The use and advancement of generative AI has exploded in the past few years. This has resulted in large tech companies seeking alternative approaches we would not have previously considered to supply the power-hungry processes that require training these advanced models. These approaches include not only placing data centers strategically near power plants, but also creating their own nuclear reactors. Quantifying the power consumption of these services is difficult, and skeptics of the value of generative AI could argue that quantification is intentionally difficult. One analysis compared the generation of an image to powering a smartphone 3. Another analysis compared generating 100 words consumes up to 3 bottles of water 6. There have already been warnings that energy consumption from data centers, AI, and cryptocurrency could double in just four years 1. While this new demand is affecting some of big tech's climate goals, what is clear is that they are also looking for a competitive edge in obtaining power at a lower cost and are willing to take approaches we would not have previously considered.
Brian Krupp (Mon,) studied this question.