The infrastructure that powers AI agents isn't just energy-hungry anymore — it's making enemies.
The Summary
- Police Scotland warned that a proposed AI datacentre near Edinburgh will require "robust security measures" to prevent "incursion" due to expected public opposition
- The warning, submitted during a planning consultation for a facility in Larbert (30 miles west of Edinburgh), signals escalating resistance to AI infrastructure beyond typical NIMBY complaints
- Growing pattern: AI datacentres are becoming political rallying points in the US and triggering security concerns internationally
The Signal
When police start writing letters about "incursion" during planning consultations, you're not dealing with normal development opposition. Police Scotland's warning about the Larbert datacentre marks a new phase in AI infrastructure deployment: physical resistance substantial enough to require preemptive security planning.
This isn't about noise complaints or traffic patterns. The language is operational, specific, and speaks to genuine physical threat assessment. Police don't typically submit formal security warnings for office parks or warehouses.
"The infrastructure costs of AI include security budgets previously reserved for critical national infrastructure."
The Larbert site sits 30 miles west of Edinburgh, close enough to urban populations to matter but far enough out that someone thought it would avoid scrutiny. That calculus is breaking down across Scotland and beyond. The pattern Police Scotland is responding to includes:
- Mounting local opposition to AI infrastructure worldwide
- AI datacentres becoming major political organizing points in the US
- Growing sophistication of resistance from single-issue activism to coordinated campaigns
Here's what makes this different from past tech infrastructure fights. Datacentres for cloud storage or enterprise computing drew complaints, sure. But AI training and inference facilities carry different symbolic weight. They're visible manifestations of automation anxiety, energy consumption debates, and questions about who profits from AI deployment while communities bear costs.
The energy demands are real and quantifiable. Large language model training runs can consume megawatts for weeks. Inference at scale means constant load. Local grids strain. Energy prices potentially tick up. Meanwhile, the jobs these facilities create are primarily construction (temporary) and facilities management (minimal). The AI work happens elsewhere. The economic bargain that made datacentres tolerable starts looking worse when the application is explicitly about automating human work.
The Implication
Companies building AI infrastructure need to add security and community relations to their deployment budgets, not as afterthoughts but as core planning elements. The days of quietly dropping compute facilities into semi-rural locations are ending. Expect longer timelines, higher costs, and the need for genuine community engagement beyond token consultation.
For regions trying to attract AI investment, this creates a new competitive dimension beyond tax incentives and power availability. Can you actually build and operate the facility without it becoming a flashpoint? That's now part of the pitch.
Watch for developers shifting strategy toward existing industrial sites with established security, more remote locations with sparse populations, or purpose-built AI zones with pre-negotiated community agreements. The easy sites are gone.