AI is set to double data centre electricity use by 2030

Upgrades to electricity grids might not keep up with the demands of data centres

A new from the International Energy Agency (IEA) warns that electricity demand from data centres worldwide is set to more than double by 2030, driven largely by the rapid growth of artificial intelligence. The IEA’s Energy and AI report projects that data centres could use around 945 TWh per year by 2030, up from about 415 TWh in 2024. That is slightly more than the entire electricity consumption of Japan today.

AI is identified as the most important driver of this increase, with electricity demand from AI-optimised data centres alone forecast to more than quadruple over the decade.

What’s driving the surge?

Several trends are converging:

• Explosive AI adoption, larger models, more frequent training and deployment across consumer and enterprise services.

• Rising rack power densities, tens of kilowatts per rack are becoming standard in AI clusters, with even higher densities under development.

• Always-on digital services, 24/7 operation for cloud, streaming, storage and inference workloads.

Under the IEA’s central scenario, data centres could account for more than 20% of electricity-demand growth in advanced economies by 2030.

Why this is an energy and thermal problem

Every watt consumed by servers eventually becomes heat that must be removed. As AI accelerators push to higher power levels, traditional approaches (air cooling and basic liquid loops) are under pressure:

• Cooling systems must handle higher heat fluxes at chip, module and rack level.

• Data-centre operators face rising Power Usage Effectiveness (PUE) challenges, as more electricity is used just to keep hardware within safe temperatures.

• In some regions, grid capacity and connection delays are becoming bottlenecks for new AI campuses.

On top of this, water use for certain cooling technologies is attracting increased scrutiny, particularly in areas already facing water-stress.

Consequences for energy systems

The projected growth raises system-level questions:

• How will electricity networks accommodate large new loads that may switch on and ramp quickly?

• What mix of renewables, nuclear, gas and storage will supply these centres while staying compatible with climate targets?

• Can waste heat from data centres be reused—for example in district heating or industry—to improve overall efficiency?

Some operators are already exploring long-term contracts for nuclear power and other low-carbon sources specifically for AI and cloud facilities.

A growing role for simulation and design optimisation

As power and cooling demands increase, there is a strong need for:

• High-fidelity airflow and thermal simulations of halls, racks and cold-plate designs

• Optimised cooling architectures (liquid, two-phase, immersion and hybrid systems)

• Integrated energy-system modelling linking data centres with local grids, renewables, storage and potential waste-heat users

These tools help operators and designers understand where hotspots will occur, how different cooling strategies perform, and how best to integrate new facilities into already-stressed energy networks, before committing to major capital projects on the ground.

From risk to design challenge

The IEA also notes that AI can be part of the solution, with potential to optimise grids, industrial processes and building energy use.

Whether AI becomes a net burden or a catalyst for smarter, lower-carbon systems will depend heavily on how new data centres are designed, how they are powered, and how effectively heat is managed and reused.

For engineers, planners and policy-makers, the takeaway is clear: AI is not just a computational revolution; it is a major energy and thermal-engineering challenge that will shape infrastructure decisions for the rest of this decade. Source: IEA (https://shorturl.at/H5ZYE) Nature (https://shorturl.at/tMpMN)