Buyers are starting to optimize for the cost of experiment. They’re not asking how much a GPU costs; they’re asking how much it costs to get a reproducible, safe, useful result — fast enough to matter ...