ComputerWeekly

The fast fix for Europe’s datacentre gap may already be plugged in


Most of us can remember burning a finger on an old-style incandescent light bulb.

The typical bulb back then drew 60 watts of power. You might even remember that they had the wattage stamped on the bulb itself. What is truly intriguing is that those old bulbs released about 90% of their energy as heat, and only about 10% became light.

Fast forward to today and that old 60-watt light bulb’s current-day LED replacement tells a different story. It produced about 800 lumens of visible light from its 60-watt rating, whereas a new LED can produce that same 800 lumens from about 9 watts. It almost stays cool enough to hold in your hand while illuminated. Same light, about one seventh of the power.

Keep that bulb in mind. It is the best picture I know of to explain what is happening in datacentres. It also shows why the gap between countries matters to people who may never see a datacentre.

Watts a household can picture

Megawatts mean little to most people, but watts in light bulbs do. So, I divided several countries’ installed datacentre capacity by their populations and households, then flipped the result back into 60-watt bulbs.

These are installed IT capacity figures from national industry bodies and brokers, for 2025 or early 2026. Definitions differ a little, so read them as scale, not precision. They count capacity built, not power drawn.

An American household has nearly four bulbs of datacentre capacity behind it. A British household has one. A Spanish or Italian household has less than half of one, roughly a tenth of the American level.

Why should we care? Think of datacentre capacity the way earlier generations thought about power lines. Towns that got the lines got a head start. Studies of the 1930s US rollout found farm output held up better and property values rose where power arrived.

Capacity is the road that jobs and patients travel on.

Take the clinic. AI software helps read a scan sooner, and in England stroke software now reaches every stroke centre. An early NHS analysis linked it to treatment more than an hour sooner. Then there are the records of every patient, student and taxpayer, which sit in a building under some country’s laws.

People in Madrid don’t need to understand a megawatt, but they do need those services to be close, fast and affordable. Whether they get them depends on the capacity a country has built.

Waiting for Watts

Europe made hard choices about power. Germany closed its last three nuclear reactors in April 2023. Britain closed its last coal plant in September 2024. There were good reasons for both. But now datacentre builders face a long line. According to commercial property agents Colliers, UK datacentres wait more than eight years on average for a grid connection, while Ireland and Germany are also acutely grid constrained.

It would not be fully accurate to blame all the delay on power plant closures. Grid build-out, a sharp rise in demand and slow planning systems all play a part. But the effect on a community is the same. A site that waits eight years for power is eight years of services that never reach the community.

But the power is already at the door

With waits like that, the fastest solution sits in buildings that already have light bulbs lit up. In very basic terms, it means upgrading select older datacentres so the same grid connection can provision far more of its power for compute. 

Here’s where the engineering matters. Older datacentres often have substantial power provisioned for cooling and other supporting systems. Upgrading that infrastructure with modern cooling equipment, variable-speed compressors and fans, and more effective power distribution can reduce the power required for those supporting systems, to allow more of the existing provisioned capacity to be allocated to compute. 

It is not the only route, since some new sites are needed and some countries lack older stock to convert. But where the queue is the bottleneck, conversion can be and often should be the quickest first step.

Reading the meter

How do we know a conversion worked? A 50MW connection doesn’t tell us how much of that power is set aside for compute versus cooling and other overhead.

The datacentre metric PCE, Power Compute Effectiveness, was built for this question. PCE divides the IT compute power a site provisions by the total datacentre power it provisions. 

And PCE is a design-stage number that can be declared on a stated basis. Say a site has a PCE score of 0.6. That tells us that it sets aside 60% of its power for potential computing. The rest of the power is provisioned for cooling, conversion losses and other overhead. Say that same site scores 0.85 after conversion.

Think of the meter on your house wall. It sets a fixed supply. Heating, pumps and wiring losses take a share, and the lamps get what is left. A higher PCE doesn’t mean a better bulb. It means more of the supply at the meter is set aside for the bulbs.

The picture is still loose. The LED story earlier was about efficiency; a better device making the same light from less power. PCE is about allocation. Nearly all the power a server uses ends up as heat too. PCE does not measure heat, and it does not measure real-time use. It measures how provisioned power is split. The design decides how much of the connection is set aside for the job that needs to be done.

The Open Compute Project published the PCE white paper this month. It is free for anyone to use. It is also new, but its true value depends on whether governments, utilities and operators actively adopt and use it.

When utilities and governments ration scarce connections, they should ask for a PCE figure alongside megawatts. Two projects asking for 50 MW are not the same project if one sets aside 60% for computing and the other 85%. The second provisions 12.5 MW more of the same scarce connection for compute, and that difference is what a PCE figure makes visible. A conversion still takes investment in cooling and electrical upgrades. But it starts with power that is already connected, so the wait for a connection should be shorter.

Back to the question

A burned finger taught us all that every watt goes somewhere. The meter on the wall decides how many there are to spend. Every provisioned watt is either set aside for compute or for supporting the datacentre itself. The design decides how the connection is divided.

So, what is each country’s opportunity to advance, learn and establish a data-ready society? Part of the answer sits in the power it already has. The rest depends on how it divides that power.

Paul Quigley is chief strategic relations officer at Airsys Cooling Technologies and co-author of the PCE white paper published by the Open Compute Project.



Source link