Aren't those simulations run on clusters of similar size (or atleast within an order of magnitude) and an agreeably responsible use of computational resources?
Feels like there should be an xkcd about this. I asked Claude to write one, but it wasn't very funny. Actually, Claude agrees with you that the energy used for training AIs is mere pocket change compared to climate simulations. (Can I trust Claude? Seems far from disinterested.)