Power Usage Effectiveness, or PUE, is the ratio of total facility power to IT equipment power. A data center drawing 15 MW at the meter to run 10 MW of servers has a PUE of 1.5. The Green Grid introduced the metric in 2007 and it was later standardized as ISO/IEC 30134-2. Every operator quotes it, and buyers misread it more often than any other figure on a tour deck, usually because the version they are shown was measured under conditions they never asked about.
A PUE of 1.0 would mean every watt entering the building reaches a server, which no facility achieves. The best modern buildings run between 1.1 and 1.3, and older halls sit at 1.5 or higher.
What Is a Good PUE in 2026?
Here is the number that should recalibrate expectations. The global weighted average PUE is about 1.54, and it has stayed there for six straight years, according to Uptime Institute’s 2025 global survey.
Six years of flat efficiency, in an industry that talks about little else. The reason is unglamorous. Most of the world’s IT load still runs in buildings designed a decade or more ago, in climates where free cooling is unavailable for much of the year, and no amount of new construction elsewhere changes what those buildings draw. New hyperscale campuses pull the headline numbers down, but they don’t move the average much, because they remain a small share of the installed base.
So when an operator quotes 1.2, the useful follow-up is not whether that is a good number, but which building it came from, at what load, and over what period.
Why Does PUE Matter?
At scale, small differences in the ratio compound into real money on the utility bill.
Take a facility consuming 10 MW of IT power. At a PUE of 1.5 it draws 15 MW total, so 5 MW goes to cooling, lighting, UPS losses, and everything else that isn’t compute. Get that facility to 1.2 and the overhead drops to 2 MW. You’ve recovered 3 MW of continuous draw, which at commercial power rates works out to a seven-figure annual line item for the life of the lease.
PUE also feeds directly into carbon reporting. Scope 2 emissions scale with total facility draw, not with IT draw, so a poor PUE inflates the reported footprint of every workload in the building. For any tenant with a public emissions target, the landlord’s cooling design has become part of their own disclosure.
How Facilities Are Improving PUE
The gains come from a short list of well-understood moves.
Free cooling is the big one. Facilities in cool climates run economizer hours for much of the year, using outside air instead of mechanical cooling. Location does a lot of the work here, which is why the Nordics, the Pacific Northwest, and parts of Canada post the strongest numbers.
Hot and cold aisle containment stops supply air mixing with exhaust, so the cooling plant does less work for the same result. It’s cheap, it retrofits into existing space, and plenty of older halls still haven’t done it.
Liquid cooling changes the arithmetic again. Direct-to-chip and immersion systems remove heat at the source rather than conditioning an entire room, and they are becoming necessary rather than optional as rack densities climb past 30 kW. High-density AI halls with liquid cooling can post better PUE than the air-cooled space next door, even while drawing far more power per rack.
Higher operating temperatures help too. ASHRAE has widened its recommended envelope repeatedly, and running the hall warmer cuts chiller load without hurting equipment life the way operators once feared.
Where PUE Misleads Buyers
This is the part that matters when you’re comparing facilities, and it’s the part most marketing material skips.
Design PUE is not operating PUE. The number on the tour deck is often what the plant achieves at full load in ideal conditions. The number on the bill is what it achieved in August, at whatever load the hall was actually carrying.
Partial load wrecks efficiency. A half-occupied hall runs its cooling plant at poor efficiency because much of the overhead is fixed. A brand new facility with three tenants in it can post a worse PUE than a fifteen-year-old building that is full.
Annualized and instantaneous are different numbers. A facility in Texas has a very different PUE in January than in July, so a single-point measurement taken in the right month will flatter the building by a margin that never shows up in the annual figure.
The measurement boundary moves. Does the number include the substation, transformer losses, or office and support space? Two operators can each measure honestly and still hand you figures that were never comparable in the first place.
PUE says nothing about what you pay per kW. A facility with a 1.2 PUE in an expensive power market can cost more to run than a 1.4 facility somewhere cheap, because efficiency is a ratio and your bill is a dollar figure.
How to Use PUE When Comparing Facilities
Treat the quoted figure as an opening claim, then ask for four things.
Ask for trailing twelve-month PUE rather than design PUE. Ask whether it was measured at the hall you’d occupy or across the whole campus. Ask what sits inside the measurement boundary. And ask what the current occupancy of that hall is, because that one fact explains most of the gap between a quoted number and a delivered one.
Then put the answers next to the delivered cost per kW in that market. PUE is one input into total cost of occupancy, and on its own it will not tell you which building is cheaper to sit in.
FAQ: PUE
Q: What is a good PUE for a data center?
Under 1.3 is genuinely good for a modern facility. Between 1.3 and 1.5 is normal for a well-run existing building. Above 1.6 suggests older infrastructure, a hall running well below capacity, or a hot climate without free cooling hours. The global weighted average is about 1.54.
Q: What is the lowest PUE possible?
The theoretical floor is 1.0, meaning zero overhead, which is unreachable. The best operating facilities in cold climates with liquid cooling report figures in the low 1.1s. Treat anything claimed below 1.1 as a design figure or a narrow measurement boundary rather than an annual operating result.
Q: Does a low PUE mean lower colocation costs?
Not by itself. PUE tells you how much overhead power the building consumes per watt of IT load. What you pay depends on that ratio and the local cost of power, and the second factor varies far more between markets than the first varies between buildings.
Q: Is PUE still a useful metric?
Yes, with limits. It’s the only efficiency figure the whole industry reports the same way, so it’s useful for comparison. It just doesn’t measure water use, carbon intensity, or cost, and it can be gamed by choosing a favorable load, season, or boundary.
Q: How does AI change PUE?
High-density AI deployments raise absolute power draw sharply, but they don’t necessarily raise PUE. Liquid cooling at 50 to 100 kW per rack can beat air cooling at 10 kW on the ratio, so an AI hall can post a better PUE while consuming several times the power of the room next door. Total consumption and PUE move independently, and conflating them is a common mistake in capacity planning.
Compare Real Efficiency Across Facilities
PUE is one line in a much longer comparison. Search 4,562 data centers by market, power, tier, cooling type and AI readiness, then send one requirement and get matched options back within 24 hours.