Data Centres
The data-centre sector is in the middle of one of the largest private infrastructure investment cycles in modern history. Hyperscalers and their investors are committing capital on a scale previously associated with national grid or motorway programmes: individual campuses now carry price tags in the billions of dollars, and industry-wide capital expenditure by the largest cloud providers is running at tens of billions of dollars per quarter, much of it driven by demand for AI training and inference capacity. Wherever capital moves that fast, delivery risk follows.
The risks are concrete and familiar to anyone who has delivered large construction or engineering programmes. Power is the binding constraint in most major markets: grid connection queues in Northern Virginia, Dublin, Frankfurt and parts of Southeast Asia now stretch for years, and several jurisdictions have imposed moratoria or tighter planning conditions. Construction capacity — labour, switchgear, generators, transformers — is under strain, with long-lead equipment times lengthening materially since 2021. Planning and community opposition have blocked or delayed projects over water use, noise and land. And the demand side carries its own uncertainty: if AI workloads or efficiency gains shift faster than expected, some capacity being financed today could be stranded or repriced.
This category investigates the delivery side of the data-centre boom. PIA examines how individual projects and portfolios actually perform against schedule and budget, what the contractual structures — pre-leasing, powered shell, build-to-suit — do to risk allocation, how power procurement and grid constraints are reshaping site selection, and where the financing assumptions look most exposed. We draw on planning records, regulator and grid-operator data, company filings and established industry and business-press reporting. The sector's promotional language is loud; our interest is the quieter record of what gets built, when, at what cost, and for whom the risk is carried. For investors, boards and programme leaders, the question is not whether data centres will be built — they will — but which projects will be delivered on time, on budget and into real demand.
Data Centres — Frequently Asked Questions
Primarily cloud computing growth and, since 2023, AI workloads, which require far denser compute and power per site than traditional enterprise demand. The largest cloud providers have committed capital expenditure programmes measured in the tens of billions of dollars annually.
Power availability. Grid connection lead times, transformer and switchgear supply constraints, and local planning restrictions are the most frequently cited causes of delay in major markets.
It is a live risk. If model efficiency improves faster than capacity is absorbed, or if demand concentrates in fewer sites, some speculatively financed capacity could be underutilised. PIA treats this as an open question and reports the evidence on both sides.
Get the Next Investigation First
Receive forensic analyses of billion-dollar failures every Monday. No fluff. Just lessons.
Subscribe to The Weekly Brief