What Happened

Dutch caretaker minister for Spatial Planning Mona Keijzer told parliament that she cannot intervene to stop a large new data center in Amsterdam that will be fully used by Microsoft, because the permit was granted about five years ago. (NOS)

The renewed political attention followed reporting that the Amsterdam facility will be dedicated to Microsoft. Members of parliament criticized the project in the context of scarce grid capacity, arguing that local housing and businesses face delays while a large compute project appears able to proceed.

The case also highlights a technical and legal nuance. The facility has a total grid connection of 78 megawatts, but it is designed as three towers about 85 meters tall, allowing it to fit on roughly 23,000 square meters of land. That footprint is well below the national hyperscale threshold tied to site area.

ALSO READ: Lelystad and the Energy Cost of AI Compute

Structural Context

This debate sits at the intersection of three realities.

First, AI and cloud demand increasingly expresses itself as power demand. Data centers scale in megawatts, and the constraint is more often grid connection and delivery timing than capital availability. When grid capacity is tight, allocation becomes politically visible because it forces tradeoffs across housing, industry, and digital infrastructure.

Second, policy definitions often reflect older facility archetypes. The national hyperscale restriction introduced in 2022 focused on centers with both a large electricity connection of at least 70 megawatts and a very large surface area of at least 100,000 square meters. That model maps well to low rise campus layouts. It maps less well to high rise, compact designs that compress power density into a smaller footprint while still drawing large loads.

Third, permitting timelines and legal certainty matter in infrastructure. When permits are granted before a policy shift, the state is constrained in how it can respond without undermining the predictability that all long lived infrastructure depends on. The political debate therefore tends to move from the individual project to the rulebook itself, including the question raised in parliament of whether the definition should shift from an and test to an or test, so that either land footprint or power connection could trigger national restrictions. (NOS)

The Enki Perspective

The Amsterdam case is a useful example of why energy alignment is becoming a first order design principle for AI infrastructure.

The public discussion focuses on who gets scarce grid capacity. That is understandable. But it also reveals a deeper point: the fastest path to more compute is not always to compete for constrained nodes in already congested urban grids. When compute demand is accelerating faster than transmission upgrades and substation buildouts, the system naturally pushes developers toward creative workarounds, such as denser vertical designs, while leaving the underlying power bottleneck unresolved.

Enki’s lens starts with the premise that the bottleneck is structural and persistent. If policy aims to manage where and how large facilities can be built, it must incorporate power intensity as a primary variable, not only land use. A compact footprint does not reduce grid impact if the power connection remains large. In that sense, the Amsterdam design is not an anomaly. It is a preview of a broader market response to land constraints, permitting friction, and the need to deploy quickly.

This is where energy aligned deployment becomes strategically important for Europe. Stranded and curtailed renewable energy represents underutilized economic capacity. When infrastructure is deployed closer to generation, it can reduce dependence on congested urban nodes and shorten time to power relative to projects that require major transmission reinforcement. It also supports a more distributed compute map, which is one ingredient in practical digital sovereignty. Sovereignty is not only about software choices or cloud contracts. It is also about where the electrons and compute capacity actually sit, and how resilient those supply chains are under stress.

From a finance perspective, the lesson is discipline and repeatability. Institutional capital favors models that can be replicated across sites with clear permitting pathways and predictable power delivery. If definitions and national rules lag behind modern facility forms, the market will keep encountering stop start cycles and localized political conflict. Energy aligned infrastructure helps shift the conversation from scarcity allocation to capacity creation.

What This Signals?

The first signal is that power based thresholds will become more central in regulation. The parliamentary question of moving to an or test is not only a legal tweak. It reflects an emerging consensus that electricity connection is the most accurate proxy for system impact. (NOS)

The second signal is that high density data center forms will proliferate. When land is expensive and scarce, verticalization is a rational response. Policymakers who want to steer outcomes will need tools that address power, heat, water, and grid services, rather than relying primarily on surface area.

The third signal is that grid capacity allocation will remain politically salient. When housing targets and industrial electrification collide with compute expansion, the debate will be framed as competition for a fixed pie. The more durable solution is to expand and diversify supply, including by monetizing curtailed renewables and locating modular compute where power is available.

ALSO READ: Public Acceptance as Europe's Next Compute Constraint

Finally, this episode reinforces a strategic European opportunity. If AI infrastructure is treated as critical infrastructure, Europe benefits from accelerating build models that match generation realities and avoid over dependence on a few constrained metropolitan nodes. That is consistent with a future where compute growth is inevitable and positive, and where the main task is to build the energy system and infrastructure interfaces that can sustain it.

Source: author not listed, 27 January 2026
https://nos.nl/artikel/2599925-keijzer-kan-niks-doen-tegen-enorm-datacenter-voor-microsoft-in-amsterdam