The queue is the moat.
A hyperscale-grade shell can be built on a predictable construction schedule. Getting utility-scale power to that shell is another matter: in Europe's legacy data-centre hubs — Frankfurt, London, Amsterdam, Paris, Dublin — connecting a data centre to the grid takes an average of 7–10 years (Ember, 2025), and several municipalities have imposed outright moratoria. Meanwhile global data-centre electricity demand is set to more than double by 2030, to around 945 TWh (IEA, 2025), so every new campus asks the grid for more than the last one did.
The result is an inversion of the industry's economics. For two decades the scarce asset was location — proximity to fibre, to exchanges, to users. Today the scarce asset is a signed, dated interconnection agreement. Buildings are replicable. Queue positions are not.
Power-zone economics.
The second shift is where AI can be built at all. Frontier training runs are latency-tolerant: a model does not care whether its gradients converge 40 milliseconds from Frankfurt. That releases AI infrastructure from the metro locations that defined the industry and lets it move to where power is abundant, affordable, and connectable — industrial zones with headroom, regions with surplus generation, nodes where the grid wants large, steady loads.
Because energy is the single largest operating cost line of an AI campus over its life, the spread between a congested metro node and a well-chosen power zone compounds into a structural cost advantage for the tenant — one that no amount of clever building design can recover.
Time-to-power is the real currency of AI infrastructure. Everything else can be bought on a schedule.
What pre-secured power changes.
This is why Castellan starts from the power side. Drawing on a decade of operating experience at the Bitkern Group, we have access to more than 5 GW of power through our network, available as demand requires, alongside established energy-supply relationships. With power secured before the first tenant conversation, the sequence flips: siting follows the megawatts, long-lead electrical equipment can be ordered early, and a lease can be signed against a date a tenant can plan around.
In practice, the path from signed lease to power takes up to 24 months — against years of grid-connection queues in Europe's legacy hubs.
Implications.
For tenants, the lesson is to scrutinise the power position, not the rendering: a beautiful campus behind a multi-year queue is a multi-year problem. When choosing an infrastructure partner, look at the layers beneath the building — land, interconnection and energy relationships — because those layers, not the shells on top of them, decide whether your capacity actually arrives on schedule.