What Is Power Compute Effectiveness (PCE)? How AI Data Centers Maximize Deployable Compute

Emerging metrics including Power Compute Effectiveness (PCE) will be critical to guiding the next wave of AI data center design — helping operators allocate provisioned power more effectively while meeting the sustainability concerns of local communities.

Power Compute Effectiveness (PCE) is a new data center metric measuring the ratio of provisioned power allocated to compute versus total provisioned facility power. Developed by Airsys with industry input, PCE helps operators design and retrofit AI data centers to deploy more compute from existing power capacity — addressing blind spots in the legacy PUE metric that don’t account for how power is actually allocated on-site.

It’s been acknowledged that some of the first wave of AI data centers were suboptimal from a design and operations perspective. A lack of experience, standards, and intense time pressures were all contributory factors. These inefficiencies — including energy and water usage — may have also contributed to the wave of anti-data center feeling in the U.S. and other regions.

Airsys Chief Strategic Relations Officer Paul Quigley was part of a recent webcast with leading industry analysts hosted by DCD. The panel looked at why the industry must focus on maximizing compute from provisioned power in order to meet not only its own growth targets but also the growing environmental and social expectations of local communities.

Joining the session along with Paul were Vlad Galabov, independent data center analyst, and Moises Levy, chief executive officer, DCMETRIX. The session was hosted by DCD Cloud & Hybrid editor Georgia Butler.

Click here to watch the webcast.

The webcast took a deep dive into how to ensure current and future facilities are more environmentally and financially sustainable, which in turn should help with community acceptance. The panel highlighted that new metrics are a critical part of this process for optimizing both the design and operation of AI facilities. These metrics include Power Compute Effectiveness (PCE) and Return on Invested Power (ROIP), both developed by Airsys with industry input and oversight. Other related topics covered included the opportunity from retrofitting legacy facilities, and future innovation including the impact of agentic AI on data center design and operation.

What Constraints Are AI Data Center Operators Facing Today?

Vlad Galabov, analyst: They’re really struggling to find power. They’re really struggling to find water. And I think we’ve seen that in the U.S. There’s protests actually against data centers. And this is not isolated to the U.S. It’s happening all around the world. There was a protest in India just a couple of days ago, the first ever against the data center, so we have to figure out how we can actually make it work with the resources we have.

Moises Levy, analyst: Look at the scale of data center expansion. Nowadays, it’s impressive, based on the latest report from Lawrence Berkeley Lab to Congress June 2026, by the end of the decade, U.S. data centers could consume up to, give or take, 12 percent of the national electricity. But let’s make it clear: AI is here. It’s not going anywhere, and we need to figure out how we use resources. Nowadays we are like toddlers. We are learning how to walk. So the more effective and the more efficient we design our data centers, the better it will be.

Paul Quiqley, Airsys: AI was a surprise to everybody in terms of demand. I talked to AWS, Meta, and really the top seven, and everybody was surprised. You think how could that have happened? We’ve been building data centers for 25, 30, years. We knew something with AI was coming, but everybody was surprised. And unfortunately, because of that, some of the designs were sloppy. The designs were, what do we have? How can we get these going fast? We have to be stewards of this. We have to be smarter. I’ll be the first to admit I was greenwashed in the past. I just got tired of all of the talk, and it wasn’t really sinking in. Now it’s a reality. We must pay attention to it. The metrics that we used to have, are they still accurate? Are they still valid? Do they give us what we want?

Why Isn’t PUE Enough Anymore?

Vlad, analyst: So Power Usage Effectiveness (PUE) is very well intended, but the problem is that it is outdated. PUE does not really address how much of the computing capacity is being consumed in what way. So it is a little bit oversimplified as a measure, it does not really address any of the water usage either. So, at the moment, it’s not that we should not use PUE, but I think that there is a sense of urgency to look at other metrics that can better show us if we’re efficient.

Moises, analyst: We need a more multidimensional basket of metrics. Because at the end of the day, what are we trying to do with metrics? We’re trying to solve a multi-objective optimization, because we need real, true end-to-end optimization. And how do we start doing that? At the start, when we’re planning and designing a data center. Depending on how we design a data center, it will tell us the story of how we’ll be building it and how we’ll be operating it.

Paul, Airsys: So we’re supporting PCE full force. Uptime has published on it already. OCP should be publishing here in Q3. There’s nothing wrong with PUE. There’s nothing wrong with WUE, and they were really good, strong metrics when they first came out. But that surprise of AI changed the dimensions a little bit. So now PUE sort of has some blind spots on it. One would be of timing, and maybe the other one would be of what I’ll call territory. So the timing blind spot is the real PUE really doesn’t happen until that meter’s spinning, until that thing’s up and operating. There really is no true judgment of the PUE in there. So that’s a part of it. You’ve heard a lot about people using the word stranded power all the time. People use that differently, but that’s again where PUE may not be quite exactly what we need. We’re almost looking at PUE as though we’re driving through the rearview mirror.

What Does Power Compute Effectiveness (PCE) Actually Measure?

Related Reading: From PUE to PCE: How High-Density AI Infrastructure Demands New Metrics

Power Compute Effectiveness is the ratio of provisioned power allocated to compute against total provisioned facility power — a measure of how power is allocated at the design stage, not how efficiently it’s used once delivered.

Paul, Airsys: What PCE is intended to do, and hopefully does for everybody, is it makes us look through the windshield as we’re driving. It gives us that design capability. So right up front, we can say, “Hey, we are being better stewards here in the way the energy is flowing into this data center through the data center, and then how much is actually still available for compute.”

What Is ROIP (Return on Invested Power)?

ROIP is distinct from PCE: PCE measures the physical allocation of provisioned power to compute, while ROIP measures the financial return on the capital invested in a facility’s power infrastructure — the payback on that investment, as reflected on the balance sheet.

Paul, Airsys: The other part of this to address is what’s paying for all this, right? What we call ROIP. What is the Return on Invested Power (ROIP)? People are looking at the balance sheet, and when they look at the power and they say we need five more gigawatts in the United States in this area or whatever, they’re looking at what’s going to pay back all of the money. Now we can say, “All right, here’s the metric that can give you your ROIP on your balance sheet.” The first time you’ve ever been able to see that.

How Effectively Is the Data Center Industry Using Provisioned Power Today?

Georgia, DCD: I’m curious to know what your thoughts are on if the data center industry as a whole today were to figure out the PCE of their facility. How well do you think they’re doing? Do you think they are generally running quite efficiently?

Paul, Airsys: It’s an age-related question and a design-related question. I think you’re seeing cleaner designs right now, but if you’re looking at a facility that’s three to four years old, something like that, it’s definitely almost mandatory. You know, PCE isn’t a solution for everything. There’s an awful lot out there in what we call the brownfield, the legacy data centers that we can now go clean up, stop using the water, really use the higher temp on the servers themselves, convert them to a realistic liquid cooling.

Vlad, analyst: There is a huge opportunity in being smart about how you can improve or retrofit an existing infrastructure because there really are a lot of existing data centers. One of the things that I think we need to be talking about is okay, because PCE is a very interesting metric. How does a data center improve their PCE, even if they have existing infrastructure? And there are many things that you can do. Sometimes we think about adopting liquid cooling in a legacy data center, particularly an extra old data center. Something that’s very very difficult. It’s not impossible, and there are a lot of different ways to achieve that.

What’s Next: Agentic AI and Energy Reuse

Paul, Airsys: The big shift that’s coming that I really love is agentic AI. CPUs are making a comeback along with GPUs for agentic AI. NVIDIA has a nice new Vera CPU out too that runs at a higher temperature than the Vera GPU does. When you push hot water to cool a chip, you don’t need a chiller. Now you have hot water out there with air coolers out there. That just means they return a whole bunch of energy to the grid, or a whole bunch of energy that can become new compute instead of new green fields. Right? That’s all very important. And then hopefully soon we’ll be talking about the next metric, which is the energy reuse metric. So now we have all of this heat, this energy that we pulled out of the data center.

Remember, energy can’t be created or destroyed. So what the heck are we going to do with that? We have all kinds of great ideas, and when we get into containerized data centers, we can start putting these into locations where that actual energy can be reused into hospitals, campuses, and just a ton of applications. That’s the next big thing to look for in the next three years. It’s incumbent upon all of us to look to that design and say, “Yeah, we’re going to be stewards. We need to. We have an important role to play here in the future.”