Tools

Data-centre project

Take a colocation site from brief to a Stage 1 report: accommodation, power and cooling, cost plan, programme, operating plan and business case, with every default drawn from a named, dated source.

Design, cost plan and programme: method and sources

One project runs through every step: the brief sets the IT load and rack densities the customers need; the design sizes the building and plant around them; the cost plan, programme and operating plan follow from the design; the appraisal reads all three. Customers' IT equipment is outside the scope.

  • Space. White space per rack, electrical and mechanical rooms follow the Schneider Electric TradeOff capital cost model (2026): its fitted functions of capacity, density and redundancy, which reproduce its own default case to the square foot. Its electrical figure covers UPS and switch rooms with the transformer and generator yards. Ancillary space (10% of technical space) and the plot ratio (0.4 for one storey) are ours.
  • Power. Grid connection = IT capacity × design-day PUE ÷ 0.95 power factor; generators on the facility peak, UPS on the IT load, transformers on the connection, each multiplied by the redundancy chosen. Unit ratings are ours.
  • PUE. An estimate: 1 + electrical losses (by UPS topology) + fans and pumps + compressor energy + 0.02. Compressor energy is the cooling load × the share of the year above the free-cooling limit (from the city's ERA5 hours, 2016-2025) ÷ a COP of 4. A designer's figure should replace it.
  • Cost plan. The level is the market's survey cost per MW for construction. The Schneider unit costs give the split by element and the effect of the design against a reference design. Its V10.1 sheet (September 2026) prices most rows in 2017 dollars and raises them all by one 32% adder; here each 2017 row is brought to 2025 by the US producer-price index of its own equipment (switchgear and transformers × 1.90, busway × 1.86, chillers × 1.74, CRAHs × 1.70, generators × 1.40, UPS × 1.42, labour by construction wages × 1.36; BLS via FRED); core and shell at Schneider's own 2025 figure of $300 per ft² (the US industrial-building index, × 1.66, overstates Europe), while rows priced in 2025 (liquid cooling, containment, HV switchgear, lithium-ion) stay. Materials are localised with Schneider's country factors; labour with Eurostat's 2024 construction labour cost per hour, indexed on Germany. A liquid-cooled share adds the premium builders measure for a fully liquid build (Turner & Townsend 7-10%, JLL 10%) in proportion. The reference design is N+1 power and cooling, 2N distribution, air-cooled chillers with free cooling, room CRAHs, 8 kW racks: cost = survey × MW × model(design) ÷ model(reference). Core and shell cost 15% more per m² for each storey above the first (ours). Fees, contingency (10% for a new site, 5% for an expansion), land and the grid connection (the network operator's published charge per MVA, where one is published, plus €0.2 M per MVA for the developer's own substation, ours; 20% paid on application, the rest with land when the first hall starts) are added outside. The shell, HV intake, half the heat-rejection plant and 5 points of site works, from the design's own cost lines, are built with the first hall; later phases fit out halls. The estimate is AACE Class 5: an order of magnitude.
  • Programme. Pre-application, then permits (the longest stated permit step for the country, or a stated total; six months of consent and no pre-application on an existing campus); long-lead equipment ordered at permit, or reserved before it; the grid wait for the market. The latest of construction, equipment plus six months of installation, and the grid sets the first hall's date, plus three months of commissioning. Construction is timed to finish when power arrives.
  • Operating plan. Engineers on shift around the clock (three posts up to 20 MW, one more per further 20 MW; one post is 8,760 ÷ 1,720 = 5.1 FTE) and day roles; contract guarding; maintenance, insurance and lifecycle reserve as shares of construction cost.
Climate and programme defaults by market
MarketHours above 20 °C a year1% design dry-bulb, °CPermits, monthsGrid wait, years
Amsterdam79526.71210
Athens364234.9124
Berlin59124.874
Copenhagen59124.8288.8
Dublin11020.5279.0
Frankfurt139330.579
Helsinki55324.7124
Lisbon163426.5123.5
London68425.8308.0
Madrid265836.2122.9
Milan265032.9132.4
Oslo43524.3127.5
Paris130129.8128.5
Stockholm51323.8124
Vienna129729.1124
Warsaw129729.194
Zurich112828.5126.5

Berlin and Vienna take the climate of the nearest city with ERA5 hours in the dataset (Copenhagen and Warsaw). Grid waits are the stated waits for a new connection; where a market has none, JLL's four years for primary markets applies. Operators who have already secured capacity can say so and the grid leaves the critical path.

Method, defaults and sources

The model runs year by year from the start of construction to the exit, in nominal euros, before debt and tax. It is the same arithmetic a lender's base case starts from, set out so that every line can be checked.

  • Construction = cost MW × cost per MW × (1 + contingency + fees), spent evenly over the build months at today's prices raised by inflation. The first phase builds the shell and site plant for every hall (35% of construction by default), later phases fit out halls only. A hall opens the month after it is finished, so built and contracted MW are averages over the year
  • Contracted MW = min(built × stabilised utilisation, last year × (1 − churn) + new contracts + pre-sales of halls opening): retail halls never sell the last kW
  • Revenue = contracted kW × capacity price × 12 × indexation + cabinets × interconnection per cabinet × 12 × indexation, with cabinets = contracted kW ÷ kW per cabinet from the design, plus customers' electricity when it is billed at cost
  • Electricity = contracted kW × average load × PUE × 8,760 h × price, plus a 5% idle draw on unsold capacity
  • Staff = FTE at 10 MW × (MW ÷ 10)scaling × cost per FTE; maintenance and insurance as a share of construction cost; property tax per MW; overheads (sales, network, customer operations, head office) as a share of capacity and interconnection revenue
  • Cash flow = EBITDA − lifecycle reserve − capex; IRR and NPV on these flows plus the exit
  • Exit = next year's (EBITDA − lifecycle reserve) ÷ cap rate, or next year's EBITDA × multiple, less sale costs
  • Yield on cost = stabilised NOI ÷ all-in capex, both at today's prices, stabilised meaning every built MW sold to the stabilised utilisation; the exit is also shown per MW in today's money and compared with European deals
  • For an existing site, the headline is the expansion alone: the plan's cash flows less those of the site left as it is. The whole site, with its book value as the year-0 investment, is shown beside it
  • Where a market has no published price, the price is an estimate: Digital Realty's EMEA new-lease rents scaled by its own in-place rent in that metro against its European median (metros with 20 MW or more, 60% or more let); otherwise, and for the 250–500 kW band, the European figure scaled by the market's construction cost against the FLAP-D average

How defaults are chosen. Each default is the median of the latest edition of every publisher that covers the market, with a range counted at its midpoint. A market with no figure of its own takes the European benchmark, and the chip under the field says so. Figures in dollars, pounds or other currencies are converted at the ECB annual average for their year. Rows from aggregator sites that contradict themselves are kept in the dataset but never used as a default.

Defaults by market
MarketConstruction, €M/MWPrice over 1 MW, €/kW-moPrice 250–500 kW, €/kW-moElectricity, €/MWhCost per FTE, €kExit yield, %
Amsterdam10.0145146*1641016.00
Athens6.7111146*1561016.00
Berlin10.5111146*2161136.00
Copenhagen10.3111146*1761396.00
Dublin9.7145146*2021146.00
Frankfurt10.91182212161136.00
Helsinki11.4111146*59876.00
Lisbon9.0133146*128696.00
London11.1149175941585.50
Madrid9.2126146*126966.00
Milan9.2111146*172786.00
Oslo11.0111146*571296.00
Paris10.0145146*921046.00
Stockholm10.9111146*89986.00
Vienna10.11081081881236.00
Warsaw8.3111146*233476.00
Zurich12.4111146*981016.00

* No public price for this market: an estimate, scaled as described above. Prices are new-lease and asking rents with power billed separately; in-place rents on older contracts are lower.

European operating-asset deals since 2023, the exit check
DealDate€M per MWBasis
Redcentric Data Centres (8 UK data centres)2025-103.641 MW across eight operating sites
AtlasEdge: eight small European edge data centres (carve-out)2025-0915.0current MW (a sale process, not a closed price)
Data4 StableCo (stabilised portfolio, c.244 MW)2025-0815.2stabilised portfolio
Frankfurt data centre (Digital Realty, additional 15.1% interest)2024-1213.8one operating 34 MW facility
Nabiax2024-1128.635 MW installed
Vantage EMEA investment partnership (six stabilised European data centres)2023-1014.1177 stabilised MW

Whole-platform deals divided by secured or planned power, development partnerships and deals reported by a single aggregator are left out: value per MW only compares on operating capacity. The exit check uses the median and the range with the cheapest and dearest deal left out.

Limits. No debt, tax, depreciation or working capital. Survey construction costs exclude land, grid connection and tenant IT fit-out. Staffing and maintenance defaults are benchmarks for a multi-tenant site, not a quote. The tool is a screen for a business case, not an investment recommendation.

Sources

The underlying rows, each with its source, date and the quoted passage, are kept in ScienceShot's economics dataset (data/reference/2026-10-dc-economics).