I am honored that Harvard Business School asked for my perspective on one of the hardest questions in energy right now: where the electricity to run the AI boom is going to come from.
At this point, everyone is aware of the growing gap between data center build outs and the limited power available on today’s grid. Why is this? Getting an answer from generative AI takes roughly ten times the electricity of a traditional Google search, and data centers are on track to double their electricity demand in two years. In my view, the alarm is warranted but the answer is simpler than most people realize.
Three things give me confidence that the grid will not collapse due to power shortages. First, efficiency curves are relentless, and I expect the cost of training these models to fall by as much as 90% in two to three years as chips and algorithms improve (Moore’s Law remains in full force). Second, more of the inference work is moving onto the phones and laptops we already own, which pulls load off the data centers entirely. And, third, the hyperscalers are buying renewables at enormous scale, including Microsoft’s $10 billion power purchase agreement.
Read the full article, Where Will We Find the Energy to Power the AI Revolution?, by Julianne Elaine White for Harvard Business School.
Illustration by Richard Borge for Harvard Business School, December 2024.
Leave a Reply