Even the most efficiently designed modern facilities experience energy loss in small, often invisible ways. Individually, these losses are insignificant and rarely worth writing in a sustainability report. Headline initiatives may be about the major strategic shifts, but the uncomfortable truth is that data center sustainability is not about the big transformation; it is about the follow-through.

For boards overseeing data center investments, sustainability conversations tend to gravitate to the initiatives that are easier to present and promote: capital projects with clear before-and-after metrics, declarations and commitments to renewable energy, and efficiency targets. But they are not where most of the avoidable costs lie.

The bigger financial exposure is often in the less glamorous, less visible details: the ongoing operational discipline required to keep a data center efficient after it's been built. Without it, even a well-designed, well-capitalized facility will bleed money, unevidenced as a single line item that anyone can point a finger at.

This is a different kind of risk than a cost overrun or an outage. It doesn't announce itself. It accumulates.

Risk profiles

Capital initiatives: Replacing aging cooling infrastructure and redesigning airflow containment are conspicuous, fundable, and accountable initiatives. These go through a capital approval process, have a defined ROI case, and produce a measurable result. Organization is generally good at executing this kind of work because it fits existing governance structures.

Operational discipline: The continuous, dull work of maintaining what's already been built does not have an approval gate. There often isn’t an obvious accountable owner, and, without a dramatic failure that forces it onto the agenda, it is most likely to be underfunded and under-governed. But it can represent a comparable share of total energy waste.

Where the losses lie

Physical waste: While small gaps around rails, floor tiles, and rack openings may not cause immediate problems, over time they add up to significant wasted energy.

There’s no failing equipment, so no incident report gets filed. However, left unmanaged, this energy waste can compound, forcing cooling systems to work harder for the same output. This is easy to deprioritize, but expensive to ignore.

Strategic waste: Strategic inefficiencies exacerbate the physical ones. Zombie servers draw power without delivering value, and outdated hardware continues to run well past the point where consolidation or replacement would be cheaper. But to avoid these inefficiencies means the component candidates need to be identified and the resources committed – projects often deprioritized amid daily operations.

Sustained investment

In both cases, the financial exposure isn't caused by a bad decision. It's caused by the absence of a decision. A gap that nobody is responsible for closing.

Capital expenditure gets one moment of intense scrutiny, and then it's done. Operational discipline doesn't work that way. It's a continuous process with no single sign-off, no individual decision, and no moment where someone needs to ask, "Is this still efficient?"

One large, visible cost is easy to capture. What’s not easy is to catch a hundred small ones, none of which individually justifies an investigation. The result is a facility that looks the same on paper as the day it was commissioned while performing materially worse in practice.

The big issues

1. PUE decay

A facility commissioned at a PUE of 1.3 rarely stays there. Modern hyperscale facilities are routinely engineered toward a PUE near 1.1–1.2. In 2026, enterprise data centers typically operate between PUE 1.6–2.0 (higher for legacy sites), while modern colocation facilities generally achieve PUE in the 1.4–1.6 range and that gap often isn't a design flaw; it's accumulated drift. Common contributors are:

• Containment breaches. Missing blanking panels, gaps around cable cutouts, and unsealed floor tiles that let hot and cold air mix. Each individual gap is tiny. Dozens of them, collected over years of moves, adds and changes, are not.
• Cooling systems. Reduced airflow and heat transfer efficiency creep in gradually, and the cooling system compensates by working harder rather than failing outright, so there's no alarm, just a slowly rising kWh.
• Sensor drift. Miscalibrated temperature and humidity sensors cause CRAC/CRAH units to overcool, undercool, or fight each other across zones.
• Orphaned adjustments. Changes made during an incident (a hot spot, a vendor visit, a one-off load event) that are never reverted once the original condition resolves.

None of these triggers a ticket on its own. They show up as anomalies that are easy to attribute to "weather" or "load” or "just how the building runs."

2. Stranded capacity

Operators frequently overprovision power and cooling per rack to be safe, then never revisit allocations as the real draw becomes clear. The result is that the facility reads as fully allocated, i.e., power and cooling spoken for, while the actual utilization sits at 40–60 percent of that allocation.

Every additional kW of usable capacity therefore ends up costing far more than the original design model assumed. It can also distort expansion decisions, with operators concluding more space might need to be found when the real problem is that existing capacity has never been reclaimed.

3. Keeping the lights on

Industry surveys have repeatedly found large shares of physical servers are "comatose” – drawing power while doing little or no useful work. An asset goes idle when a project ends or a migration completes, and nobody formally retires it. This results in both wasted power draw and wasted rack space that could otherwise be reclaimed as usable capacity. This is a tracking and decommissioning process problem.

4. Deferred maintenance masquerading as savings

Skipping or stretching maintenance cycles can produce a visible short-term lower maintenance spend but can become an invisible long-term liability. At best, nothing bad happens for years. The deferral looks like a good decision, until a single failure event such as an unplanned outage or an emergency vendor callout presents a loss that dwarfs every dollar saved.

The dangerous part of this pattern is that it's self-reinforcing. Because the early years look fine, deferred maintenance gets read as evidence the schedule was conservative to begin with, rather than as a risk that hasn't come due yet.

5. Vendor/contract creep

Service contracts, licensing, and managed-service agreements go unreviewed. Costs can escalate via auto-renewals, by increased pricing that hasn't been renegotiated against market rates, or because the subscription is just not needed anymore.

Once again, no single renewal is the problem. It's the absence of a periodic audit.

The common thread

All these issues share the same consequences: small, individually defensible decisions or omissions that are repeated or left unaddressed over time. They amalgamate into a real and substantial cost. This is not a failure of engineering. It is a failure, or absence, of an ongoing process designed to catch drift before it compounds.

This is also why it's organizationally hard to fix. Capital projects have natural owners and natural review cycles built into how they're funded. Operational discipline doesn't. These are metrics that don't show up until something has already gone quietly wrong for a long time. 

The bottom line for the board

Sustainability commitments are typically judged by the visible initiatives a company announces. The actual cost discipline that ascertains whether those commitments translate into real savings is determined by something much quieter: whether routine operational maintenance is treated as a governed business function or an informal activity.

The most effective oversight the board can provide here isn't by approving the next capital project. It's ensuring the organization has clear and durable ownership of the unexciting operational diligence so that any efficiency gains achieved through capital investment aren't eroded by the absence of follow-through.

The one big question any data center needs to ask itself: is there a named owner for ongoing efficiency, with a recurring review, independent of incident response?

If the answer is no, drift isn't a risk. It's the default state.