Europe's Hidden Solution to the AI Energy Crunch

Jan de Vries · 2026-09-21

Earlier this year, I wrote about the AI power crisis and argued that coordination matters more than just building new capacity. Honestly, the numbers since then haven't gotten any better—they're still pretty daunting. Research from the International Energy Agency (IEA) shows global data center electricity demand is projected to roughly double, from 485 terawatt-hours in 2025 to around 950 terawatt-hours by 2030. Think about that for a second. It's growing more than four times faster than total electricity demand from all other sectors combined. But here's the interesting twist I discovered with my colleague Anna Trendewicz: much of the capacity Europe desperately needs already exists. It's just sitting there, unused most of the time. The real unlock isn't waiting years for new power lines and substations. It's software that can orchestrate what we already have, right now. We dug in and found a small group of companies quietly building this exact software across four key layers: the grid, the facility, the compute hardware, and the software itself. ### The Grid: Seeing What's Already There This layer is about making invisible capacity visible. Companies here are turning big energy users—like data centers—into flexible assets for the grid. - Amsterdam's **Sympower** and Dublin-based **GridBeyond** have each raised over $75 million. They aggregate demand from data centers and industrial users, then bid that flexibility into energy balancing markets across Europe. - Munich's **Entrix** is one of Germany's established battery storage optimizers, trading flexible capacity across different energy markets. - Finland's **Capalo AI** runs a virtual power plant that had contracted over 1 gigawatt of battery storage by the end of 2025. They've expanded from the Nordics into Poland and Bulgaria. - London's **Piclo** operates a flexibility marketplace connecting energy assets with grid operators, while Vienna's **enspired** automates AI-driven trading on European power spot markets. Here's the thing—none of these companies started as data center specialists. Most built their platforms for industrial and renewable energy flexibility first. That tells you how new and specific the AI version of this power problem really is. ### The Facility: The Physics of Efficiency This is where you optimize everything inside the four walls of a data center. It's the nuts and bolts layer. Darmstadt-based **etalytics**, a spin-off from TU Darmstadt, is a prime example. Their etaONE platform uses AI with physics-based digital twins to optimize cooling, heating, and power systems. Operators can let it adjust settings automatically or approve recommendations manually. Their customers include big names like Equinix, Digital Realty, NTT, and Telehouse. The results speak for themselves. NTT Global Data Centers cut cooling energy by 19% using their software. Telehouse Germany reduced cooling electricity by 10.4% at its Frankfurt campus—a site they already considered well-optimized. Across their customer base, etalytics reports savings up to 50%. In 2025, Microsoft's venture fund M12 led an $8.5 million extension, bringing their Series A to about $17 million. They've since opened their U.S. business. For AI operators, this isn't just about lowering the energy bill. It's about capacity. A site with a fixed grid connection can shift every kilowatt saved on cooling directly to its IT load. That turns efficiency into sellable rack space without needing a new site or interconnection. It's a game-changer. Rotterdam's **Gradyent** takes this a step further. Their real-time digital twin optimizes district heating networks in over 35 European cities. They raised about $30 million in a 2025 Series B. Their platform can actually route a data center's waste heat into a city's heating grid, warming homes instead of just venting it into the air. Now that's a clever use of resources. ### The Compute Layer: Smarter Workload Management This final layer is about recovering efficiency at the hardware level—making sure the actual AI computation happens in the most efficient way possible. Paris-based **FlexAI** launched in 2024 with a $30.5 million Seed round. Their software places AI training and inference workloads across different hardware from various vendors and clouds. The goal? To ensure customers aren't locked into a single generation of GPUs. Bristol-based **YellowDog** takes a similar approach, scheduling workloads across on-premises, hybrid, and multi-cloud fleets to maximize efficiency and minimize cost. So what's the big takeaway here? Europe isn't just waiting for new power plants to solve its AI energy challenge. A new ecosystem of software-first companies is emerging to unlock the hidden capacity we already have. They're working across every layer of the problem, from the grid down to the individual server. And in doing so, they're not just fixing a power problem—they're building a more resilient, efficient foundation for Europe's AI future. The coordination I talked about earlier? It's already happening, one software platform at a time.