The power of AI

Although data center workloads nearly tripled between 2015 and 2019, the sector’s power consumption remained essentially flat at roughly 200 Terra Watt hours (TWh) per year. Data centers have become more power efficient though a combination of systems architecture and the steady cadence of semiconductor design improvements. However, according to research from Goldman Sachs, data center efficiency gains have slowed in recent years and the firm forecasts that AI will contribute to a more than doubling in data center energy usage by 2030. According to the Goldman report, by 2028, AI is expected to account for 19% of global data center power demand – roughly equivalent to the current power consumption of the Netherlands. Although AI models are themselves becoming less power hungry and recently released AI semiconductors are delivering up to 75% improvements in power efficiency, usage of AI applications could still drive accelerated growth in energy requirements.

Source: Goldman Sachs, May 2024