The Download: introducing the AI energy package

AI's "greedy appetite" for power has long been documented, but its full extent remains underexplored. MIT Technology Review's investigative team offers a gripping look into this issue, revealing how inference—AI's ability to process queries and generate content—will soon dwarf the energy required to train models. Their journey takes readers from the arid deserts of Nevada, where data centers demand water for cooling, to Louisiana's Meta facilities, which promise to fuel future ambitions but are already burdened by hidden secrets. This exposé also sheds light on why transitioning AI power needs away from clean energy remains a daunting challenge.