Better experiences. Better-informed decisions.Work with Smarter BI

Software · Research

AI infrastructure research: a guide to the series

Seven analyses on AI infrastructure: workload, evidence, planning, power, delivery, capacity and commercial commitment. A reading route, not a status ladder.

A large megawatt figure is a starting point, not a complete investment story. Understanding an AI infrastructure project means asking what workload it could serve, why its location matters, what has been authorised and what can actually be delivered.

Read the seven analyses in order for the broader argument, or use this guide to find the question behind a particular claim. This is a reading sequence, not a project-status ladder: the dimensions overlap, planning and power can develop in parallel, and commercial commitments can precede physical delivery. Capacity boundaries matter throughout.

1. Workload: start with the competitive question

Can the UK compete for the next wave of AI infrastructure?

Different workloads place different demands on location, connectivity and electricity costs. This opening analysis considers the UK’s position against selected overseas examples and includes an electricity-cost sensitivity tool. It is not a league table of investment returns.

Read for: why the relevant question is not simply which country announces the most megawatts, but which locations can serve which workloads at what cost.

2. Evidence: separate the project claims

Why data-centre project status is harder to verify than headline MW suggests

Four examples show why planning decisions, power arrangements, customer commitments and delivery forecasts need different evidence. A source may establish one narrow fact while leaving other important questions unanswered.

Read for: how to replace a broad status label with a precise statement of what is known, who says it and what remains unverified.

3. Planning: read consent within its scope

Planning is not delivery: reading data-centre consent properly

A resolution, permission, reserved-matters approval and condition approval answer different questions. The article uses two Slough applications to explain why an application’s identity, the relevant phase and the actual decision matter more than a generic portal label.

Read for: why consent is not proof of construction, why an application withdrawal is not necessarily abandonment of a site, and why procedural windows should not be mistaken for completion dates.

4. Power: unpack “power secured”

What does “power secured” actually mean for a data centre?

Manor Farm, Wapseys Wood and Pinewood illustrate three different evidence problems: attributed supply arrangements, alternative routes and a timetable tied to an unverified starting event. The article keeps incoming supply, generation and IT capacity separate.

Read for: the difference between an application, a reported agreement and energisation, and why a future power tranche does not establish operating IT load.

5. Delivery: identify the completed milestone

A completed shell is not an operating data centre

A building shell, customer fit-out, energisation, commissioning and live workloads establish different milestones. Retained Manor Farm and Wapseys Wood evidence shows why the scope of a completion forecast, responsibility for fit-out and phased operation matter. Completion of one building does not establish operation of a whole campus.

Read for: what has actually been completed, what remains for the developer or occupier to do, and why practical completion or a future opening date does not prove operating IT capacity.

6. Capacity: establish what the MW measure

Why data-centre MW figures cannot simply be added together

IT load, incoming supply and generator ratings can all describe the same project without being additive. Manor Farm, Wapseys Wood and Pinewood illustrate different measurement boundaries, phase-versus-total figures and alternative proposals. A voltage rating is not a capacity figure, and unknown capacity is not zero.

Read for: when figures are comparable, when adding them would double-count or mix different measures, and why a partial subtotal is not a market census. The article supplies neither a cross-project total nor a conversion between supply and IT load.

7. Customer / commercial commitment: identify the relationship

What does “fully leased” actually establish for a data centre?

GTR GB One’s reported whole-campus letting and later Phase 2 handover are different disclosures, not a complete contract-to-live-workload history. Manor Farm supplies historical negotiation context; Akzo shows why a site-lease reference is not automatically an end-user capacity commitment. Identify the relationship, source and phase before interpreting the commercial claim.

Read for: how to separate reported commercial positions from inspected agreements and delivery milestones. The checklist asks what is committed, by whom and on what evidence; it does not score readiness or establish customer acceptance or live workloads.

The common thread: match the claim to the evidence

Across the series, the same questions recur: which project or phase, which capacity measure, which source, which date and what exactly was verified? Planning, power, physical delivery and commercial commitments are distinct dimensions, not interchangeable steps on one universal status ladder.

Unknown does not mean absent, zero or adverse. Equally, a forecast is not an observed outcome, and an applicant’s statement does not become independent confirmation merely because it appears in an official document.

For the wider collection, browse all Smarter BI articles.