IEA Says Global Grid Spending Must Rise 50% to Meet Power Demand Growing at 15-Year High

IEA Says Global Grid Spending Must Rise 50% to Meet Power Demand Growing at 15-Year High The agency puts power-demand growth at its fastest pace in 15 years through 2030, with data centres only one part of a wider electrification surge straining grids. Global electricity demand is rising at its fastest pace in 15 years and will keep climbing through the end of the decade, the International Energy Agency said in its latest assessment (2026-05-19), driven by AI infrastructure, advanced manufacturing and electrification.⁷ That matters because the grid is not built to absorb it. The IEA puts annual average demand growth at 3.6% between 2026 and 2030, fed by industry, electric vehicles, air conditioning and data centres, and says meeting that load would require lifting annual grid investment by roughly 50% from $400 billion.⁷ In the United States the squeeze is already visible, with grid engineers, utility executives and regulators describing permitting, supply chains and interconnection queues that cannot match the speed at which data centres want to connect.⁵ Data centres are the headline, and the numbers behind them are large. The IEA reckons a single server rack in an advanced facility could draw peak power equivalent to 65 households by 2027, and that US data centres now account for about half of the country's incremental demand growth.^4,5 America needs roughly 5,000 miles of new high-voltage transmission to keep pace, by the account grid engineers gave Quartz (2026-05-19).⁵ The EIA's longer view confirms the trajectory without the hype. In its Annual Energy Outlook 2026, the agency projects server electricity use climbing from about 7% of commercial-sector consumption in 2025 to between 22% and 33% by 2050, with standalone data centres growing faster than all other server rooms combined.² By mid-century, server consumption alone reaches somewhere between 446 and 818 billion kilowatt-hours, a range wide enough to flag how uncertain the forecast really is.² That uncertainty cuts against the loudest demand bulls. The EIA's counterfactual case assumes that after 2040 servers keep getting more efficient, shaving 10% off average operational power draw every three years beyond historical trends.² Efficiency per AI task is improving fast even as aggregate consumption rises, a paradox the agency's framing makes plain.⁶ The bullish case for power demand rests on the bet that compute growth keeps outrunning those gains. There is a counterweight worth taking seriously. The Economist argues that Americans' rising electricity bills should not be pinned on AI, and notes that large new loads can actually help.³ PG&E, the big Californian utility, estimates that adding a gigawatt of demand could lower customer bills by up to 2%, because fixed grid costs spread across more consumption.³ If that holds, the political backlash building around data-centre power use rests on a shaky premise. The hyperscalers are not waiting for the grid to catch up. Alphabet, Google's parent, paid $5 billion in December (2025-12) for Intersect Power, a developer of utility-scale solar and battery storage, a deal that buys generation rather than just contracting for it.³ The pattern of tech firms securing nuclear restarts and backing reactor startups points the same way: demand that cannot find clean firm supply will take whatever is available.¹ That is the unresolved tension for traders. The same buildout pushing power demand higher is, on the margin, keeping fossil generation in the mix where renewables and storage cannot deliver round-the-clock load.¹ One forecaster captured the IEA's own message bluntly: the world is getting greener per unit of compute and consuming more power than ever, and it is not ready for either.⁶ Watch the investment number. The gap between the $400 billion of annual grid spending and the figure the IEA says is needed is the binding constraint, not chip supply or model demand.⁷ Until transmission build rates and interconnection timelines move, every gigawatt of new data-centre load lands on infrastructure that engineers already say cannot keep up.⁵ The next signal is whether 2026 grid-spending commitments start closing that 50% gap or merely acknowledge it.⁷