How much energy can artificial intelligence consume before the electricity system begins to strain? How much water can it draw before the conflict with agriculture becomes impossible to ignore? Who decides where these infrastructures are built, and on the basis of what kind of analysis?

These questions remain open in almost every country where the phenomenon has accelerated. By approving Italy’s first regional law on data centres in May 2026, Lombardy has at least attempted to put them on the table. Yet the paradox common to regulations that chase phenomena already in motion is this: the law exists, but it starts from the assumption that data centres are an indispensable investment. It then builds its rules around that premise without ever questioning the trajectory of growth itself.

---

TAKEAWAYS

---

Why the Lombardy Case Matters

Lombardy is a useful case study precisely because it is an extreme one. According to the Data Center Observatory of Politecnico di Milano, the Milan area alone accounts for 68% of the nominal energy capacity installed in the sector across Italy, with 414 MW IT already active and a target of exceeding 1 GW by 2028. By the end of 2025, electricity connection requests in the region had reached 34.9 GW. If all of them were approved, the footprint of the resulting facilities would be equivalent to the perimeter of the entire city of Bergamo.

The law intervenes on three fronts: land, water and energy. It rewards projects built on abandoned or degraded sites, doubles charges by 100% for those built on agricultural land, and increases them by 200% in green areas. On water, it prohibits withdrawals from public aqueducts and from water intended for drinking use – a far from obvious measure in a plain under growing water stress. It also introduces a regional observatory and an environmental impact assessment procedure for facilities above a certain power threshold.

According to Michela Palestra, regional councillor for Patto Civico, who followed the legislative process from within, the text has an original flaw: “The problem is that the need for these facilities was never really questioned. We do not have an analysis of how many data centres are actually needed, of what kind, or of what size. There is no trace of critical reasoning in this direction.”

The mechanism of increased charges is the most contested point. “When you are dealing with very well-capitalised actors, able to offer even small municipalities millions of euros, an additional economic burden is not enough to shape territorial planning,” Palestra explains. “It is a toll, not a veto. And tolls are not how you govern the concentration of infrastructure in a territory that is already saturated.”

Several elements remain outside the scope of the law: the cumulative assessment of environmental impacts, electrical works and connected infrastructure, and minimum distances from residential areas, despite the fact that noise pollution is a documented issue. The planned regional observatory will collect data only from the moment the law enters into force, not before. “Yet we already know that data centres bring a series of critical issues with them,” Palestra comments. “Starting to collect data in an orderly way only now is inadequate, considering that Lombardy already has sixty-seven data centres in operation.”

There is also a deeper and more structural national problem. “There is no specific land-use category, no dedicated ATECO code recognising the particular nature of these infrastructures. The result is that developers are required to build car parks because that is what the rules for productive facilities require, even though a data centre does not create many jobs and those car parks are not needed.”

The Big Energy Bet

Global figures describe a trajectory that is difficult to keep up with. In 2025, data centres consumed around 415 TWh – 1.5% of global demand – with a 17% increase compared with the previous year, six times the growth rate of total global electricity demand. According to the IEA, by 2030 that figure could double to 945 TWh: three times Italy’s electricity needs. In Italy, data centres could account for between 7% and 13% of national electricity demand by 2035.

Gianluca Ruggeri, president of ènostra, the Italian renewable energy cooperative, does not hide his unease at the range of forecasts currently circulating. “In a single day, I received two scenarios with forecasts ranging from a doubling to a quadrupling by 2030. We are in a situation of genuine forecasting chaos.”

The logic of investment, Ruggeri explains, is structurally oriented towards overcapacity. “Time is the key factor. Whoever manages to provide a certain service before others gains a competitive advantage. But this pressure leads to forecasts constructed as if everyone needed five cars: one for shopping, one for the evening, one for holidays, and so on. Nobody actually has them, but infrastructures are being sized as if they were all necessary at the same time. And even if only 10% of those forecasts materialised, we would still be facing the largest increase in electricity consumption of the last forty years.”

America as a Laboratory of Protest

In the United States, resistance has already produced figures that deserve attention. According to a Gallup survey from May 2026, 71% of Americans oppose the construction of a data centre in their area, compared with 42% nine months earlier. More than half identify data centres as the leading cause of rising electricity bills. In the first quarter of 2026 alone, at least twenty projects were cancelled following local opposition, for a total value of more than 41 billion dollars.

“What is happening in the United States,” Ruggeri observes, “is that a form of grassroots opposition has emerged and is capable of achieving results. In some cases, investors prefer to move the project elsewhere rather than face months of territorial conflict. It is a sign that community pressure works, at least locally.”

Technology Is Advancing, But Towards What?

In Italy and in the rest of the world, opposition to data centres exists but is not yet as well organised as it is in the United States. The tension between citizens and large-scale data warehouses is becoming increasingly visible. If the rules governing coexistence are lagging behind and revealing significant gaps, what role does technological development play, or could it play? Does innovation mitigate growth, or does it feed it?

Liquid cooling – whether direct-to-chip, immersion-based, or through CDU units – is the main response to the increase in power density, which in 2025 was around 120 kW per rack and is expected to reach 600 kW by 2027. It reduces reliance on evaporative water cooling and increases the number of free-cooling hours, lowering consumption per unit of computation. But it does not necessarily reduce total consumption if installed capacity continues to grow.

On waste heat recovery, Palestra argues that the reality is less convincing than it often appears in presentations. “The water temperature in data centres is between 35 and 45 degrees,” she explains. “It is not useful for heating unless it is artificially heated before being fed back into a circuit. Whenever we talk about heat recovery, we should always ask: at what additional energy cost? The net saving is far less significant than it seems in presentations.”

The underlying question remains open: do mitigation technologies reduce absolute impact, or do they enable larger facilities to be built without reducing overall pressure? This is the Jevons paradox applied to digital infrastructure.

Upstream, as Ruggeri notes, there is a choice that no institution is yet making: “AI can be used for early cancer diagnosis or to interpret climate models more effectively. But it can also be used to make funny videos or something else entirely. How do we decide when it makes sense and is worth using? And who gets to decide? At the moment, the market decides. And the market does not distinguish between objectives, values and priorities, except in terms of profit.”

Who Governs, Who Decides, Who Pays

Read within a global context, Lombardy’s law remains a real step in a regulatory vacuum that is still largely unresolved. But the perimeter of that step is narrow compared with the scale of the problem. At European level, a clause included in a 2024 directive allows data centres not to make their energy consumption public. Only 36% of those subject to disclosure obligations have actually published their data.

“In Italy, we have never had real planning for this type of infrastructure,” Ruggeri says. “It happened with shopping centres, which are now closing and leaving behind enormous empty spaces. It happened with logistics. Each time, we started from the idea that the investment was indispensable, and regulation came afterwards. Here, the aggravating factor is that land consumption is accompanied by energy and water consumption on an unprecedented scale.”

There are questions that continue to remain unanswered. The real problem is that, in many cases, they are not even being seriously asked. How many data centres are actually needed? Of what kind? Where, and for what purpose? While regulations and technologies struggle to limit their effects, who is deciding – and who should decide – which applications are really worth that collective cost?