“Companies with AI products at a reasonable scale are transitioning from per-token pricing models to paying for dedicated underlying GPU hardware.”