Technically yes, that's a good point. But at the same time, on-premise servers are not good for data science problems, as you usually need high computational powers. This especially could be a problem for not-yet established companies and startups, buying a server will be very expensive for them.

Big companies can definitely use on-premise servers, but that is only the case when they have stable income and see a big need in their own servers.

Right now cloud-based solutions are much more popular for AI stuff, as they are much easier to scale, use and maintain.