Training generative AI is particularly energy-intensive.
The study only took into account large language models, as they are the most computationally intensive.
AI data centers require substantial amounts of water to maintain optimal temperatures for the servers that handle complex computations.
This has led to a significant increase in its water and electricity consumption.
By 2022, its electricity usage reached 11,508,131 MWh a 60% increase from 2020.
Having said that, Meta has also made significant water restoration efforts.
The company claimsits goal is to be water positive by 2030.
How Is Water Restored?
In 2017, Meta started collaborating with NGOs across the world to fund and support variouswater restoration projects.
These advancements are closely associated with a constant increase in the companys electricity and water consumption.
For instance,its corporate electricity usage increased by 35% from 2020 to 2023, reaching 3,487,000 MWh.
Like Meta,Apple has maintained 100% renewable energy use in corporate facilities since 2018.
This shows Apples commitment to reducing the environmental impact of its AI-related operations.
These innovations have led to a considerable expansion of the companys resource demands.
Its electricity consumption grew by 43% between 2020 and 2022, reaching 21,776,200 MWh.
Like Meta,Googles water stewardship strategyis all about improving watershed health in the communities where it operates.
In 2023, Google restored 18% of the water it used.
In 2023, Samsungs electricity consumption reached 29.9 million MWh.
Moreover,Samsungs water consumption was the highest by far, reaching 97.5 million m3in 2023.
Google comes at a distant second, consuming 32.7 million m3of water that same year.
Now, we must keep in mind thatbeing 100% renewable and being carbon-free are not the same thing.
The calculator takes into account whether the provider fully offsets its total emissions or not.
Servers without carbon offset dont compensate for their greenhouse gas emissionsthrough any form of carbon credit purchase.
The majority of the other highly polluting servers are in India and Australia.
How Did We Do This Analysis?
First, we downloaded the data containing the identification information of the different servers and their service providers.
This is equivalent to driving a conventional internal combustion car for about 150 km.
Perhaps one of the most notable regulatory frameworks passed so far is the EU AI Act.
It also recommends the development of mandatory reporting frameworks by companies offering AI products and services.
However, like the EU AI Act, its recommendations are not binding for member states.
hey, comment on how to improve this article.
