Grid queues, not capital, are now the binding constraint on data centres

Capital is not what is slowing the data centre build-out. Electricity is. Connection queues in several major markets now stretch past five years for large loads, and the queue position, rather than the balance sheet, determines who builds and when. Operators with legacy grid rights are trading them at valuations that would have looked absurd three years ago.

The consequence for buyers is a widening regional spread in what compute costs. Capacity in power-constrained markets carries a premium that is not going to compete away, because the constraint is physical and the remedy is transmission infrastructure on a decade timescale. Meanwhile regions with surplus generation — the Nordics, parts of the American Midwest, increasingly the Gulf — are pricing aggressively to attract load.

Finance teams encounter this indirectly, through cloud bills and through vendor pricing that quietly varies by region. Most enterprises set a data residency default years ago and have never revisited it. For genuinely latency-insensitive workloads — batch reconciliation, document processing, overnight model runs — the residency requirement is often a policy preference rather than a legal obligation, and revisiting it is one of the few AI cost levers available without touching the model.

For anyone running a tender, the useful question has changed. Floor space and rack density are no longer the scarce inputs. Ask for contracted power availability dates and the evidence behind them. A provider who cannot answer that precisely is selling you a queue position they may not hold.