Concerns over artificial intelligence's (AI) potential effects on the environment are growing as AI increases power consumption. According to The New Yorker, just OpenAI's well-known generative AI, ChatGPT, is thought to need more than half a million kilowatt-hours a day to process 200 million user requests every day. This is more than 17,000 times more energy than the typical US household uses on a daily basis.
The energy drain may worsen as generative AI is applied more widely. Data scientist Alex de Vries calculated that if Google included generative AI in every search, the company might use almost 29 billion kilowatt-hours a year—more than the annual electricity use of whole nations like Kenya, Guatemala, and Croatia.
De Vries draws attention to how energy-intensive AI is, pointing out that an AI server may already use as much energy as more than a dozen UK homes put together. However, because of model operating unpredictability and a lack of openness from large tech businesses driving the AI boom, quantifying the overall electricity usage of the AI industry is difficult.
According to De Vries' report, the AI industry as a whole may need between 85 and 134 terawatt-hours per year by 2027. This estimate is based on data from Nvidia, a major chipmaker in the field. This might account for as much as half of the world's electricity usage by 2027, which would be a substantial amount.
In contrast, some of the most energy-intensive companies in the world don't use as much AI as is anticipated. For example, Samsung uses almost 23 terawatt-hours to power its data centres, networks, and user devices, but Google and Microsoft, two of the biggest IT companies, use about 12 and 10 terawatt-hours, respectively.
In order to reduce its negative effects on the environment, AI must manage its energy consumption as it spreads over more and more industries. The development of AI hardware and algorithms that consume less energy is essential to striking a balance between the advancement of technology and environmental objectives.
