We often speak about AI as software, models, and data. But beneath the surface, something far more physical will determine who leads and who follows. Europe’s AI future will not be decided by algorithms alone. It will be decided by electricity.
By: Per Imer, CEO, Homerunner
Contains: 634 words
Every technological wave reaches a moment when its true limitation becomes clear. In the AI era, that limitation is not chips, talent, or capital. It is power.
What we are currently witnessing in PJM Interconnection, the grid at the heart of the United States’ AI expansion, is not an isolated American issue. It is a preview.
Data center demand is rising faster than reliable generation capacity can be brought online, even when accounting for effective load carrying capability, the power that is actually available during grid stress. The result is structural shortages, rising capacity prices, and growing political anxiety.
This is the moment when Europe should pause and think carefully.
For years, we have spoken about digital infrastructure as if it were weightless. Cloud, compute, and AI workloads sound abstract and intangible. The reality is physical and industrial. AI consumes megawatts.
In the PJM region, even aggressive expansion of gas, solar, wind, and battery capacity is failing to keep pace with projected data center demand through 2030. This is not a temporary mismatch that can be resolved with incremental supply. It reflects a deeper structural miscalculation in how energy systems have been designed relative to exponential digital growth.
AI does not merely transform software. It reshapes energy architecture.
We continue building data centers as if the grid were unlimited, resilient, and inexpensive. It is none of those things. Especially not in Europe.
Local grid bottlenecks are common rather than exceptional. Grid expansion frequently takes ten to fifteen years. Public resistance to new transmission infrastructure is rising. Assuming that we can meet AI-era demand simply by constructing larger and more centralized data centers is strategic complacency.
It resembles trying to solve urban congestion by adding one more highway through the city center.
Europe operates a sophisticated electricity market. What we lack is an energy system designed for exponential demand growth. Denmark offers a clear example. It is not an energy island but a transit country between Scandinavia and continental Europe.
In practice, it is often unclear which electricity is actually consumed domestically, which has been traded contractually, and which merely flows through the system. Electricity does not follow contracts. It follows impedance. Predictability therefore becomes a scarce resource, precisely what AI infrastructure requires most.
There is frequent discussion of large interconnected high voltage direct current supergrids as Europe’s energy future. On paper, this vision appears elegant. In reality, it is technically and economically complex. HVDC converter stations are extremely expensive, operationally sensitive, and difficult to scale.
Over shorter distances and within the same frequency domain, alternating current remains more efficient and robust.
HVDC solutions make sense for long subsea cables, typically in point to point configurations. Large scale multi terminal HVDC networks are still largely theoretical. This is one reason why many energy island projects ultimately rely on alternating current internally. Not because it is visionary, but because it works.
In grid constrained regions, power generation should not simply be something data centers connect to. It should be something they integrate and partially own.
On site energy strategies may include transitional gas generation, small modular nuclear reactors, local battery storage, thermal storage, or direct integration with industrial waste heat. These approaches fundamentally change the equation. They reduce pressure on transmission networks, shorten deployment timelines, and create real accountability for consumption.
Decentralization is not ideological. It is architectural.
One critical detail in the AI economy is often overlooked. A single AI query consumes significantly more energy than a traditional web search. Training large scale models requires vast amounts of stable baseload power sustained over extended periods.
AI cannot simply wait for surplus electricity. It competes directly for peak capacity. If energy becomes structurally constrained through 2030, the price of compute will not merely increase. It may become volatile. Marginal prices during peak demand could escalate rapidly.
If disruption occurs in the AI economy, it will not be because models failed. It will be because architecture failed.
The next conceptual shift may be uncomfortable. Not all compute must remain earthbound.
Space based data centers may sound speculative, yet they are increasingly discussed within realistic timeframes. In orbit, solar energy is constant, cooling challenges are reduced, local grid bottlenecks do not exist, and there are no land use conflicts. If launch costs decline substantially, it may become economically viable to position compute infrastructure closer to abundant energy rather than expanding terrestrial grids indefinitely.
When energy is abundant but land and transmission are constrained, infrastructure migrates. History demonstrates this repeatedly.
Europe still has an opportunity, if it chooses a different path than the United States. Fewer mega facilities and more distributed edge compute. Energy and compute designed together rather than separately. Regulation that incentivizes local energy responsibility. Faster permitting for on site generation.
If Europe simply replicates a centralized data center model, it will replicate the bottlenecks, price volatility, and political tensions that accompany it.
AI is not Europe’s greatest challenge. Electricity is.
If Europe intends to compete strategically and geopolitically in the AI era, energy must be treated as foundational architecture. Not as an afterthought. Not as marketing. Not as market theory. As infrastructure design.
Otherwise, we risk pairing advanced models with an energy system incapable of delivering when it matters most.
This is not fundamentally a technology problem. It is a design decision.
