AI’s soaring energy use threatens sustainability, NTT data warns

A new study by NTT Data has raised concerns over the growing environmental toll of artificial intelligence, warning that its rapid expansion is becoming unsustainable.

Researchers predict that AI workloads will account for more than 50% of global data centre power consumption by 2028.

The report stresses the urgent need to embed sustainability into every stage of AI’s development and deployment. The technology’s massive energy requirements stem from the computational power needed to train large language models, run inference pipelines, and maintain always-on services.

Beyond energy consumption, the study highlights other major environmental issues, such as high water usage for data centre cooling, electronic waste, and the extraction of rare earth minerals used in hardware manufacturing.

NTT Data’s AI experts and sustainability consultants recommend that organisations adopt holistic sustainability goals instead of focusing solely on traditional AI metrics like accuracy and speed. “Efficiency must be a core design principle,” the study says.

The report calls for the creation of standard and verifiable measures for AI’s environmental footprint, including energy consumption, carbon emissions, and water use. It points to emerging benchmarks such as the “AI Energy Score” and “Software Carbon Intensity for AI” as tools to integrate sustainability into governance, procurement, and compliance.

According to the study, sustainable AI requires a full-lifecycle approach — from raw material extraction and hardware manufacturing to deployment and disposal. Key steps include extending hardware lifespans, optimising cooling systems, and applying circular economy principles.

The responsibility for building sustainable AI, the report notes, should be shared among hardware manufacturers, data centre operators, software developers, cloud providers, policymakers, investors, and consumers.

Many organisations, it says, still focus only on energy or emissions while neglecting water usage, rare material depletion, and e-waste. Even when environmental goals are set, they often lack practical strategies for implementation.

To address these challenges, the report outlines several best practices.

These include applying green software engineering patterns to reduce resource use, scheduling AI workloads to align with renewable energy availability, and leveraging both remote GPU services and on-premises AI. It also recommends reducing e-waste by prioritising modular and upgradable components and extending hardware life through refurbishment, reuse, and responsible recycling.