The commercial reality I’m sure they’re dealing with is that the percentage of users who care enough to deal with everything required for local inference is nowhere close to large enough to make it into an initial launch.