Well it does say that compute is a current bottleneck, but I doubt that'll stay that way forever. There's a ton of resources going into making AI run locally, quickly. It's already gotten loads better just last year.