AI Data Centre Optimisation May Be Missing a Critical Sustainability Metric: Water

Dubai, UAE: As artificial intelligence drives an unprecedented expansion in data-centre capacity, a new research review is raising questions about whether current approaches to optimising data-centre energy consumption are sufficiently accounting for the wider environmental footprint of AI infrastructure.

The study, “Artificial Intelligence for Energy Optimization in Data Centers,” reviewed roughly 194 research papers, with 63 studies subjected to detailed coding. Its authors found a significant gap in how AI-driven data-centre optimisation is currently assessed: none of the 28 primary control-oriented studies examined accounted for water withdrawal, while none accounted for embodied carbon.

The findings are particularly relevant as data centres become increasingly energy- and cooling-intensive, with AI workloads requiring high-density computing infrastructure and generating substantial amounts of heat.

Optimising energy is only part of the equation

The researchers argue that data-centre sustainability and AI optimisation have often been treated as two separate challenges.

On one side, AI is increasingly being used to optimise infrastructure operations, including energy consumption and cooling. On the other, the growing demand for AI computing is increasing the amount of infrastructure required to support workloads.

According to the study, existing research has largely treated the workload entering a data centre as an external factor while modelling infrastructure as a fixed component. This can make it difficult to understand the net environmental benefit of an optimisation strategy when increased computing demand is considered.

The authors also found that reported energy savings across four broad optimisation technique families overlap substantially. This suggests that the current body of research does not yet provide enough evidence to clearly determine which approaches consistently deliver the greatest benefits.

Water remains largely outside optimisation models

Perhaps the most significant finding for sustainability and facilities-management professionals is the lack of water considerations in the reviewed control studies.

Data centres require cooling infrastructure to remove heat generated by servers and other IT equipment. Depending on the technology and location, cooling can have implications for both electricity consumption and water use.

Yet the researchers found that water withdrawal was not incorporated into any of the 28 primary control-oriented studies examined.

The omission is significant for regions where water availability is already under pressure. In the Middle East, where data-centre investment is accelerating while freshwater resources remain constrained, the relationship between computing demand, cooling systems and water consumption is becoming an increasingly important consideration.

Embodied carbon also missing

The review identified another major gap: none of the primary control-oriented studies accounted for embodied carbon.

Operational energy consumption is only one component of a data centre's environmental footprint. Equipment, servers, cooling infrastructure and other components also carry environmental impacts associated with manufacturing, construction and replacement.

The researchers therefore argue that optimisation should move beyond measuring direct energy savings and consider a broader set of environmental indicators.

A proposed framework for broader measurement

To address these gaps, the researchers propose CLEAR-DC, a framework designed to connect data-centre workload demand with infrastructure control decisions.

Rather than measuring only the direct energy benefit generated by an optimisation strategy, the proposed framework is intended to report a broader set of indicators covering energy, carbon, water, embodied impacts and the validation environment.

The researchers stress that CLEAR-DC is currently an architectural and methodological proposal rather than a trained system. The empirical contribution of the paper is primarily its review and analysis of the existing research landscape and the reporting framework developed from that analysis.

Why this matters for data-centre FM

For facilities managers and data-centre operators, the findings point to a broader shift in how infrastructure performance may need to be measured.

Energy efficiency remains important, but an optimisation strategy that reduces electricity consumption while increasing water requirements or relying on resource-intensive infrastructure may not deliver the same sustainability benefits when the full system is considered.

The issue is particularly relevant as AI workloads push data centres towards higher-density computing and more sophisticated cooling systems.

The study suggests that the next generation of data-centre optimisation will need to look beyond the question of “How much energy can we save?” and ask a broader question: “What is the total environmental cost of meeting the computing demand?”

For the Middle East, where data-centre investment is growing alongside some of the world's most challenging water and climate conditions, that distinction could become increasingly important for developers, operators and FM teams.