There is a number the data center industry watches more closely than almost any other. It appears in annual reports, keynote slides, and sustainability pledges. It's how operators defend their efficiency to investors, regulators, and enterprise tenants. It's the metric that's supposed to prove the industry is getting better.
That number is PUE: Power Usage Effectiveness. And for six consecutive years, the global average has barely moved.
It sits at approximately 1.5. It was 1.5 before the AI boom. It was 1.5 when hyperscalers announced trillion-dollar infrastructure plans. It is 1.5 today, as data center power demand is projected to double by 2030 and grid operators in the world's largest markets are quietly preparing for a supply crisis.
That persistent 1.5 is not a coincidence. It is not a measurement artifact. It is a structural signal, and most of the industry is misreading what it actually means.
What PUE Is, and What It Hides
PUE is deceptively simple. Divide total facility power consumption by the power consumed by IT equipment alone. A perfect score is 1.0, meaning every watt entering the building goes directly to computing. A score of 1.5 means that for every watt powering a server, an additional 0.5 watts are consumed by cooling, lighting, power conversion losses, and everything else that keeps the facility running.
The metric was introduced by The Green Grid in 2007 as a baseline tool to bring some standardization to an industry that had none. For its original purpose, establishing a common language, it worked. But the industry then did something predictable with a simple, single-number metric: it optimized for the number itself, rather than for what the number was supposed to represent.
Why the Gains Stopped
When PUE became the dominant metric, facilities discovered that the fastest way to improve their score was to focus on overhead reduction: more efficient cooling infrastructure, better airflow management, economizer modes that leverage outside air, higher operating temperatures that reduce chiller load. These are real improvements, and they delivered real gains, from average scores above 2.0 in the mid-2000s down to the 1.5 range by the mid-2010s.
Then the gains stopped.
Not because the industry ran out of ideas. Because the improvements that remained required addressing something far more difficult than infrastructure upgrades: the way facilities are actually operated on a continuous basis.
Cooling efficiency isn't determined once when a system is designed. It degrades the moment conditions change, and in a data center, conditions are always changing.
Load patterns shift. Seasons change. Equipment ages and fails in ways that alter airflow. New hardware with different thermal profiles gets installed in racks that weren't designed for it. Hot spots migrate. Cold air bypasses. Cooling systems respond to static setpoints rather than dynamic conditions.
Every one of these deviations from design intent represents energy waste. And every one of them is invisible unless someone is actively watching, interpreting, and acting on the signals, continuously and in real time.
Most facilities aren't doing that. They're monitoring. And monitoring is not the same as managing.
The Numbers Behind the Stagnation
The Uptime Institute's 2025 Global Data Center Survey, the most comprehensive annual benchmarking study in the industry, confirmed what operators already suspected but rarely say publicly: meaningful improvement in facility efficiency has stalled. The survey, now in its 15th year, found that sustainability metrics collection across the industry has not materially improved, even as the commercial pressure to demonstrate efficiency has intensified.
Gartner projects data center electricity demand to grow 16% in 2025 alone and to double by 2030. Goldman Sachs estimates a 165% increase in data center power demand by the end of this decade. At a PUE of 1.5, one-third of that doubled load (tens of gigawatts of generation capacity, hundreds of billions of dollars of grid infrastructure) will produce no compute output. It will simply be overhead.
Find out what your facility is actually costing you.
Book a demo ↗Gaming the Metric
There is a second dimension to PUE stagnation that the industry rarely addresses in public: the growing gap between reported PUE and operational reality.
PUE, as a metric, can be optimized in ways that improve the number without improving the underlying operations. Facilities can measure PUE during optimal conditions and report that figure. They can locate facilities in cold climates that allow extended economizer operation, effectively subsidizing efficiency with geography rather than operational discipline. They can exclude certain load categories from the denominator. They can time measurements to coincide with periods of higher IT utilization, which mechanically reduces the overhead ratio.
None of these practices are fraudulent. But collectively, they mean that industry-wide PUE figures systematically understate the actual efficiency gap. The reported 1.5 is likely better than the lived operational reality for a significant portion of the installed base.
What the EU Already Knows
Regulators have moved faster than the industry on this question. The European Union's Energy Efficiency Directive, now in force for facilities above 500 kilowatts, requires annual reporting of PUE, WUE, energy consumption, temperature set points, waste heat utilization, and renewable energy share. But the EU didn't stop at reporting requirements.
New facilities in cool climates must achieve a PUE of 1.3 or better. New facilities in warm climates must achieve 1.4. The EU's Data Centre Energy Efficiency Package, expected in 2026, is anticipated to introduce a binding performance rating scheme and potentially minimum standards for the existing installed base.
Reporting becomes baseline. Baselines become targets. Targets become mandates.
The US is following, with state-level legislation accelerating sharply: over 238 data center bills were considered across all 50 states in 2025, with 40 or more enacted. In the first six weeks of 2026 alone, more than 300 bills were filed across 30 states.
Beyond PUE: The Metrics That Will Replace It
The industry has begun to recognize that PUE is insufficient as a standalone measure of operational quality. Several successor metrics have emerged, each attempting to capture a dimension that PUE obscures:
WUE (Water Usage Effectiveness) tracks water consumption per unit of computing energy, now mandatory under the EU EED and increasingly requested by enterprise tenants with supply chain sustainability obligations.
CUE (Carbon Usage Effectiveness) measures carbon dioxide equivalent emissions per computing energy consumed, relevant as carbon pricing and Scope 2 emissions accounting reach data center buyers.
ERE (Energy Reuse Effectiveness) credits facilities that export waste heat for productive use (district heating, agricultural applications, industrial processes), rewarding what PUE ignores entirely.
GEC (Green Energy Coefficient) captures the proportion of consumption met by renewable sources, separating efficiency from sourcing in a way PUE cannot.
And increasingly, operators and analysts are proposing PCE, Power Compute Effectiveness, which asks the most operationally honest question: how much sustained, usable compute output is produced per unit of power consumed under real operating conditions?
PCE is uncomfortable because it forces a reckoning that PUE allows facilities to avoid. Two buildings with identical PUE scores can deliver radically different compute yield depending on how aggressively their cooling is managed, how tightly their power infrastructure is controlled, and how effectively their operational teams respond to dynamic conditions. PCE makes that variance visible. It rewards operational discipline, not just infrastructure investment.
The Real Question PUE Stagnation Raises
Six years at 1.5 is not a technology failure. The technology to operate at 1.3 or better exists and has existed. Best-in-class hyperscale facilities routinely achieve annualized PUEs in the 1.1 to 1.2 range. The knowledge exists.
The gap is operational. It is the difference between knowing what a facility is doing and actively controlling what a facility does.
Monitoring infrastructure tells operators what PUE is right now. It does not identify which specific cooling decisions, thermal conditions, or operational patterns are driving that number up. It does not detect the drift between design intent and operational reality. It does not act on the signal that a cooling loop is running ten degrees warmer than optimal, or that airflow through a particular row has degraded since a maintenance cycle, or that a weather shift tonight creates an opportunity to reduce chiller load by 15% without any risk to thermal margins.
Those decisions require continuous, real-time interpretation of thousands of simultaneous variables, and then action. Not a report. Not an alert for a human to investigate and eventually respond to. Action. Coordinated adjustment across the interdependent systems that collectively determine whether a facility runs at 1.3 or 1.7.
This is the gap PUE stagnation reveals. Not a hardware gap. Not a knowledge gap. An operational governance gap: the absence of active, continuous control over the systems that determine efficiency.
That gap is not getting smaller as AI workloads increase density, as liquid cooling introduces tighter operational tolerances, and as regulatory scrutiny requires defensible, auditable proof of efficiency performance.
It is getting larger.
And that is why the number that hasn't moved in six years should be the most urgent number in the room.