Many data centre operators recognise the scenario. A customer wants to expand. A new AI environment needs to be deployed. An opportunity arises to accommodate additional workloads. But according to the available figures, there is little or no power capacity left. As a result, discussions begin about expanding infrastructure, investing in additional capacity, or even building a new facility.
In many data centres, power is still allocated based on nameplate ratings, theoretical peak loads and generous safety margins. That is understandable. No operator wants to risk running out of capacity when customers need to grow.
However, this can create situations where a rack is allocated 25 kW while its actual average load remains closer to 8 kW.
On paper, the capacity appears occupied.
In reality, a significant portion of the available power may never be used.
This phenomenon is often referred to as stranded capacity, and it is becoming increasingly relevant as power emerges as one of the most valuable resources in the data centre industry.
For many years, data centre development was primarily focused on space, cooling and connectivity.
Today, the conversation is increasingly centred around power.
Grid congestion is limiting new connections. Energy availability is becoming more challenging. At the same time, rack power densities are rising faster than ever before.
Where traditional enterprise racks often consumed between 3 and 8 kW, deployments of 20, 30 or even 50 kW per rack are now commonplace.
For AI and HPC environments, power densities of 80 to 100 kW per rack are rapidly becoming a reality.
As a result, every kilowatt has become more valuable.
Not only for operators, but also for customers looking to expand their IT infrastructure.
Data centres are naturally risk-averse environments.
When power capacity is allocated to a customer, calculations are often based on:
This approach makes perfect sense. The consequences of underestimating future power requirements can be significant.
However, it often results in a substantial gap between allocated capacity and actual power consumption.
For example, a server may be equipped with two 1600 W power supplies.
On paper, that suggests a potential load of 3200 W.
In reality, the same server may consume only 800 to 1200 W during most of its operational life.
When this pattern is repeated across hundreds or thousands of devices, the difference between theoretical and actual consumption can become substantial.
At first glance, stranded capacity may seem harmless.
After all, the reserved capacity remains available for future growth.
However, there are consequences.
When power is reserved but not actively used, it affects:
A facility may appear to be approaching its power limits, while a considerable portion of the allocated capacity remains largely unused.
This does not mean that all unused capacity can immediately be reassigned elsewhere. Contractual commitments, redundancy strategies and future growth plans remain important considerations.
However, without accurate measurement data, it becomes extremely difficult to determine how much capacity is genuinely required and how much is simply being reserved as a precaution.
The rise of AI is fundamentally changing the way data centres approach power management.
With traditional workloads, generous safety margins often had relatively little impact.
With AI racks consuming 50 to 100 kW or more, every planning assumption carries far greater consequences.
When multiple racks are oversized based on theoretical loads, operators may end up investing in:
The higher the rack density, the greater the value of accurate operational data.
Increasingly, data centre operators are moving beyond traditional capacity planning towards a more data-driven approach.
The question is shifting from:
“How much power could this equipment consume?”
to:
“How much power does this equipment actually consume?”
That may sound like a subtle distinction, but it fundamentally changes how capacity is managed.
Accurate measurement data allows operators to identify trends, understand growth patterns and make better-informed infrastructure decisions.
Rack-level power monitoring provides valuable visibility into overall rack consumption.
However, as environments become more complex and power densities continue to rise, operators increasingly require more detailed insight.
Questions now include:
These questions cannot always be answered through rack-level monitoring alone.
This is one of the reasons why outlet-level monitoring is attracting growing interest across the data centre industry.
By measuring the consumption of individual servers, storage systems and network devices, operators gain a far more detailed understanding of how power is actually being used.
This supports:
The objective is not simply to collect more data.
The objective is to make better decisions.
The challenges surrounding power are real.
Grid congestion is real. AI workloads are driving unprecedented demand. Access to additional capacity is becoming increasingly difficult.
However, alongside the search for more power, another challenge is emerging:
How effectively are we using the power capacity that is already available?
Power monitoring will not solve grid congestion.
It will not create additional capacity.
But accurate energy measurement can help distinguish between genuine constraints and conservative assumptions.
In a market where power is becoming one of the most valuable resources available to data centre operators, visibility may ultimately prove to be just as important as capacity itself.
Data Center Energy Management Schleifenbauer introduces the EnerTree software and the powerful PDU 5.0. This innovative combination gives data centres extensive control over energy management, secure scalability, and is available at no extra cost with your PDU 5.0. EnerTree offers real-time monitoring, comprehensive security, and an integrated interface, enabling data centres to work smarter and […]
What Is HIB? The Handbook for ICT Infrastructure and Cabling (HIB) defines the standards for passive and active ICT infrastructure in government buildings. Its goal: to create a future-proof, secure, and standardised ICT environment for all Dutch government agencies. HIB specifies, among other things: • The design of ICT rooms (EF, BD, FD); • Requirements […]
Request a Quote