Environmental impact of AI - statistics & facts

Data centers boosting electricity demand

Energy consumption is probably the most talked-about environmental impact of AI. In 2023, artificial intelligence power consumption worldwide was estimated at 4.5 gigawatts, or roughly eight percent of data center power consumption that year. By 2028, this figure is expected to more than triple and could account for up to 20 percent of data center’s power usage. This demand is not likely to be evenly distributed across the globe. In the U.S., data center energy consumption could account for six percent of the national total by the middle of the decade. In Ireland, the power demand from data centers is forecast to add up to one-third of the country’s total needs.

There is a lot of investment and research towards improving the energy efficiency of AI models, both during the training and inference phases. Nevertheless, experts fear a paradox often seen in the scaling up of new technologies, in which an increase in efficiency is compensated by a much larger growth in use, as a result of the improved efficiency. For example, the average electricity demand for a Google request is estimated at 0.3 Watt-hours. With the growth in AI efficiency, there are plans to incorporate generative AI as the standard. If Google were to incorporate AI to all its searches, the electricity demand of the average request could be up to 30 times higher, drastically increasing the tech company’s power demand.

Solving climate change or aggravating it?

Along with a growth in power demand, Google’s greenhouse gas emissions soared by nearly 50 percent in 2023 compared to 2019 levels, mainly driven by the growth in AI use. AI’s carbon footprint comes mainly from the electricity produced for consumption in the data centers used to run its models, and is thus directly linked to the carbon footprint of the grid it is connected to. Although in theory this could be easily adjusted by favoring regions with a high share of clean electricity sources, in practice, global renewable capacity growth is still far behind where it needs to be for a net-zero by 2050 scenario. In 2023, fossil fuels still accounted for more than half of the global electricity mix.

AI’s thirst for more than knowledge

Although less talked about, the water demand of AI is a growing cause of concern. AI’s water usage mostly stems from the need to cool data centers. In 2023, water withdrawals by Google surged 14 percent year-over-year. For Microsoft, the annual water withdrawal growth surpassed 20 percent. Despite pledges by both companies to replenish the amount of water used, there are concerns that the growth of AI demand could significantly aggravate water scarcity in already drought-prone areas around the world. In 2023, almost one-third of Google’s water withdrawals were in watersheds with medium or high scarcity. In addition to the water used directly in the cooling of data centers, water used in electricity production to power them, and the high-water demand in the production of semiconductors used in AI servers also contribute to the technology’s supply-chain water footprint. However, there is still very little disclosure around these figures.

Looking ahead, as AI use grows exponentially by being incorporated into daily routines, so will its energy, water, and carbon footprints. AI tools could be put to good environmental use, to detect leaks from pipelines, monitor deforestation, aid in circular product design, and enhance warning systems for climate disasters, to name a few. However, clear regulation, reporting standards, and data transparency will be key in assessing and controlling the environmental impacts of AI and guarantee its sustainable use.