Why doesn’t AI Usage include co2e from storing generated assets on the models' cloud storage servers?

So that assets generated (whether a text based response, or an image or video file) can be accessed at a later date by the user via the chat interface, they are stored by the AI model provider. 

‍

However, it is extremely difficult to measure the co2e relating to the storage because of the differing (and undocumented) approaches taken by the AI provider. For example, it is not known how long items are stored for, and for image and video responses, the file size or compression type. It’s also unknown where they are stored and whether they are replicated multiple times for safety (as other cloud storage files are). 

‍

Given we cannot feasibly estimate this, we concluded that it cannot be included at present - therefore it is a known gap.