The sustainability debate around artificial intelligence is often framed in broad terms: AI may help optimise energy systems, forecast demand, identify methane leaks and improve industrial efficiency. At the same time, it requires rapidly expanding physical infrastructure—data centres, network capacity, cooling systems, electricity connections, water and construction materials.
The European Commission is now making that infrastructure more visible.
Its latest data-centre sustainability initiative is designed to improve transparency and encourage facilities to become more energy efficient, more flexible and more responsive to the wider electricity system. The direction is clear: data centres should not be assessed only by how much electricity they consume. They should be assessed by whether they can operate as responsible participants in a constrained energy, water and land system.
This is a major change in emphasis.
For years, data-centre sustainability has focused heavily on power-usage effectiveness, renewable-energy procurement and headline carbon claims. Those measures remain important. But they are no longer enough.
A data centre can have a strong annual renewable-energy narrative while still adding pressure to a local grid at the wrong time, competing for water in a stressed catchment, requiring carbon-intensive construction or relying on backup generation that creates local pollution and emissions.
The next standard is therefore more demanding: can the facility demonstrate that it supports, rather than strains, the system around it?
The shift from efficiency to system value
Data centres are becoming a larger part of electricity demand as cloud computing, AI training, AI inference, digital services and data-heavy business applications grow. This does not make them inherently unsustainable. But it does make their location, design and operating model economically and environmentally material.
A responsible data-centre strategy needs to answer at least five questions.
First, where is the power coming from?
Annual renewable-energy matching is useful, but local and time-sensitive demand matters too. A facility that operates during periods of system stress may increase reliance on marginal fossil generation, even if its operator purchases renewable certificates on an annual basis.
Second, can demand be flexible?
The Commission’s initiative places increased attention on facilities that can adjust electricity use to grid conditions. This could include shifting non-urgent computing workloads, using thermal storage, coordinating backup systems or aligning certain processes with periods of abundant low-carbon generation.
Third, what is the water impact?
Cooling requirements can make water a significant local issue. The question is not only how much water a facility uses in aggregate. It is whether the water source is appropriate for the local catchment, whether the facility is located in a water-stressed area, and whether alternative cooling or reuse options have been evaluated.
Fourth, what is the embodied-carbon impact?
New data centres require land, buildings, steel, concrete, electrical equipment, cooling systems and network infrastructure. Their operational carbon performance cannot be separated from the carbon embedded in construction and equipment supply chains.
Fifth, who benefits and who bears the cost?
Local communities may experience pressure on electricity networks, water resources, land use and infrastructure. They should also see credible economic, employment and system benefits. Transparency matters because the social licence for data-centre expansion will depend on whether the wider public can see a fair allocation of costs and benefits.
Why a European rating system matters
An EU-level rating framework can make a complicated set of technical decisions more visible to customers, investors, regulators and communities.
The value of such a framework is not that it creates a single perfect score. No single rating can capture every sustainability issue. The value is that it can establish a common language for discussing performance and comparing facilities.
A credible approach should include indicators for:
Energy efficiency.
Renewable and low-carbon electricity sourcing.
Carbon intensity.
Water use and water-stress exposure.
Waste heat recovery.
Demand flexibility.
Embodied carbon.
Equipment life cycles and circularity.
Local infrastructure impacts.
For corporate buyers of cloud and AI services, this will increasingly become a procurement issue. Businesses purchasing digital services need to ask not only whether a provider has a net-zero target, but how its underlying infrastructure performs in the regions where workloads are actually processed.
For investors, it becomes a resilience and capital-allocation question. Facilities that are poorly located, inflexible or dependent on constrained local resources may face greater regulatory risk, grid-connection delays, higher operating costs and community opposition.
AI makes the local test unavoidable
AI creates a particular challenge because demand can grow quickly and unpredictably. The economic value of AI services may be global, but the infrastructure impacts are local.
A model trained in one region may require large volumes of electricity and water there, while the commercial benefits are captured elsewhere. This makes it essential that data-centre sustainability claims are grounded in local evidence.
The key questions are practical:
Is the facility located where low-carbon power can be delivered without worsening local grid constraints?
Can it reduce demand during peak periods?
Does it rely on potable water in a water-stressed region?
Can waste heat be used by homes, businesses or public infrastructure?
Are new grid investments serving communities as well as the data centre?
Has the facility assessed its whole-life carbon impact?
The best projects will be able to answer these questions transparently.
What businesses should do now
Companies do not need to own a data centre to be exposed to this issue. Any organisation increasing its use of cloud computing, AI tools, data analytics or digital platforms should begin to understand its digital footprint.
Five actions are particularly useful:
Map your major digital suppliers.
Identify which cloud, software, hosting and AI providers support the most material business functions.Ask for location-specific information.
Annual corporate carbon targets are not enough. Ask how providers manage electricity, water, demand flexibility and infrastructure impact in the regions used by your workloads.Include digital services in procurement standards.
Add sustainability and transparency questions to technology procurement, renewal and vendor-management processes.Use AI where it creates measurable value.
Prioritise applications that reduce waste, improve forecasting, support efficiency or strengthen climate-risk decisions—not technology adoption for its own sake.Treat digital growth as an infrastructure decision.
As AI use expands, it should be considered alongside energy, carbon, water, data governance and financial planning.
The real test
Europe’s new data-centre framework signals that digital infrastructure is entering the same accountability era as buildings, transport and industrial assets.
The question is no longer whether the digital economy needs more infrastructure. It clearly does.
The question is whether that infrastructure can be designed as a grid asset, a water-responsible neighbour and a transparent contributor to the energy transition.
That is the sustainability test for AI.