Big AI’s dirty secret: The environmental cost of generative AI

By Michael R. Wade and Jialu Shan

As artificial intelligence (AI) continues to reshape industries and revolutionize the way we work, the environmental impact of AI models and data centers is coming under increasing scrutiny. The infographic “Big AI’s Dirty Secret” sheds light on the staggering levels of electricity and water consumption by major tech companies powering AI systems.

While AI’s potential is boundless, its growing carbon and resource footprint raises crucial questions about sustainability. However, companies are beginning to address these challenges by investing in renewable energy, optimizing AI model efficiency, and developing innovative cooling technologies to reduce water consumption.

The power-hungry giants of AI

The infographic highlights the electricity consumption of five major tech companies, Google, Microsoft, Meta, Apple, and Nvidia. Google leads the pack with 25 terawatt-hours (TWh) of annual energy use, closely followed by Microsoft at 23 TWh and Meta at 15 TWh. Apple and Nvidia, though consuming less, still contribute significantly, with 3.5 TWh and 0.6 TWh respectively.

To put this into perspective, Google’s energy use alone is equivalent to the electricity consumption of 2.3 million U.S. households, while Microsoft’s consumption could power 48 Disneyland Paris parks for an entire year. This massive power demand underscores the increasing strain AI and data center infrastructure place on global energy resources.

Energy efficiency: Who gets the most bang for the buck?

AI companies are not just energy-intensive; they also vary in efficiency. The infographic examines how much energy is used per million dollars earned. Meta emerges as the least efficient, using 113.6 megawatt-hours (MWh) per million USD generated, followed by Microsoft (111.2 MWh) and Google (82.3 MWh). Nvidia and Apple, by contrast, are significantly more efficient, consuming just 10.1 MWh and 9.1 MWh per million USD, respectively.

Apple, in particular, stands out as the most efficient, making $1 million in sales using only as much electricity as needed to charge 152 electric vehicles. This stark contrast highlights Apple’s superior energy efficiency compared to Meta and Microsoft, which use significantly more electricity to generate the same revenue. In contrast, Meta’s AI operations appear to require over ten times more energy to generate the same amount of revenue.

The water crisis: AI’s thirst for resources

AI is not just an electricity guzzler, it is also a major consumer of water. Data centers rely heavily on water for cooling systems to prevent overheating. Google’s annual water consumption reaches a staggering 24 million cubic meters, which is enough to fill over 9,618 Olympic-sized swimming pools. Microsoft, while consuming less, still uses 7.8 million cubic meters, followed by Apple (6.1M m³), Meta (3M m³), and Nvidia (134K m³).

To contextualize, Google’s water usage could fill 120 million bathtubs, which is equivalent to enough baths for almost everyone in Japan. Meanwhile, Microsoft’s annual water consumption could fill 9,000 Boeing 747-400 jets, illustrating the enormous resource strain AI companies place on water systems.

Read more at IMD.org