AI data centers face soaring power and cooling demands

The rapid expansion of artificial‑intelligence workloads is driving an unprecedented surge in data‑center electricity use and heat output. Analysts forecast a 26 % rise in global data‑center power demand by 2026, with U.S. AI‑related consumption possibly reaching 123 GW by 2035. This growth strains existing grid infrastructure – transformers, high‑voltage cables and substations – causing delays in new facility deployments in both the United States and the United Kingdom. Liquid‑cooling solutions, such as those offered by Vertiv, are becoming essential to keep high‑density GPU racks from overheating. The shortage of compute capacity has led firms to lease surplus hardware at premium prices and to seek more efficient platforms; NVIDIA’s Vera Rubin system claims to cut GPU requirements for post‑training by 75 % through integrated design. Start‑ups like France’s Infinity are developing software that adapts AI inference workloads to any chip, aiming to reduce the ecosystem’s reliance on a single hardware stack. At the same time, the sector’s carbon footprint is drawing scrutiny. Researchers warn that data‑center emissions could quadruple by 2030 without carbon‑capture measures, while UN‑UNEP officials stress the need to balance energy, water and mineral use against AI’s societal benefits. The combined pressure on power grids, the premium cost of compute, and mounting environmental concerns underscore a critical infrastructure bottleneck for the AI era.