The ScienceDaily piece also shows that AI emissions can vary massively depending on the model, how it’s designed, and how people use it. Smaller, more concise models produce far less CO₂, and users can lower the impact even further by asking for shorter outputs or choosing more efficient systems.
The WRI analysis is basically saying the same thing: data centers do bring problems, especially around electricity demand, water use, land use, and where these facilities get built. Yet it doesn't automatically prove AI is evil or destructive. It means the infrastructure around it has to be handled more responsibly. The takeaway is that AI has an environmental cost and that cost can be reduced through smarter design and responsible use which is exactly what AI companies are pushing towards and putting their focus on conserving and reusing energy resources.