AI Is Changing What a Data Centre Is, And Even How It Uses Power

As AI pushes computing demand higher, data centres are beginning to evolve from isolated power-hungry facilities into infrastructure that can interact with electricity grids, reuse heat and play a larger role in the energy system.

For most people, a data centre is something they never see.

Yet almost every digital action, making a card payment, joining a video call, streaming content, accessing cloud software or using an online government service, eventually touches one.

Artificial intelligence is now putting these largely invisible buildings under unprecedented pressure.

The rapid expansion of AI computing means data centres need more power, more cooling and more physical infrastructure than before. But the conversation around them is also beginning to change. Instead of treating data centres simply as huge consumers of electricity, the industry is increasingly exploring whether they can become a more integrated part of the wider energy ecosystem.

That was one of the themes to emerge from the European Data Center Associations Summit hosted by Vertiv, where representatives from several European markets discussed how AI is changing the infrastructure behind the digital economy.

And the shift could be significant.

Data centres may have to work with the grid

Electricity has become one of the defining issues facing the next generation of data centres.

AI workloads require enormous amounts of computing power, while the GPUs used to run them are becoming increasingly dense and energy intensive. At the same time, electricity grids in many markets are already facing pressure from electrification, renewable-energy expansion and ageing infrastructure.

Traditionally, the relationship was fairly straightforward: a data centre needed electricity, and the grid supplied it.

That model is beginning to evolve.

Stijn Grove, Managing Director of the Dutch Data Center Association, highlighted how data centres could increasingly be located closer to renewable-energy generation, helping make use of electricity where it is produced rather than relying entirely on new transmission infrastructure.

The bigger idea is that a data centre could eventually become more than a passive electricity customer.

With batteries, backup power systems and sophisticated energy controls already part of many facilities, data centres may be able to support the grid during periods of high demand or fluctuating renewable-energy output.

Ronan Kelly, CEO of Digital Infrastructure Ireland, similarly pointed to battery energy-storage systems and existing backup infrastructure as technologies that could potentially contribute to grid stability.

That changes the question.

Instead of simply asking, “Where can we find enough electricity for the next data centre?”, operators may increasingly ask, “How can the data centre become part of the energy system itself?”

Even the heat could have another life

Then there is the heat.

Servers generate a huge amount of it, and removing that heat is one of the biggest challenges in operating a data centre.

Historically, that heat was largely treated as something that had to be removed and discarded.

Increasingly, it is being viewed as a resource.

Isabelle Kemlin, Vice Chair of the Board at Swedish Datacenter Industry, discussed examples in which waste heat from data centres is fed into district-heating networks or used in other local applications.

Similar ideas are being explored elsewhere in Europe.

Grove pointed to projects in the Netherlands where heat generated by data-centre infrastructure can help reduce the use of natural gas in buildings and potentially support greenhouse operations.

It turns a familiar criticism of data centres on its head.

A facility may consume a large amount of electricity and produce heat, but if that heat can be captured and reused, the data centre becomes part of another energy loop rather than simply an isolated industrial building.

That does not eliminate its environmental impact. But it changes how the overall system can be designed.

AI is also changing how efficiency is measured

For years, one of the most familiar ways of assessing a data centre has been Power Usage Effectiveness, or PUE.

Put simply, PUE compares the total electricity used by a data centre with the amount of electricity that actually reaches its computing equipment.

The closer the number gets to 1.0, the less power is being consumed by supporting infrastructure such as cooling.

But AI is complicating the picture.

Imagine, for example, that a facility installs additional equipment to capture waste heat and send it into a local heating system.

That equipment itself may consume electricity.

The data centre’s PUE could therefore become slightly worse, even though the larger energy system may have become more efficient because another building no longer needs as much separate heating.

This is one reason the industry is beginning to explore other ways of thinking about efficiency.

One increasingly discussed idea is tokens per watt.

AI models generate and process tokens — essentially units of text or information. Tokens per watt therefore asks a different question: how much useful AI output can be produced from a given amount of electricity?

It shifts the focus from simply minimising the overhead of a data centre to maximising the amount of useful computing generated from the energy being consumed.

As AI infrastructure becomes larger, that distinction could matter considerably.

Where your data lives still matters

There is another reason Europe cannot simply place all of its computing infrastructure wherever electricity is cheapest: digital sovereignty.

Governments, financial institutions, healthcare providers and other regulated organisations often face rules or requirements concerning where information is stored and processed.

Latency matters too.

A workload running thousands of kilometres away may technically be accessible, but some services need data and computing resources much closer to their users.

Europe’s collection of individual countries, regulations and public-sector requirements therefore creates a more complex infrastructure landscape.

The result is that AI does not necessarily mean every workload will be concentrated into a handful of gigantic global computing hubs.

Local and regional data centres will remain important.

The industry still has an image problem

The technical challenges may be difficult, but public perception could prove just as important.

Data centres are increasingly scrutinised for their electricity consumption, water usage, physical footprint and the relatively small number of people required to operate very large facilities.

Several speakers at the summit acknowledged that the sector has struggled to explain its wider role.

Michaël Reffay, Managing Director of France Datacenter, argued that discussions about data-centre consumption can overlook the economic activity and digital services the infrastructure enables.

Kelly raised a similar point around carbon emissions.

A data centre clearly consumes electricity. But the services operating through it can also reduce other activities — business travel through video conferencing, physical distribution through digital services, or visits to physical branches through online banking.

That does not mean digital infrastructure has no environmental cost.

It means its impact is more complicated than looking at the electricity meter outside the building.

And as AI drives that electricity consumption higher, explaining the relationship between energy use and the economic or social value being produced will become increasingly important.

AI is turning infrastructure into strategy

For years, computing infrastructure largely sat in the background.

Businesses cared about their applications, their software and their digital services. The buildings, cooling systems and electricity networks underneath them rarely entered mainstream conversation.

AI is changing that.

Access to electricity can now influence where billions of dollars of computing infrastructure are built.

Cooling technology can determine how densely AI hardware can operate.

Grid connections can affect how quickly an AI project comes online.

And the ability to turn electricity into useful computing output is increasingly becoming a competitive consideration.

The next stage of the AI race may therefore depend on far more than who owns the best algorithm or the most powerful GPU. It may also depend on something much more physical!