i think inference is the thing, that also fast inference, so enterprises can just host their own and run, ig vercel do it, many more would. but zs it thinks toooo much idk how fast we can make it.