According to figures published by the International Energy Agency, power demand at data centres grew by 17% in 2024, while total electricity consumption at those same facilities expanded at a markedly steeper rate, reaching a 50% increase over the same period. Worldwide power drawn by data centres is projected to reach around 950 TWh before the decade closes — nearly twice the volume the sector consumed at the time the agency conducted its assessment. Projections place data centre electricity use at roughly 3% of worldwide power demand before 2030.

At a local level, the figure carries considerably more weight than its global standing might indicate. Power demands for data centres built around artificial intelligence workloads typically run into the hundreds of megawatts, and the largest projects now taking shape are designed to operate at gigawatt-level capacity. Power-hungry hyperscale data centres — those drawing between 300 MW and well over 1,000 MW — are placing mounting pressure on local distribution networks, with the Department of Energy noting that utilities generally face a window of only one to three years to fulfil those grid connection requests.

Data centres running artificial intelligence workloads have moved to the forefront of factors pushing electricity consumption higher. Power drawn by data centres built around artificial intelligence workloads is forecast to reach roughly three times today's levels before 2030, the International Energy Agency has indicated. Building a data centre does not, on its own, make a facility part of the artificial intelligence supply chain — cloud computing, storage, networking, and standard digital services all fall within their operational portfolio as well.

Advances in the efficiency of both AI hardware and software have not slowed that expansion. Each AI task may demand less power than it once did, yet the overall energy footprint keeps climbing as adoption broadens and workloads shift toward far heavier applications — video generation, reasoning systems, and agentic AI among them. Depending on the task, certain applications draw power at a rate orders of magnitude above what straightforward text generation requires — sometimes reaching into the hundreds or thousands of times greater per query. As usage of the technology rises, so too does the volume of energy drawn to power it — and with that, the environmental footprint tied to its operation tends to grow.

The IEA's 2025 Energy and AI report documented a stark disparity in power consumption between conventional data centres, which typically draw between 10 MW and 25 MW, and hyperscale AI-oriented facilities, which can exceed 100 MW. Ben Szuhaj, an AI strategist at a management consulting and engineering firm, argued that organisations stand to reduce wasteful deployment by establishing internal governance frameworks, noting that such structures naturally steer adoption toward applications whose environmental costs are more readily offset by tangible value.

Two factors shape how power networks handle any given load: the physical location where demand arises and the speed at which that demand builds. Where data centres cluster in relatively small areas, the strain they place on nearby power networks can run well beyond what their slice of worldwide energy use would imply. Analysts note that the country faces a notably higher degree of exposure than many of its counterparts, with projections from the IEA putting data centres on track to drive roughly half of all new electricity demand in the country by the end of the decade.

Berkeley Lab, part of the U.S. Department of Energy, attributes the mounting electricity burden on data centers largely to the proliferation of AI-driven applications, projecting that this demand will account for more than two-fifths of all U.S. power consumption growth before 2030. The CEO of GridRails, a distributed-energy coordination and settlement platform, noted that AI-linked load can scale within years while the infrastructure needed to support it — generation, transmission, and storage — typically requires far longer to plan, permit, and construct.

Figures drawn from the IEA's energy and technology assessments indicate that renewables, led by solar, wind, and hydropower, are positioned to cover close to half of the additional electricity that data centers worldwide will require through that same period. According to the IEA's 2025 Energy and AI report, renewable sources supplied around 27% of the electricity drawn by data centres globally, with coal the leading fossil fuel contributor at roughly 30%, natural gas responsible for about 26%, and nuclear accounting for the remaining 15%.

Szuhaj added that large-scale technology operators are channelling substantial resources into clean power generation and storage, a trend he described as speeding up the rollout of key clean energy infrastructure. That trajectory is expected to hold, with one observer describing direct procurement as the quickest route to guaranteed electricity at foreseeable costs — and the IEA noting that clean energy cannot expand in isolation, leaving low-carbon dispatchable sources such as nuclear and geothermal, together with natural gas as a bridging fuel, embedded in near-term supply planning.

Advances tied to the same wave of artificial intelligence development carry potential applications for the energy sector — among them, folding more renewable sources into the grid, refining how stored power is managed, and handling the growing complexity of interconnected electricity networks. IEA projections put the emissions-reduction potential of AI tools already in use at somewhere between three and four times current data centre output, provided uptake spreads broadly across the industry. Methane leak monitoring, faster identification of faults across the grid, and efficiency improvements in industrial processes each represent areas where meaningful reductions in emissions are within reach.

Forecasting how much electricity wind turbines and solar panels will deliver at any given moment has become a mounting operational challenge for those who manage the grid. Forecasting tools built on machine-learning frameworks have shown potential for strengthening how consistently renewable power sources perform. Predictive platforms drawing on live grid signals, historical asset behaviour, and meteorological pattern data can refine output forecasts — though the broader claim that such tools lift overall energy system efficiency rests, at this stage, on the assessment of a single sector participant.

Markus Levin, who co-founded XYO — a decentralised data network built around AI and robotics applications — noted that artificial intelligence holds potential for forecasting renewable energy output with greater accuracy, including the timing of that generation. The IEA puts the potential annual savings from deploying machine learning across power plant operations and upkeep at up to $110 billion by 2035, with fuel efficiency gains and lower baseline overhead driving most of that figure.

Grasso added that artificial intelligence holds promise for cutting emissions and strengthening energy-system performance, though realising those benefits depends on treating it simultaneously as a new draw on electricity supply and as a tool for managing that same demand. By redistributing electricity use away from the hours when grid pressure peaks, the technology allows the network to run more efficiently. Wider deployment of AI across more resource-intensive applications has kept overall power consumption climbing even as the energy required per individual computation has come down.

Deploying AI across energy networks faces two concrete barriers — uneven availability of dependable data and a shortage of workers with the digital skills needed to implement these tools at any meaningful scale. Levin made the case for shifting attention away from simply building ever-larger data centres and waiting for power infrastructure to catch up, citing Bitcoin as an illustration of how processing power can be spread across multiple independently run locations rather than pooled into a single centralised hub. Productivity improvements, on their own, are unlikely to keep pace with the pressures of rising consumption.

Projections point to continued growth in AI-related power consumption, which analysts say would place mounting pressure on generation capacity, storage infrastructure, and transmission networks. Clean energy sources stand to cover a notable share of that extra electricity load, while the same AI technology has the potential to help power companies balance consumption patterns and smooth out the variability of renewable output. Neither factor, on its own, holds the tools to fully resolve it.