AI Data Centers Strain Power Grids as Compute Demand Surges

Artificial‑intelligence driven data centres are expanding at an unprecedented pace. Forecasts foresee up to 100 GW of new capacity by 2030, with AI workloads potentially accounting for half of all data‑center activity. This rapid growth is outpacing the ability of existing electricity grids to supply power, creating a timing mismatch where data‑centre construction can be completed in years while grid interconnection, transformer delivery and transmission upgrades may take five to ten years. Reports from the United States and Europe highlight transformer shortages, prolonged interconnection studies and local bottlenecks that delay or cancel projects.

The cost of building AI‑ready facilities is also rising, with next‑generation data‑centre sites estimated at $15‑$20 million per megawatt. While hyperscalers are beginning to sell more compute hardware, the fixed costs of new power infrastructure threaten to squeeze margins, as noted by Goldman Sachs. At the same time, the environmental impact is growing: a recent Allianz Trade study estimates data‑centre emissions at 286 million tonnes of CO₂ in 2025—57 % higher than previous IEA figures—and predicts water consumption comparable to Switzerland’s annual use by 2030.

Industry players are responding with a mix of strategies. Nvidia and Dell are positioning new AI‑focused hardware and storage solutions; Microsoft and Salesforce are launching secure cloud PC services for AI agents; Eco Wave Power is piloting wave‑energy generators to supply coastal data centres; and regulators and investors are warning of a potential AI‑investment bubble that could amplify financial and macro‑economic risks. The convergence of soaring power demand, grid constraints, rising capital costs and climate concerns is reshaping the trajectory of AI infrastructure development worldwide.