Is PUE Dead? Are There Better Ways to Measure Data Center Efficiency?

Data center efficiency is top of mind for IT leaders and industry observers alike, thanks to increased AI adoption and data center density. Higher density simply means that today’s data centers consume more energy in the same physical footprint. Denser workloads are also driving a shift toward liquid cooling, which raises the question of how this shift will impact global data center efficiency.

To continue progressing toward their sustainability goals, business leaders need their data center workloads to run as efficiently as possible. Like everything else in the business world, we can’t truly understand data center efficiency unless we can accurately measure it. Data center operators need meaningful insights into how efficient their facilities are and how they can make improvements. For almost 20 years now, power usage effectiveness (PUE) has been the preferred metric to provide these insights.

Since PUE has been around so long, and since so much has changed during that time, some have suggested that it’s now obsolete. At Equinix, we recognize that PUE may not be a perfect metric, but we still consider it a valuable metric. Below, I’ll summarize some of the reasons that the supposed death of PUE has been overstated.

Everyone has their own definition of “performance”

PUE compares the amount of energy consumed by IT systems with the total energy consumed throughout the data center. Notably, it doesn’t consider IT performance: the amount of useful work that systems perform while consuming that energy. According to PUE’s detractors, you can’t get a complete picture of efficiency without measuring performance—especially not in the AI era. That’s because AI data centers consume significantly more energy than conventional facilities, but they also perform significantly more work.

To measure the performance-to-consumption ratio, some have proposed alternative metrics such as total usage effectiveness (TUE). The problem with these kinds of metrics is that there’s no standardized measure of IT performance, because all organizations want something different from their IT hardware. There are ways to quantify raw compute, such as tracking the amount of LLM tokens processed in a set time period, but this is just an abstract concept. It looks good on paper, but real performance is about business outcomes.

For instance, if your company makes shoes, then each compute cycle you run presumably contributes to your ability to manufacture and sell more shoes. However, it’s highly unlikely that you can quantify that connection in terms of data center energy consumption. There’s a good reason that “shoes per kilowatt-hour” isn’t a widely used metric. Directly linking energy consumption to business outcomes would be just as impractical in any other industry.

For a global colocation provider like Equinix, a performance-efficiency metric would be especially difficult to calculate, because we serve thousands of different customers who all have their own unique goals. For this reason, we believe that PUE continues to be the best efficiency metric available for our purposes.

How you use PUE matters

One reason detractors may fail to see the value of PUE is that they’ve been using it the wrong way. For example, consider two hypothetical data centers:

• Data Center A has a PUE of 1.45.
• Data Center B has a PUE of 1.35.

On paper, Data Center B appears to be more efficient. However, there are many unknowns:

• Where are these data centers located? Are they in different climates that have different cooling requirements?
• Who uses them, and for what purposes?

Without this missing context, the comparison is ultimately meaningless.

Now, consider a different example:

• In 2022, Data Center A had a PUE of 1.45.
• In 2024, Data Center A had a PUE of 1.35.

This is using PUE in a valid and useful way. Instead of comparing one data center to another, we’re comparing a data center to itself over time. We can see that the PUE decreased significantly over a two-year period, which tells us that the data center operator must have implemented meaningful efficiency improvements. With PUE, it’s not the absolute number that matters; it’s the trend line over time.

At Equinix, efficiency and sustainability go hand in hand, and we use PUE to help us pursue continuous efficiency improvements. Each time we calculate PUE, it’s a snapshot of a moment in time. It establishes a baseline for us to work on improving. In spite of how much our business might change, with new data centers opening in new places and more high-density workloads, our goal of continuous PUE improvement stays the same.

Here’s a quick look at some of our recent global annualized average PUE results:

• In 2019, we established a baseline PUE of 1.54.
• In 2024, our PUE was 1.39.

This represents an efficiency improvement of about 28% over a five-year period. This shows that our culture of continuous improvement and the site-level targets and initiatives we’ve implemented have been effective. Without PUE, we’d have no way of knowing this.

Liquid cooling may impact PUE, but not yet

While it’s true that data center density is increasing and more workloads are using liquid cooling, this trend isn’t enough to warrant throwing out a tried-and-true metric like PUE.

I’ve seen first-hand the growing demand for liquid cooling among Equinix customers, and I’ve seen the investments we’ve made to meet that demand. Despite this, liquid cooling is still an emerging solution. It’s primarily used for workloads with very high data, compute and power requirements, and it hasn’t fully crossed over into the mainstream for generic data center and IT workloads. For this reason, I don’t expect liquid cooling to move the needle on PUE in the short term.

Liquid cooling or not, businesses should not get bogged down in the details of their deployment-level PUE numbers. Building a data center environment that meets your business needs should be your top priority. If you deploy liquid cooling, it should be because it’s a good fit for your workloads, not because you think it’s going to improve your PUE.

At Equinix, we believe that the best way for our customers to make their deployments more efficient is to collaborate with us on building a solution that’s properly sized for their needs. Deploying more cabinets than you need results in underutilization, which is the ultimate PUE killer. After you build the data center solution that’s properly sized for your business requirements, then you’ll be able to benefit from Equinix’s continuous efficiency improvements that could lead to better PUE over time.

Helping customers navigate efficiency and sustainability

At the end of the day, PUE is still a valid metric to track because it’s still important to our customers. We know that many of these customers are facing pressure from regulators and stakeholders to continue reporting PUE as part of their sustainability strategies. There’s no reason to believe this pressure will go away anytime soon, which means there’s no reason to believe that PUE will go away.

Despite this, we recognize that there will always be opportunities to improve PUE, and we’re dedicated to participating in industry working groups that pursue those opportunities. For instance, during the ISO/IEC 30134-2 review in 2024,[1] we actively participated in the committee that helped update and improve the existing best practices for PUE.

To learn more about how you can pursue your business goals alongside your sustainability and efficiency goals, read our solution brief Aligning data center strategy with sustainability goals: Five key considerations.

[1] ISO/IEC 30134 Part 2: Power usage effectiveness (PUE)