Europe's ambition to build its own artificial intelligence infrastructure is colliding with a hard reality: the electricity system may not be able to keep up. Data-centre power consumption across the continent is projected to jump from 96 terawatt-hours (TWh) in 2024 to 236 TWh by 2035, according to the UK-based energy think tank Ember. That surge comes as the EU and national governments push to expand AI capacity, but every new facility adds pressure to grids already struggling with ageing infrastructure and rising demand from electrification.
The result is a paradox: Europe needs more data centres to compete in AI, yet the very power they require makes it harder to connect them. Grid connection queues are lengthening, and in some regions, new projects face years of delay. But a growing number of startups across the continent are betting that smarter use of existing capacity can ease the crunch.
Cooling: the hidden energy hog
Servers generate enormous amounts of heat, and keeping them cool accounts for a significant share of a data centre's electricity bill. Traditional air conditioning is energy-intensive, but several European companies are rethinking the approach.
UK-based EkkoSense has developed a wireless sensor system that monitors temperature, airflow, and power usage inside data centre rooms. Its software creates a three-dimensional map of the environment, allowing operators to spot servers that are being over- or under-cooled. In 2024, the company deployed its system across 20 sites operated by Virgin Media O2, one of Britain's largest telecoms firms. The result: cooling energy dropped by an average of 15%, saving more than £1 million (around €1.17 million) annually and cutting carbon emissions by the equivalent of 760 tonnes, according to the company.
German startup etalytics takes a different tack. Instead of watching the room, it simulates the entire cooling system—chillers, pumps, heat exchangers—in software. The tool tests how these components would perform under various conditions and calculates the most efficient operating settings while keeping temperatures safe. Japanese tech giant NTT trialled the software at its data centre in Bonn, and the electricity used by its chillers fell by 19.1% in the first months. NTT expects savings of up to 25% over a full year.
Spanish startup Submer goes further by changing the hardware itself. Its systems immerse servers in a non-conductive liquid that carries heat away directly, eliminating the need for energy-hungry fans and air conditioning. Telefónica, which uses Submer's technology, says the liquid cooling technique can boost energy efficiency by up to 50%, reduce refrigerant-related emissions, shrink the physical footprint, and improve total cost of ownership.
Batteries with a second job
Data centres already rely on batteries to keep servers running during outages. Irish startup GridBeyond has found a way to make those batteries work harder. Its software controls when batteries charge and discharge based on real-time grid conditions. When electricity is cheap and abundant, the batteries charge; when demand spikes, they can supply the facility, reducing the amount drawn from the grid at peak times.
GridBeyond has installed its software at two Dublin data centres owned by Singapore's Keppel DC REIT, providing a combined 8 megawatts of flexible capacity. This kind of demand-side management is becoming increasingly valuable as grids integrate more intermittent renewables and face tighter supply margins.
Turning waste heat into a community resource
Another approach is to stop treating heat as a problem and start using it. UK startup Deep Green places compact computing units next to buildings that need heat—swimming pools, district heating networks, even greenhouses. Liquid warmed by the servers carries that heat into the building's heating system, displacing gas or electric boilers.
In a trial in southwest England, Deep Green told the Exmouth Journal that donating heat to a swimming pool would cut its gas use by 62%, save more than £20,000 (around €23,300) a year, and reduce annual carbon emissions by 25.8 tonnes. CEO Mark Bjornsgaard explained: "By moving data centres from industrial warehouses into the hearts of communities, our 'digital boilers' put waste heat to good use, saving local businesses thousands of pounds on energy bills and reducing their carbon footprint. Pools are just the start—around 30% of all industrial and commercial heat needs could be provided by this technology."
This concept is gaining traction beyond the UK. Several European cities are exploring ways to capture waste heat from data centres and feed it into district heating networks, a practice that could help reduce reliance on fossil fuels. As Europe's overlooked answer to cutting gas reliance becomes more prominent, startups like Deep Green are proving that the idea can work at a local scale.
Efficiency is not enough
These innovations are valuable, but they cannot replace physical expansion. As AI workloads grow, Europe will need more electricity generation and stronger grid infrastructure. According to the International Energy Agency, new grid projects can take between five and 15 years to plan, approve, and complete. That timeline is out of step with the pace of AI development.
Some policymakers are already looking at ways to speed up grid connections and prioritise projects that support digitalisation. The European Commission has proposed measures to improve grid planning and cost-sharing, but implementation remains uneven across member states. Meanwhile, startups are showing that immediate gains are possible—if utilities and data centre operators are willing to adopt them.
The challenge is not just technical but also regulatory. In many countries, connecting a new data centre to the grid requires lengthy permitting processes, and the rules for using waste heat or selling flexibility back to the grid vary widely. Harmonising these frameworks could unlock more of the potential that startups are demonstrating.
As the EU pushes forward with its digital agenda, the question is whether the power system can keep pace. The startups profiled here offer a glimpse of a more efficient future, but they are not a silver bullet. Europe will need a combination of smarter operations, new generation capacity, and grid investment to avoid becoming the bottleneck of its own AI ambitions.
