Capacity strategy under a 4 MW utility cap
A 2.0 MW IT load data centre is at 62% of its 3.2 MW electrical design capacity; actual PUE is 1.58. Demand has grown 15% per year for three years, and sales forecasts the same for five more years. The utility will not approve more than 4.0 MW site demand before 2032; the 4.5 MW transformer is derated to 4.0 MW by a planning condition. Expanding the existing hall to 3.2 MW costs €6.8m and takes 14 months. A nearby colocation option costs €135/kW/month with six-month lead time. A new edge site for latency-sensitive customers costs €2.4m for 400 kW and 18 months. The board has a 2030 carbon reduction target of 50% and wants an 'AI-ready' 20 kW/rack zone. Finance uses a seven-year horizon and a 9% discount rate. Operations is already short of two shift engineers.
1. Quantify the capacity gap and evaluate the three options. Show the demand calculation and the main constraints.
Read the worked answer
At 15% CAGR, 2.0 MW becomes 2.0×1.15^5=4.02 MW in five years and 8.1 MW in seven years; the 4.0 MW utility cap is hit around year five. At PUE 1.58, 4.0 MW IT would need 6.32 MW facility demand, already above the cap, so the cap binds earlier-roughly 2.53 MW IT at current PUE. Efficiency first: if PUE falls to 1.35, the same 4.0 MW site supports about 2.96 MW IT, buying about one year. Expanding to 3.2 MW IT costs €6.8m and takes 14 months, but it exceeds the utility cap at current PUE and does not solve the 2032 limit. Colocation at €135/kW/month for 1 MW is €1.62m/year, flexible and fast, but has no carbon benefit and may not meet the AI density. Edge at €2.4m for 400 kW is €6,000/kW capex and only justified for latency revenue. No single option meets growth, carbon and latency; the board needs a portfolio.
2. Recommend a phased strategy to 2030, including the AI-ready zone, and state what you would not build.
Read the worked answer
Phase 1 (0-12 months): cut PUE from 1.58 to ≤1.4 through containment, free cooling and setpoint discipline; this releases about 0.3 MW IT within the existing cap. Phase 2 (12-24 months): build a 400 kW AI-ready zone at 20 kW/rack inside the existing hall using close-coupled cooling; do not build it at 20 kW/rack across the whole hall. Phase 3 (24-48 months): contract 1-1.5 MW of colocation for general compute rather than expand beyond 4.0 MW; use the existing hall for high-value, latency-sensitive load. Phase 4: build the 400 kW edge site only if a customer contract covers at least 60% of its capex. Do not build a 3.2 MW expansion that cannot be powered or cooled within the planning condition. Do not chase 8 MW on-site; the utility cap makes it undeliverable before 2032. Recycle stranded UPS and space with metered chargeback.
3. Define the governance, KPIs and triggers that would make this plan revisable.
Read the worked answer
Governance: a quarterly capacity board with sales, finance, operations and sustainability; one owner for the utility constraint; stage gates at PUE, contract backlog and carbon. KPIs: IT MW, MW contracted vs installed, PUE, carbon intensity (kgCO2e/kWh), utilisation by rack density, colocation cost per kW-month, edge contract coverage, and engineering vacancies. Triggers: if PUE does not reach 1.45 by month nine, pause the AI zone; if colocation price exceeds €150/kW/month or lead time exceeds nine months, re-open the on-site expansion with a utility exception; if AI backlog exceeds 300 kW, accelerate the close-coupled zone; if carbon trajectory is missed, shift general compute to lower-carbon colocation. Finance: use a seven-year NPV with a 9% discount rate, but include a carbon shadow price and the cost of stranded UPS. People: do not approve the plan without two additional shift engineers or a managed-service contract; the current team cannot operate a more complex site.