AI Drives Surge in Data-Centre Electricity Demand

According to the International Energy Agency's Energy and AI analysis, global data-centre electricity consumption could reach about 945 terawatt-hours (TWh) annually by 2030, up from roughly 485 TWh in 2025, per the IEA report. Nature's coverage of the IEA analysis framed that rise as roughly a doubling of data-centre energy use by 2030, and S&P Global notes the 945 TWh figure is equivalent to Japan's current total power consumption. Per the IEA executive summary, 945 TWh by 2030 represents just under 3% of projected total global electricity consumption that year.

Editorial analysis - technical context: The IEA report breaks down data-centre electricity demand by equipment and functions, estimating that servers account for around 60% of modern data-centre electricity use, with storage and networking contributing smaller shares and cooling varying widely by facility efficiency. Per the IEA, inference already accounts for roughly 80-90% of AI computing today and is expected to represent approximately 75% of total AI energy demand by 2030 in the base case, as AI features are embedded into everyday products and services. The dynamic reflects an industry pattern where per-query cost reductions and broader deployment increase aggregate compute demand even as hardware and cooling efficiency improve.

Editorial analysis: For practitioners, the projections shift where operational risk and cost pressure sit in the stack. Harvard Business Review notes that executives are beginning to treat electricity and power contracts as strategic inputs, and HBR references CBRE's warning about power shortages constraining data-centre growth. Industry-pattern observations show that as compute scales, constraints migrate from access to models and GPUs toward site-level power availability, grid permitting, interconnection capacity, and cooling water supplies. That creates cross-disciplinary requirements spanning ML ops, site selection, procurement, and sustainability teams.

Editorial analysis: The IEA-backed projections imply rising demand for high-density power delivery, expanded cooling capacity, and potentially new grid investments. The coverage highlights three technical pressure points:

• •rising peak power at hyperscale facilities
• •water use for evaporative and closed-loop cooling
• •land and transmission footprints for new generation and substations. The coverage also invokes the Jevons paradox - improvements in efficiency can lower unit costs and thereby expand total consumption - as a plausible mechanism that could counteract efficiency gains

• •Regional grid and permitting signals where new data centres can be sited without triggering long lead times for power upgrades.
• •Cloud and hyperscaler disclosures on power purchase agreements (PPAs) and onsite generation capacity, which will indicate how providers plan to match compute demand with contracted supply.
• •Advances in inference efficiency (model distillation, quantization, specialized accelerators) and software-level throttling or scheduling that change per-request energy footprints.

Editorial analysis: Observers should also track regulatory and market responses that could reshape costs, including capacity markets, prioritized transmission for data-centre clusters, and water-use restrictions in stressed regions. Public reporting from the IEA, academic analyses in venues such as Nature, and business commentary in outlets like HBR together create a consistent picture: accelerating AI deployment scales energy and resource needs in ways that matter to ML engineers, infra teams, and sustainability officers.