Behind every search, every video call and every answer from an artificial intelligence model there is a physical infrastructure that is almost never seen: the data centre. In Italy it is much talked about — not least after the recent declaration of strategic interest for the €8 billion programmes — but often without an overall picture. Let us try to take one: how many there are, where they are, who runs them and which forces are redrawing the market.
How many there really are
According to the mapping by Data Center Map, at the date of publication Italy has around 250 facilities: about 190 operational, some twenty under construction and more than 30 planned. National operational capacity stands at around 600 MW (figures from the Data Center Observatory of Politecnico di Milano, Osservatorio Data Center) — for comparison, a single-digit fraction of Europe's total capacity.
An honest caveat about the numbers: the surveys differ from one another (what counts as a ‘facility’? is a campus with four buildings one or four?), and the ‘planned’ projects often declare remarkable target capacities — from 50 up to 120-240 MW IT per campus. Behind many of these announcements lie waits for permits and, above all, for connection to the high-voltage electricity grid: as we explained in our analysis of the strategic programmes, over the past three years almost a third of announced investment never materialised. The map, in short, should be read in pencil: what is switched on is certain; what is announced is a hypothesis.
The geography: Milan, then everything else
No surprises about the centre of gravity: Lombardy — and the Milan metropolitan area in particular — concentrates 68% of national capacity (414 MW), on course for the gigawatt by 2028. The reasons are structural — the density of interconnections and Internet exchange points, proximity to European markets, and the (contested) availability of power and land.
But the paradigm is broadening: interest is growing in second-tier sites, closer to industrial districts and local public authorities. The reasons are sound: cutting latency for the uses that genuinely feel it, easing congestion on the grid, and making the national infrastructure more resilient than a system with a single, enormous centre of gravity can ever be.
Who moves the market: three families of operators
Simplifying a crowded landscape, the players fall into three families with different roles:
- ‘Neutral’ colocation — operators that let space, power and connectivity while remaining independent of the carriers: long-standing national players (Aruba, Retelit), European and global platforms (Data4, Digital Realty, Equinix, OVHcloud) and a generation of new entrants targeting precisely those second-tier cities. This is the backbone of the ‘visible’ market.
- The telcos — TIM Enterprise above all, with Wind Tre and, on the enabling connectivity front, Open Fiber: infrastructures born for telecommunications that remain a widespread presence across the country.
- The developers for the hyperscalers — the least-known family and today the most dynamic. The cloud giants (AWS, Google, Microsoft, Oracle) rarely build in their own right: they rely on specialised operators that deliver made-to-measure campuses in the hinterland of Milan and Pavia — names such as CloudHQ, CyrusOne, Stack, Vantage, Virtus. The model is called anchor tenancy: you build only with a multi-year contract already signed by the large tenant.
The economics of digital bricks and mortar
Why anchor tenancy? Because the sums involved leave no room for gambling. Industry estimates put the basic infrastructure investment — building, power, cooling — at between €12 and 14 million per MW IT installed. And for campuses destined for AI there is a multiplier that surprises anyone thinking in property terms: the hardware that goes inside the building (GPUs and accelerators) can cost several times the construction investment — with the aggravating factor that its useful life is measured in 2-3 years, against the decades of the shell.
Taken together, these two facts explain almost all of the market's behaviour: a campus of hundreds of MW ‘fully loaded’ with AI hardware represents a financial commitment in the tens of billions, recurring with every refresh cycle. No investor moves sums like that without watertight contracts — which is why announcements only materialise once an anchor tenant is already in place, and why the gap between ‘planned’ and ‘operational’ remains so wide.
Gigafactory or distributed network? Probably both
The most credible trajectory for Italy is not an either-or but a balancing act:
- on one side, a few large campuses optimised for AI — liquid cooling, high densities, PUE pushed towards unity;
- on the other, distributed and ‘edge’ capacity, federated with the Polo Strategico Nazionale (Italy's national strategic hub for public-sector data): the model that the Public Administration and critical businesses need in order to keep data within a known perimeter, reducing dependence on non-EU players.
It is the same argument as the federation of small and medium-sized providers: sovereignty does not live only in gigawatts; it lives in the structure of the market.
And for a ‘normal’ company, what changes?
Almost none of the above directly concerns most Italian organisations: their workloads are not hyperscale and never will be. For them, the right questions are simpler and more concrete: where is my data? who administers the infrastructure? how efficient is it, and who answers when something stops?
That is the ground we work on every day: a private cloud on dedicated infrastructure in a data centre we know rack by rack, the design of tailor-made data rooms and a maturity assessment method — the DCMM — that does not require you to be a hyperscaler to do things properly. Our zero-emission data centre is proof, at our own scale, that efficiency and control do not depend on size. If you are weighing up where your infrastructure should live, let's talk.
Sources and references: Data Center Map (Italy mapping, July 2026); Data Center Observatory, Politecnico di Milano — School of Management (press releases, January 2026); MilanoFinanza on Observatory data (announced vs delivered investment, 2023-2025). The cost estimates per MW and the AI hardware life-cycle figures reflect current industry values and should be read as orders of magnitude.
