After the spectacular appearance of ChatGPT last year, the thirst of technology companies to create artificial intelligence products has increased significantly, which was also influenced by the interest of investors. However, the path to creating new technologies is not without a carbon footprint, and energy consumption in the industry is growing and may reach the level of small countries in the coming years.
The study examines the supply chain of artificial intelligence devices with a particular focus on NVIDIA, a manufacturer that dominates approximately 95% of the market for artificial intelligence chips. Considering the amount of hardware expected to be delivered by 2027, the range of AI energy consumption could reach 85-134 terawatt-hours of electricity each year.
In general, public information on this is still quite limited, as technology companies do not disclose enough data.
A standard rack with a normal kit is said to be about 4kW of power, which is equivalent to the power of a family home. Whereas a rack with an AI kit will be about 20 times larger, i.e. about 80 kW.