Startups are developing drone protection systems for data centres. The news seems trivial, almost amusing. In reality, it marks a turning point: these buildings have just entered the category of infrastructure you protect like a power plant or an airport.
We've covered data centre constraints from the angles of electricity, water and land. Here's an angle we hadn't anticipated, and it's revealing.
Why this protection is becoming a topic
A data centre was long a discreet building whose location interested no one. Three developments have changed that.
The concentration of value. A site hosting tens of thousands of accelerators represents a multi-billion investment, in a context of component shortages that makes hardware hard to replace quickly.
Functional criticality. As essential services come to rely on these infrastructures, their unavailability stops being a technical incident and becomes a large-scale disruption.
Political visibility. These sites have become symbols, which draws attention they didn't have before. We saw with local opposition in India that they crystallise conflicts.
What this says about the status of these buildings
The real lesson is a change of category. A critical infrastructure is recognisable by three signs: you physically protect it, you plan its redundancy, and the state takes an interest in who controls it.
All three now apply to AI data centres. And the third carries the heaviest consequences: it's exactly what we described in our article on compute sovereignty. A country that regards these sites as critical will want them on its territory, under its law, and out of harm's way.
There's an irony worth noting. We devote a lot of attention to software risks: jailbreaks, agentjacking, models escaping their test environment. Meanwhile, the most basic vulnerability remains physical: a building, a power line, a cooling system. The most sophisticated security in the world doesn't survive a power cut.
What this points to
Three likely developments follow from this shift.
Dedicated regulation. Critical infrastructures are subject to specific obligations around resilience, business continuity and incident reporting. It would be logical for AI data centres to gradually fall under these.
An additional cost. Physical security, redundancy and compliance add to already considerable budgets. This raises the barrier to entry and thus the sector's concentration.
A question of location. If these sites are critical, where they're built becomes a matter of public policy rather than a simple corporate choice.
What to take away
A technology becomes serious when it stops being a software matter and becomes a matter of concrete, fences and contingency plans. It's the same signal we flagged regarding agents moving into production: the boring announcements are the ones that matter.
Ten years ago, computing was thought of as immaterial. We're discovering it rests on buildings that need guarding, electricity that needs generating, water that needs drawing and components that need manufacturing. AI didn't create this reality; it simply made it impossible to ignore.