Big tech emissions have been on the rise especially since 2020 due to AI. Exiobase’s latest version uses data from 1995-2020 with “now-casted” data for 2021-2022. So they don’t really cover the risen emissions due to AI. And AI is currently being pushed into every possible software and feature. This is of course problematic in many ways, one of them being companies with high IT service emissions setting climate targets with baseline around 2023-2025. But it’s really an issue that concerns every company to some degree.
Has there been any talks about this within C+A+D or with Exiobase? Or have any of you in the community found meaningful solutions? We have been discussing a bandage solution with one of our clients in which we would assess the emissions of Google, Microsoft & Amazon and see how their emissions have changed from 2020 to 2024 and use that to correct the spend-based emissions of all IT service emissions. Of course, in a case with relatively small number of IT service providers the best course of action would be to use supplier specific emissions/emission intensity.
Actually it seems that the problem is that Exiobase Computer and related services (product) EF still hugely overestimates Google and Micorosft emissions as the EF is 0,205 kgCO2e/€ for 2025 and the location-based emission factors for Google and Microsoft for 2024 are 0,03 and 0,11, respectively.
As a software providor and user of several AWS services (hosting, AI services, etc.), I can share the following info:
We know in detail the amount of computational power we use (expressed in # servers, # tokens, etc.)
To translate that to carbon emissions, you need to translate it to electricity consumption which is almost impossible as there is a lot of interdependency with the overal load on AWS server
We know in detail where our servers are based (i.e. in France for CAD) so we know what emission factor to us
There was another community post on this topic. Maybe that might be of help