AI boom strains Europe's power grids as operators look for unconventional fixes

The race for AI computing capacity has hit an unexpected obstacle: a plain shortage of electricity. Across Europe, data center developers are lining up in multi-year queues for grid connections, and infrastructure operators are scrambling to find unconventional ways to free up capacity.

The problem grows every quarter. The biggest tech companies — Microsoft, Google, Amazon, Meta* — are aggressively expanding their European infrastructure, driven by EU data localization rules and fast-growing demand for cloud AI services. In 2023–2024 alone, confirmed investment in European data centers exceeded 40 billion euros. Each large facility draws hundreds of megawatts, comparable to the consumption of a small city.

Grid operators were not ready for this. The Netherlands, Ireland and several regions of the UK have introduced temporary moratoria on new tech connections. In some cases the queue has stretched to 5–10 years. The problem is not only electricity itself: the grid infrastructure — transformers, high-voltage cables, distribution substations — needs large-scale modernization, which physically takes years even when funding and political will are in place.

In response, operators are starting to test fundamentally new connection models. One is the flexible contract: a data center gets access to the grid but commits to reducing consumption during peak load hours. In practice, this means server capacity may be temporarily cut in the most strained periods, such as cold winter evenings or when renewable generation falls short. For workloads that do not need instant response, above all training large language models, this mode is entirely workable.

Another tool is real-time demand management. Modern algorithms can automatically redistribute load among several facilities based on current congestion on grid segments and wholesale electricity prices.

Some operators are studying virtual power plants: aggregated systems that combine large consumers, energy storage and renewable generation into a single managed structure. In theory, this smooths consumption peaks without building new lines and substations.

Energy is becoming a strategic issue for the entire AI industry. Analysts forecast that by 2030 data centers will consume 8 to 15% of all electricity in Europe, versus 2–3% today. At the same time, the EU is tightening decarbonization requirements: new facilities must run mainly on renewable sources, which without large-scale backup cannot yet provide stable round-the-clock supply.

The situation exposes a structural contradiction. Governments compete for AI infrastructure investment by offering subsidies and tax breaks, but the physical limits of power systems prevent them from delivering on those commitments within acceptable timeframes. In the race for capacity, the winners are those who locked in long-term energy contracts early or are considering building their own generation. It is no coincidence that several large tech companies are already negotiating to site small nuclear reactors near their data centers.

*Meta has been designated an extremist organization and banned in Russia.