AI Energy Consumption: How Much Power Does AI Use? (toolpod.dev)

AI systems are significantly more energy-intensive than traditional online activities, consuming up to ten times more electricity per query than a Google search. This escalating energy demand is concerning, given that there are now over a billion AI queries made daily, translating to substantial electricity and water consumption. For instance, each AI interaction may use around 0.3 watt-hours, and larger tasks like video generation can reach up to 950 watt-hours. The increasing reliance on specialized GPUs, such as Nvidia's H100, compounds these energy requirements, leading to an estimated annual consumption of 32-80 million metric tonnes of CO2 emissions directly from AI workloads, which rivals the emissions of major cities like New York. The environmental implications are critical, as modern data centers require vast amounts of cooling water—up to 5 million gallons per day—highlighting the strain on local water resources, especially in drought-prone areas. Despite ambitious climate commitments from major tech companies, actual emissions are rising due to the growing energy demands from AI inference rather than just model training. With predictions that data center electricity consumption will double by 2030, the AI/ML community must reconsider usage patterns and implement strategies to reduce individual carbon footprints significantly. Addressing this transparency gap and incorporating energy metrics in user decisions will be vital for sustainable AI development moving forward.