Data Centre Costs in France, Spain and Italy Need a Comparable History

A country cost comparison can change an investment shortlist. It can also mislead if the underlying definitions change between observations.

Our review of France, Spain and Italy points to two conclusions. Electricity deserves an explicit sensitivity analysis alongside construction expenditure. Historical comparability deserves the same attention as the latest available price.

For a hypothetical 20 MW IT facility, applying the first-half 2026 electricity benchmarks below produces an annualised difference of approximately €9.73 million between Italy and France. This is a useful screening signal. Turning it into a site decision requires contracted prices, connection evidence and a consistent development scope.

The electricity comparison

Eurostat's non-household electricity dataset provides a common statistical basis. We use consumption band IG, covering customers consuming at least 150,000 MWh annually, and the X_VAT tax basis, excluding VAT and other recoverable taxes and levies.

Source: Eurostat, electricity prices for non-household consumers, nrg_pc_205. Values checked on 26 October 2026 and converted from EUR/kWh to EUR/MWh.

France has the lowest benchmark of these three countries in this period. Spain's gap to France has narrowed compared with the same semester a year earlier. Italy remains materially higher.

These are national statistical prices for a defined consumption cohort. They are not data-centre supply offers, wholesale spot prices or forecasts. The movements alone do not establish their causes.

What the gap means for a common facility

Consider a facility with 20 MW of installed IT capacity, an average electrical IT load of 70%, an annual PUE of 1.30 and 8,760 operating hours.

Annual facility consumption is therefore:

20 × 70% × 1.30 × 8,760 = 159,432 MWh.

The result falls within the selected Eurostat consumption band. Applying each H1 2026 price to that same theoretical annual consumption gives the following comparison.

The Spain–France difference is approximately €0.88 million per year. The Italy–France difference is approximately €9.73 million.

This exercise annualises a semester price. It does not estimate an actual 2026 bill or predict future expenditure. The 70% assumption describes electrical IT load, rather than commercial occupancy. Annual PUE also cannot determine peak import capacity or the contracted power requirement.

A project model must replace these assumptions with its expected load profile, cooling performance, procurement terms and commissioning ramp. When using an invoiced-price benchmark, it must also avoid adding network charges or taxes already included in that benchmark.

CAPEX requires an equally explicit boundary

A construction cost per megawatt is useful only when the denominator and inclusions are clear. IT capacity, utility import capacity and building design capacity describe different things.

The comparison should identify whether a figure includes land, grid connection, external works, professional fees, contingency, active IT equipment and financing. It should also retain the pricing date, currency, redundancy architecture and delivery scope.

The Gleeds Global Data Centre Market Report published in 2025 illustrates why benchmark methodology matters. Its city construction benchmarks have a defined cost basis and exclusions. They should not be treated as complete investment budgets or silently relabelled as costs per IT MW.

For AI infrastructure, the equipment boundary becomes particularly consequential. GPU systems, networking, storage and associated maintenance need their own assumptions. A change in hardware generation or equipment configuration can alter the budget without representing construction inflation.

My approach is to keep observed construction evidence, indexed estimates and forecasts distinguishable throughout the model. An estimated historical value should remain identifiable as an estimate after newer observations arrive.

History must preserve the meaning of each observation

A reliable time series requires more than keeping old numbers.

For electricity, switching from a smaller consumption band to IG can create an apparent price fall that partly reflects a different customer population. Changing tax treatment, currency or geography creates similar problems. Such observations belong in separate series unless an explicit, documented adjustment makes them comparable.

Each observation should retain its country or market, cost category, unit, scope, consumption band where applicable, tax treatment and source. It also needs two distinct time references.

The first describes the economic period or effective date. The second records when the observation became known to the model. The source publication date adds further context.

This distinction makes two useful questions answerable. What is the best revised estimate for a past period? What information was available when an investment decision was taken?

A correction should therefore create a linked revision with a reason and timestamp. A new period should add an observation. Rechecking an unchanged source should record verification without manufacturing another price observation. A failed refresh should preserve the last verified value and make its age visible.

These are methodological requirements for maintaining the research. They should not be read as an announcement that a new historical interface has been deployed.

How I would use the comparison

France's electricity benchmark supports a closer examination of its operating-cost position. Spain's improvement warrants testing whether comparable terms are achievable for the intended load profile. Italy's higher benchmark makes procurement strategy and the project's revenue proposition especially important.

None of these findings settles the location decision. Available power on the required date, connection expenditure, permitting, customer demand, latency and delivery risk remain separate inputs.

The practical next step is to compare specific sites using the same facility assumptions, then replace national proxies with evidenced project terms. Preserve the original assumptions when doing so. That allows an investment committee to see whether a changed conclusion comes from the market, the project design or a correction to the evidence.

A useful cost database should make that distinction visible every time a number changes.

Sources and calculation basis

Analysis dated 26 October 2026. Electricity series use Eurostat dataset nrg_pc_205, band MWH_GE150000, currency EUR and tax basis X_VAT. Direct dataset extracts are available for France, Spain and Italy.

Year-on-year changes compare H1 2026 with H1 2025. Annualised electricity costs and differences are PowerlandMap calculations using the illustrative assumptions stated above. They exclude other operating costs and do not constitute a complete total-cost-of-ownership ranking.