This is a strong reference from a respected source. Quoted often, usually out of context. Advocates usually took the highest projections as though they were facts. On target except that data center growth and renewables supply are accelerating. See the 2026 and other updates
Important facts
Data centers pulled 415 TWh of electricity in 2024 — 1.5% of global consumption. The US took the largest share (45%), then China (25%), then Europe (15%). Global data center demand has grown roughly 12% a year since 2017, more than four times the pace of overall electricity demand growth.
Global data center investment nearly doubled since 2022, hitting half a trillion dollars in 2024. AI-related firms in the S&P 500 added $12 trillion in market cap out of $16 trillion in total S&P gains since 2022. A typical AI-focused data center draws as much power as 100,000 households; the largest ones under construction now draw 20 times that.
Grid strain is already real. IEA estimates 20% of planned data center projects risk delay. New transmission lines take 4–8 years to build in advanced economies. Wait times for transformers and cables have doubled in three years. Gas turbine orders now carry lead times that could push new plants past 2030. Fifty percent of US data centers under development are clustering into pre-existing hubs, raising local bottleneck risk.
Globally, data centers account for roughly one-tenth of electricity demand growth to 2030 — less than industrial motors, home and office AC, or EVs. That share jumps to more than 20% in advanced economies specifically, where electricity demand had been flat for decades.
Computation used to train a frontier AI model has grown roughly 350,000 times since 2014. Large-company AI adoption rose from just over 15% in 2020 to nearly 40% in 2024; smaller firms lag well behind, largely on missing expertise. China controls about 99% of global refined gallium supply, a metal increasingly used in advanced chips and power electronics. Cyberattacks on energy utilities have tripled over four years. Emerging and developing economies outside China hold half the world’s internet users but less than 10% of global data center capacity.
Projections
Base case: data center electricity demand more than doubles to roughly 945 TWh by 2030 — slightly above Japan’s entire current consumption — and climbs to about 1,200 TWh by 2035.
That 2035 figure sits inside a range built from three scenarios: 700 TWh under a High Efficiency Case (faster hardware and model efficiency gains), roughly 1,200 TWh in the Base Case, and 1,700 TWh under a Lift-Off Case (faster AI uptake plus proactive fixes to energy bottlenecks). Gas buildout tied to data center demand runs four times higher in the Lift-Off Case than in the Headwinds Case. Nuclear growth to meet data center demand varies even more across scenarios.
On supply: renewables cover half of global data center demand growth to 2035, adding more than 450 TWh. Gas adds 175 TWh, concentrated in the US. Nuclear adds a comparable amount, mostly in China, Japan, and the US, with the first small modular reactors online around 2030.
Emissions from data center electricity use grow from 180 million tonnes today to 300 million tonnes by 2035 in the Base Case, and up to 500 million tonnes in the Lift-Off Case — still under 1.5% of total energy-sector emissions, though among the fastest-growing sources. Widespread adoption of existing AI applications could cut energy-related emissions by roughly 5% by 2035, offset in part by rebound effects such as autonomous vehicles pulling riders away from public transit.
AI-driven grid fault detection could unlock up to 175 GW of transmission capacity without new lines — more than the entire projected data center load increase through 2030. Broad industrial AI adoption could save energy equivalent to Mexico’s total consumption today. AI-led building optimization could save around 300 TWh, matching Australia and New Zealand’s combined annual generation.
Author and credits
International Energy Agency, published April 10, 2025, based on new global and regional modeling plus consultation with governments, tech companies, and the energy industry.
Link
https://www.iea.org/reports/energy-and-ai
Summary
This report set the vocabulary for the AI-power debate — the “Age of Electricity” framing, the 945 TWh 2030 figure, the Japan comparison. It runs five parts: the AI boom and its energy footprint, the electricity buildout needed to power it, AI’s use inside the energy sector itself (grids, oil and gas, industry, transport, buildings), AI’s role in accelerating energy R&D, and a final section on security, investment, skills gaps, and emissions.
The central argument is that energy and AI are now mutually dependent — AI’s growth depends on power availability, and AI in turn offers real tools for running the power system better. IEA frames the electricity-demand question with explicit uncertainty: three named scenarios spanning 700 to 1,700 TWh by 2035, not a single forecast. The report also complicates the “data centers are wrecking the grid” story in both directions at once. Locally, the strain is real and geographically concentrated. Globally, data centers remain a minor driver of demand growth next to motors, air conditioning, and electric vehicles — the outsized share only shows up in advanced economies, where grids had gone flat for decades before this.
On climate, the report pushes back on both ends of the argument: it says AI-driven emissions growth is real but small next to total energy-sector emissions, and that AI-driven emissions savings, while larger than data centers’ own footprint, fall well short of what’s needed to meet climate goals on their own.
Conclusions
The energy sector isn’t yet capturing AI’s upside — in grid optimization, industrial efficiency, or innovation acceleration — and needs policy movement on data access, digital infrastructure, and workforce skills to close that gap. IEA treats the demand growth ahead as manageable, not as a crisis, contingent on addressing grid bottlenecks directly rather than letting them constrain buildout by default. On climate, it lands on a dual rejection: AI is not a major new climate threat, and AI is not a climate solution on its own — it’s a tool that requires deliberate policy to matter either way.
edited and authored by Dave with close collaboration by Claude