As artificial intelligence drives unprecedented demand for data center capacity, one challenge is becoming increasingly clear: power is difficult to obtain.

In many regions, utilities are struggling to keep pace with new data center development. Electrical interconnections that once took months may now require years. New substations can cost millions of dollars and significantly delay projects.

When power becomes the limiting factor, every kilowatt matters.

This is where Power Usage Effectiveness, or PUE, becomes important.

PUE is one of the most widely used metrics in the data center industry because it helps owners understand how much of their electrical power is actually reaching the servers that generate revenue.

What Is PUE?

Power Usage Effectiveness is defined as:

PUE = Total Facility Power ÷ IT Equipment Power

Where:

• Total Facility Power includes servers, cooling systems, lighting, UPS systems, pumps, fans, and all supporting infrastructure.
• IT Equipment Power includes only the equipment performing computing functions.

A perfect data center would have a PUE of 1.0, meaning every watt entering the facility is used by the IT equipment.

In reality, every data center requires supporting infrastructure, so actual PUE values are higher.

Why PUE Matters

Consider a data center with a utility service capacity of 10 MW.

PUE = 2.0

Total Facility Power: 10 MW

IT Equipment Power:

10 MW ÷ 2.0 = 5 MW

Only half of the available power reaches the servers.

PUE = 1.5

Total Facility Power: 10 MW

IT Equipment Power:

10 MW ÷ 1.5 = 6.7 MW

PUE = 1.2

Total Facility Power: 10 MW

IT Equipment Power:

10 MW ÷ 1.2 = 8.3 MW

The difference is significant.

Improving PUE from 2.0 to 1.2 effectively increases available computing capacity by approximately 66% without requiring additional utility power.

When utility capacity is difficult to obtain, improving PUE may be one of the fastest ways to increase usable computing capacity.

What Is a Good PUE?

PUE values vary significantly depending on facility age, climate, cooling strategy, IT density, redundancy requirements, and operating conditions.

Typical ranges include:

While a lower PUE is generally desirable, the lowest PUE is not always the best business decision.

Owners must balance:

• Capital cost
• Water consumption
• Reliability
• Maintenance
• Utility costs
• Future flexibility

The optimal solution often depends on project goals rather than achieving the lowest possible PUE.

The Largest Drivers of PUE

While many factors influence data center efficiency, several design decisions typically have the greatest impact.

1. Cooling System Design

Cooling is often the largest non-IT electrical load in a data center.

The cooling strategy may include:

• Computer Room Air Handlers (CRAHs)
• Computer Room Air Conditioners (CRACs)
• Chilled water systems
• Cooling towers
• Dry coolers
• Direct-to-chip liquid cooling
• Immersion cooling

As rack densities increase, cooling systems become increasingly important drivers of overall facility efficiency.

A small improvement in cooling efficiency can produce substantial reductions in facility power consumption.

Cooling strategy also affects water consumption. A low-PUE design may use significant water if it relies heavily on evaporative cooling, while a slightly higher-PUE design may dramatically reduce water use through dry coolers or hybrid cooling strategies.

For a deeper discussion of water-efficient cooling strategies, see Reducing Data Center Water Consumption with Dry Coolers: A Practical Approach for the AI Era.

2. Airflow Management

Poor airflow management forces cooling equipment to work harder than necessary.

Common optimization strategies include:

• Hot aisle containment
• Cold aisle containment
• Blank-off panels
• Cable management
• Air leakage reduction

These relatively simple improvements can often produce meaningful reductions in cooling energy.

3. Higher Cooling Water Temperatures

One of the most effective ways to improve efficiency is operating cooling systems at higher temperatures.

Benefits may include:

• Increased economizer hours
• Reduced compressor operation
• Improved chiller efficiency
• Improved heat rejection performance

Many modern facilities are designed to operate at temperatures that would have been considered aggressive only a decade ago.

4. Liquid Cooling

The rapid growth of AI workloads is driving interest in liquid cooling technologies.

Traditional air cooling becomes increasingly difficult as rack densities increase.

Today it is common to see:

• 30 kW racks
• 50 kW racks
• 100 kW racks

Some AI deployments are already exceeding these values.

Liquid cooling technologies can reduce fan energy while improving heat removal effectiveness.

Common approaches include:

• Direct-to-chip cooling
• Rear-door heat exchangers
• Immersion cooling

These technologies are expected to play an increasingly important role in future data center designs.

5. Heat Rejection Systems

Ultimately, every watt consumed by a server becomes heat that must be removed.

Heat rejection systems may include:

• Cooling towers
• Dry coolers
• Hybrid systems
• Adiabatic systems

Each approach involves tradeoffs between energy consumption, water consumption, capital cost, and reliability.

6. Electrical Distribution Efficiency

Not all losses occur within the cooling system.

Electrical infrastructure also consumes power through:

• UPS losses
• Transformer losses
• Power conversion losses
• Distribution losses

While individual losses may seem small, they can become significant in large facilities.

Is PUE the Only Metric That Matters?

No.

PUE remains a useful metric, but it should not be evaluated in isolation.

For example, a facility may achieve an excellent PUE by using large amounts of water. Another facility may operate at a slightly higher PUE while dramatically reducing water consumption through dry coolers, thermal storage, or other cooling strategies.

For a deeper discussion of water-efficient cooling strategies, see my article: Reducing Data Center Water Consumption with Dry Coolers: A Practical Approach for the AI Era.

Similarly, some facilities may prioritize reliability, redundancy, or operational flexibility over achieving the absolute lowest PUE.

Owners should evaluate the complete picture, including:

• PUE
• Water Usage Effectiveness (WUE)
• Reliability
• Utility costs
• Sustainability goals
• Future expansion requirements

The Future of PUE in AI Data Centers

As AI continues driving higher rack densities and increasing power demand, efficient infrastructure becomes increasingly valuable.

The challenge is no longer simply reducing operating costs.

In many markets, improving efficiency directly increases the amount of computing capacity that can be deployed within a fixed utility power allocation.

For owners and developers, that may represent one of the most valuable opportunities available.

Independent Review of Data Center Energy Infrastructure

Whether evaluating a new facility or upgrading an existing one, understanding the factors that drive PUE can help identify opportunities to improve efficiency, increase computing capacity, and reduce long-term operating costs.

I provide independent technical reviews of data center energy infrastructure, including cooling systems, electrical systems, utility constraints, thermal storage concepts, and emerging cooling technologies.

If you're evaluating a data center project and would like an independent perspective, visit Fassbender Energy Advisory.

```