Europe’s data centres will consume roughly 950 TWh of electricity by 2030, nearly double last year’s figure, according to the International Energy Agency. The surge threatens to outpace grid capacity across the continent, yet much of the needed headroom already exists, sitting idle inside infrastructure that runs far below its potential.
The problem no longer centres on building more power plants. Connection queues for new generation and transmission projects stretch years beyond construction timelines in Europe. Meanwhile, software that coordinates existing assets can unlock capacity within months.
A handful of European startups have begun attacking this bottleneck across four distinct layers. Amsterdam’s Sympower and Dublin-based GridBeyond, each with over €70 million raised, aggregate industrial and data centre demand into flexible assets traded across balancing markets. Munich’s Entrix optimises battery storage across day-ahead and intraday trading, while Finland’s Capalo AI operates a virtual power plant that surpassed 1 GW of contracted battery capacity in 2025. London’s Piclo and Vienna’s enspired run marketplaces and AI-driven trading for distributed energy assets.
On the facility side, Darmstadt spin-off etalytics applies physics-based digital twins to cooling and power systems. NTT Global Data Centers cut cooling energy by 19%, and Telehouse Germany saved 10.4% at a Frankfurt campus already deemed optimised. Microsoft’s M12 led an €8 million extension in 2025, bringing etalytics’ Series A to €16 million. Rotterdam’s Gradyent, meanwhile, routes data centre waste heat into district heating networks across 35 European cities after raising €28 million.
Compute and software layers remain the earliest innings. Paris-based FlexAI and Bristol’s YellowDog orchestrate workloads across heterogeneous hardware fleets. San Sebastián’s Multiverse Computing compresses large language models by up to 95% and reached unicorn status in 2026. Munich-Paris startup Pruna AI has open-sourced model optimisation tools.
No European company yet connects grid, facility, and compute decisions into one coordinated system. That integration layer remains vacant. Stanford’s AI Index shows inference costs fell 280-fold between 2022 and 2024 through software efficiency alone, a signal that the biggest near-term unlocking of AI capacity may come not from new infrastructure, but from the startups knitting together what already exists.
