Generative artificial intelligence platforms incur direct compute costs on graphics processing units for every output generated by users