AI's impact on the environment

28 Nov 2024

While AI keeps getting better and better, getting it there is having a very negative impact on the environment. This seems far-fetched at first, but if we think about how large tech companies make the AI better it all makes sense.

Companies like Amazon, Google, and Microsoft began making investments in new data centers all around the world. While they will be used for other things too, their main focus is training AI models to make them better. Training AI models takes a lot of compute power, so these data centers need to have a lot of hardware running. The hardware needs a lot of power and a lot of cooling. The main issue is that clean water is used for cooling because it is far more effective than dirty water and will not damage the equipment. You may think how much water can they use? In 2023 Microsoft's data centers used 7.8 million cubic liters of water. That is 34% more than 2022, before they started investing heavily in AI. The water comes from the same sources as our drinking water, which is becoming more scarce even without this. One study from 2022 claimed that a mid-sized data center can use up to 1 million liters of water a day, which is enough to supply a 100,000 homes. Also, another factor is the increased power consumption. To run all the data centers requires a lot of power. Some companies are making deals to restart nuclear reactors just to run the hardware needed for AI, and that power is only for them. All of these things are kept below radar to avoid regulations from governments, but if this keeps going it will do more harm than good.

Don't get me wrong, I am not against AI in general. It can help up understand certain diseases better, it can help us create new medicines. I think we should slow down a bit and see how we can improve the process of getting to better and more efficient AI without doing more damage that the benefits we will see from them.