AI and its Power Problem

Artificial intelligence has rapidly moved from experimental technology to everyday infrastructure, embedded in search engines, productivity platforms, customer service, cybersecurity and an expanding range of business applications.

The environmental consequences, however, remain far less visible than the capabilities appearing on users’ screens.

Every Prompt Has a Footprint

Every AI request relies on physical infrastructure: processors, servers, cooling systems, network equipment and data centres consuming electricity and water around the clock. As models become larger and adoption increases, sustainable IT strategies built around efficient laptops, responsible recycling and cloud migration no longer tell the whole story.

Demand Is Outrunning Efficiency

Data centres currently account for around 2.5% of EU electricity consumption, but capacity is forecast to rise from 12 gigawatts in 2025 to 28 gigawatts by 2030. Across advanced economies, data centres could generate approximately 20% of the growth in electricity demand during the same period. The EU is now considering minimum energy-efficiency standards and a sustainability label covering measures such as water consumption and clean energy use. Reuters

The UK Faces the Same Tension

The UK faces a difficult balance between digital expansion and environmental responsibility. Government policy treats AI and data centres as strategically important engines of investment and economic growth, while questions surrounding grid capacity, water availability and carbon emissions become harder to avoid. A recent House of Commons Library briefing reflects the growing political attention being paid to the energy use, water consumption, planning and resilience of data-centre development. House of Commons Library

CIOs Need Better Answers

For CIOs, sustainability therefore needs to enter the AI conversation before deployment rather than appearing later in an ESG report. That means asking suppliers where workloads are processed, how facilities are powered and cooled, whether environmental figures are independently verified and how consumption changes as usage grows. Claims that a service runs on renewable energy also deserve scrutiny, particularly where they depend on certificates rather than additional clean-generating capacity.

Model Choice Matters

Efficiency must be considered at application level too. The most capable model is not automatically the most appropriate model for every task. Smaller models, better prompts, controlled retention, workload scheduling and limits on unnecessary generation can reduce computational demand without preventing useful adoption. Removing redundant data and abandoned cloud resources remains equally important; storing everything indefinitely carries a financial and environmental cost.

Procurement Holds the Power

Procurement teams can reinforce this approach by including energy, water, repairability, expected lifespan and end-of-life recovery in technology assessments. Sustainable IT becomes considerably more credible when those measures influence supplier selection and contract renewal rather than sitting in a separate questionnaire with little commercial weight.

AI Must Earn Its Footprint

AI can help organisations optimise buildings, transport, supply chains and energy use, but those benefits cannot be assessed without accounting for the infrastructure required to produce them. The next phase of sustainable IT will be defined by organisations that can connect digital ambition with physical consumption—and demonstrate that the intelligence they are buying delivers more value than environmental debt.