includes everything from the cooling systems and backup batteries (UPS) to the office lights
• IT equipment power: Only the energy that directly fuels the computing, storage, and network gear

What is a good PUE score?

A “perfect” PUE score is 1.0, which means every watt of electricity goes directly to computing. However, in reality, extra power is always lost to cooling systems, lighting, and power distribution.

How is PUE used?

Data centers use PUE as a benchmarking tool to track performance over time.

• Identifying waste: By monitoring PUE in real-time using data center infrastructure management (DCIM) software, operators can pinpoint “power leaks” or underutilized cooling systems
• Gauging improvements: If a facility upgrades its cooling tech or implements hot/cold aisle containment, a drop in the PUE ratio proves that the initiative worked
• Marketing and compliance: Highly efficient providers use their low PUE scores to attract eco-conscious clients. In some regions, like the European Union, reporting PUE is now a regulatory requirement

Actual data center PUE performance

Fig. 1: PUE over the last 20 years. (Source: Uptime Institute and Google Sustainability)

Over the last 20 years, PUE has followed a “hockey stick” curve, dropping sharply early on and then hitting a long plateau.

1. The era of rapid gains (2007–2013): In the mid-2000s, the average industry PUE was around 2.5, meaning for every 1 watt of IT power, 1.5 watts were wasted on cooling and infrastructure.
2. The great plateau (2014–Present): Since roughly 2014, the global industry average has remained stubbornly flat, hovering between 1.55 and 1.59.
3. The hyperscale exception: While the industry average has stalled, “hyperscalers” (like Google, Microsoft, and Meta) have pushed boundaries. Google’s fleet-wide PUE reached 1.09 by early 2025, achieved through custom-designed hardware and using AI to manage cooling in real time.

Why PUE still matters

In the current era of AI factories, PUE is shifting from a comprehensive “gold standard” to a foundational “infrastructure-only” metric with still meaningful importance:

• Infrastructure benchmarking: PUE is the standard for measuring how much power is lost to noncomputing tasks like cooling and distribution. In AI factories, where power is the ultimate constraint, minimizing this “wasted” overhead is critical to maximizing the power available for GPUs
• Regulatory compliance: Governments are increasingly mandating PUE reporting. For example, Germany’s Energy Efficiency Act requires new data centers to achieve a PUE of 1.2 or lower starting in 2026
• Cooling strategy evaluation: As AI factories move toward high-density liquid cooling, PUE helps operators justify investments in technologies like direct-to-chip or immersion cooling, which can drive PUE as low as 1.1 by drastically reducing cooling energy

What are the benefits and limitations of PUE?

Benefits

• Simplicity: It provides a single, easy-to-understand number for complex facilities
• Industry alignment: Because it is globally recognized, it allows for a common language between engineers, executives, and sustainability officers
• Cost management: Lower PUE directly correlates to lower electricity bills
• Scalability: Efficient power use allows a facility to house more servers without needing a massive infrastructure expansion

Limitations

• Climate dependent: Facilities in cooler climates can use “free cooling” from outside air, giving them an advantage over centers in tropical regions
• Sustainability gap: PUE does not track water usage or carbon sources. Experts often pair it with water usage effectiveness (WUE) and carbon usage effectiveness (CUE) for a full picture
• Ignores IT efficiency: A data center can have a perfect PUE of 1.1 while running old, power-hungry servers. This has been referred to as the PUE Loophole because PUE only measures power up to the server’s plug, completely ignoring the massive energy losses that happen inside the hardware during AC-to-DC conversion

Fig. 2: Benefits and limitations of PUE, summarizing why the metric remains useful for infrastructure benchmarking and where it falls short as a full “grid-to-chip” efficiency measure.

The PUE loophole: A data center blind spot

PUE is widely used to track facility efficiency, but it is not perfect. It measures how efficiently a data center delivers power and omits one crucial element: power conversion in the server itself.

In this era of “grid to chip” views of data center performance, that omission creates a significant blind spot for operators.

To understand the loophole, it is important to recognize that power arrives at the server as AC, usually at a voltage of around 240 V. However, the semiconductors in the server use low-voltage DC power.

So, as a first step, the server must convert AC power to DC power, which typically takes place within an element called a power supply unit (PSU), sometimes referred to as a common redundant power supply (CRPS), a standardized, hot-swappable power module designed for servers, storage, and networking equipment. With conventional power design, this conversion process wastes significant energy.

Yet the PUE metric ignores AC/DC conversion – and this results in the PUE Loophole.

In the PUE calculation, any power consumed inside the server chassis is categorized as “IT equipment power.” This includes the high power losses resulting from moderately efficient power supplies in the server equipment.

These power supplies can be as low as 95% efficient, 5% of power wasted as heat that requires fans to keep things cool. Because the power supply doing the AC to DC conversion is counted as “IT power” (the denominator) rather than “facility power” (the numerator), the PUE looks better than it should.

The implications of the loophole are significant.

Since the PUE does not consider AC/DC conversion efficiency in the server, there is little incentive for those measured on PUE to want AC/DC converters to be more efficient. Many servers still use AC/DC converters that are 95% efficient or less.

While this may sound impressive, consider that 5% or more of all energy in a data center is lost, with measures in place to hide it. Not only does this significantly increase costs and CO2 emissions, but it also creates extra waste heat, placing additional demands on cooling systems.

What can be done

It is up to the industry itself to recognize this loophole and take action to add another measurement point or create another data center metric whereby data centers cannot have a near-perfect PUE of 1.1 while running old, power-hungry, inefficient power supplies.

Until then, winning companies will make a change to more innovative power topologies and transistor technologies, such as GaN, resulting in power supplies that are achieving efficiencies of 97% or better, which means that more than 50% of the wasted energy can instead be used effectively.

Across the entire industry, this could translate into more than 37 billion kilowatt-hours saved every year, enough to run 40 hyperscale data centers!