AMD, NVIDIA raise bar on AI energy efficiency

Two of the biggest makers of computer chips for artificial intelligence, AMD and NVIDIA, are promising big cuts in the electricity their data center hardware uses. AMD says that by 2030, it will take just two of its server racks to do work that needed 570 racks in 2024 — a 20-fold drop in electricity use and a 28-fold drop in carbon intensity. The company says it's already made a fourfold efficiency gain between 2024 and 2026. NVIDIA says its newest chip generation, Vera Rubin, is 10 times more energy efficient than its previous one, and it aims to cut emissions per unit of computing power by 75 percent by 2030.

This is industry and investor news, not something that changes anything about your own home or utility bill directly. But it matters for the bigger picture: AI data centers are a major driver of rising electricity demand nationwide, including new natural gas plants built to power them, which affects overall grid strain and costs. More efficient chips could slow that growth in demand somewhat.

Some analysts and investors caution that efficiency gains don't automatically mean lower total emissions, since companies could still build and sell far more chips overall. Both AMD and NVIDIA have separate, smaller-scale pledges to cut their own direct emissions and those from purchased electricity by half by 2030, with AMD further along on that goal than NVIDIA so far.