Of course I’ve taken into account model training costs, was that supposed to be your gotcha? You don’t actually think the energy cost amortization from training still accounts for the bulk of energy expended per image generated do you?