IEA Quantifies AI’s Energy Constraint: Data Centers to Consume as Much Power as Japan by 2030
Global data center electricity demand will nearly double to 945 TWh within four years, with energy scarcity replacing silicon shortages as the binding constraint on AI deployment velocity.
Global data center electricity consumption will surge from 415 TWh today to 945 TWh by 2030—equivalent to Japan’s total power demand—according to the International Energy Agency‘s authoritative Energy and AI report released in April 2026. AI is the primary driver, with accelerated server electricity consumption projected to grow 30% annually through the decade. The IEA’s analysis reveals energy availability, not semiconductor supply or capital, as the new hard constraint on AI deployment.
The scale shift is immediate. In the United States, Data Centers will account for nearly half of electricity demand growth through 2030 and consume more power than all energy-intensive manufacturing combined—aluminum, steel, cement, and chemicals—by decade’s end, according to the IEA. A single hyperscale AI facility requires 100-300 megawatts of continuous power, equivalent to a mid-sized city. AI inference tasks consume up to 1,000 times more electricity than traditional web search.
“AI is one of the biggest stories in the energy world today—but until now, policy makers and markets lacked the tools to fully understand the wide-ranging impacts,” said Fatih Birol, IEA Executive Director, in an IEA statement. “Global electricity demand from data centres is set to more than double over the next five years, consuming as much electricity by 2030 as the whole of Japan does today.”
Grid Constraints Force Infrastructure Pivot
Energy availability now dictates deployment timelines. The IEA estimates 20% of planned data center projects face delay risk due to grid connection bottlenecks. US data center electricity demand currently sits at 41 gigawatts—a 150% increase over five years—with the Department of Energy projecting potential tripling by 2028 under high-growth scenarios, according to Tech-Insider.
Hyperscalers are responding by bypassing grids entirely. Oracle announced a 2.8 gigawatt fuel cell commitment with Bloom Energy on 14 April 2026, expanding an existing 1.2 GW contract. The deal reflects Oracle’s strategy of building on-site generation to eliminate interconnection queue delays, reported by Bloomberg. Oracle Cloud Infrastructure revenue is growing 84% year-over-year, requiring infrastructure deployment at scale.
“By rapidly deploying Bloom’s reliable, efficient fuel cell energy, we are quickly meeting the demands of our customers across the United States. Together, Bloom and Oracle Cloud Infrastructure are building the power foundation and AI infrastructure to accelerate American AI leadership.”
— Mahesh Thiagarajan, Executive Vice President, Oracle Cloud Infrastructure
Grid operators face structural limits. Capacity market clearing prices in PJM—the largest US grid operator—jumped to $329.17/MW for the 2026-2027 delivery year, more than 10 times the $28.92/MW clearing price two years earlier, according to PJM. Communities near major data center clusters in Virginia, Texas, and Georgia are already experiencing electricity rate increases of 8-15% due to infrastructure strain, per March 2026 analysis from Consumer Reports.
Fossil Fuels Fill Near-Term Gap
Renewables will supply half of global data center electricity growth to 2035—around 450 TWh—but cannot meet total demand due to intermittency and infrastructure constraints. Natural gas will expand by 175 TWh over the same period, with Nuclear contributing a similar amount, primarily in China, Japan, and the United States, according to the IEA. Small modular reactors are projected to come online around 2030, but deployment timelines remain uncertain.
The United States has doubled the amount of gas- and oil-fired capacity in development over the past year, driven partly by AI industry energy demand. Projects face long lead times and sharply rising costs as utilities compete for materials and skilled labour.
Through 2030, 40% of additional data center energy consumption will still be supplied by gas- and coal-based sources, the IEA estimates. At the LNG 2026 conference in Doha this month, leaders from Shell, ExxonMobil, TotalEnergies, ConocoPhillips, and QatarEnergy concluded that AI, data centers, electrification, and population growth are pulling the energy system to a new scale faster than grids and policy can adapt, reported by FNArena.
Nuclear Contracts and Uranium Bottlenecks
Big Tech signed more than 10 gigawatts of new nuclear capacity contracts over the past year, with Goldman Sachs projecting three plants could come online by 2030. Nuclear offers firm, dispatchable carbon-free electricity—critical attributes for data centers requiring 24/7 uptime.
Uranium supply is emerging as a new bottleneck. NexGen Energy entered talks with data center providers in February 2026 to help finance its Rook 1 uranium project in Saskatchewan, Canada, according to Tom’s Hardware. OpenAI has called on the US government to build 100 gigawatts of additional power-generating capacity annually—roughly 100 nuclear reactors per year—for AI development.
- By 2030, AI data centers could represent 2-3% of global demand for copper and related Critical Minerals
- Uranium supply chains face acute pressure with 10-17 year lag times for new mine capacity
- Copper, lithium, and uranium demand overlaps between AI infrastructure and energy transition, creating competition
- Hyperscalers moving to direct supply contracts with mining companies to secure long-term access
The IEA analysis positions critical mineral supply chains as a geopolitical chokepoint. AI data centers could account for 2-3% of global demand for copper and related elements, creating direct competition with electrification and renewable energy deployment.
Regional Impacts Vary Sharply
In Ireland, data centers accounted for 21% of electricity demand in 2023. The IEA forecasts the sector could reach 32% of total electricity demand by 2026, forcing grid operators to implement connection moratoria. The Netherlands and Singapore have imposed similar restrictions. Southeast Asian nations are positioning themselves as alternative data center hubs, but face their own infrastructure constraints, per analysis from the Brookings Institution.
US electricity sector capital expenditure rose from 13% year-over-year growth in 2023-2024 to 23% in 2025, reflecting utility attempts to keep pace with data center demand. Global investment in data centers nearly doubled since 2022, reaching $500 billion in 2024.
Efficiency Gains Insufficient to Offset Growth
Power demand for inference tasks is projected to increase at 122% compound annual growth through 2028. Token generation volumes have grown 5x to 9x year-over-year at major tech companies, according to analysis from IO Fund. While chip efficiency improves, total compute demand is growing faster.
The IEA’s sensitivity analysis projects data center electricity consumption could range from 700 TWh to 1,700 TWh by 2035, depending on deployment velocity, efficiency gains, and workload intensity. The high-end scenario would represent nearly 5% of global electricity consumption.
What to Watch
Grid interconnection queue lengths in major US markets will signal whether utilities can keep pace with hyperscaler expansion plans. Oracle’s fuel cell strategy may become standard if queue delays persist beyond 18-24 months. Nuclear capacity contracts will test whether small modular reactor timelines hold or slip further—2030 deployment remains aspirational without regulatory streamlining.
Uranium spot prices and long-term contract volumes will indicate whether hyperscaler direct financing of mines becomes widespread. Capacity market clearing prices in PJM and other regional operators will show if April’s 10x spike was transitory or structural. Regional electricity rate trajectories in Virginia, Texas, and Georgia data center clusters will determine political viability of continued expansion.
IEA scenario updates in Q3 2026 will refine the 700-1,700 TWh range for 2035, incorporating actual deployment data from the first half of 2026. The gap between hyperscaler capex commitments and available Grid Capacity remains the critical constraint on AI scaling velocity.