@Allan2700, it’s an interesting question, but a complicated one, and honestly not one we have the resources to answer in a quantitative fashion (assuming that’s even possible). @alex estimates that we spent ~1,400 kilowatt hours training our last computer vision model, and if you believe the 2.21 pounds / kilowatt hour stat for US coal-fueled power plants (and you might not, given the current state of the US government), that would be 3,094 pounds of carbon emitted, ~1.4 metric tons. However, that electricity was all consumed at the California Academy of Sciences, which, as Cassi pointed out in her links above, derives all of its electricity from hydroelectric and solar power, not coal (with the usual caveats about the impossibility of sourcing electricity on an electrical grid), so you might say the carbon footprint of the energy consumed to create that model was zero.
You might also say that electricity in a grid comes form a diverse array of sources, and that most electricity production in California is from burning natural gas. You might further say that the components in the solar arrays on the roof of the California Academy of Sciences might have their own pollution problems, to say nothing of the electrical infrastructure required to keep the building provisioned with electricity. You might even say that hydroelectric power from Hetch Hetchy Reservoir comes at the cost of destroying one of the most beautiful natural areas in California, the controversy over which galvanized (pun intended!) the environmental movement in the United States. So, it’s complicated.
Overall, I think the electricity consumption of the iNat team in the CAS building is relatively small, and that CAS does a good job trying to use renewable sources of electricity, so take that for what it’s worth.
We probably consume much more electricity on our virtual servers at Microsoft Azure, but they claim to be carbon neutral in their electricity consumption and on their way to being carbon negative, though a significant portion of that is due to purchasing carbon offsets according to their whitepaper. We use machines in Microsoft’s “West US 2” data center, which is in Washington state, where hydroelectric plants are the biggest electricity providers.
The carbon footprint of iNaturalist as a whole is an even more ponderous and hard/impossible-to-answer question. Does iNat encourage people to consume more carbon through travel? To consume more electricity through more use of digital devices? To contribute to more pollution by supporting the market for electronics? My guess is that we’re a part of all these problems, but probably a very small part.