Should with start building AI products that run locally ?

The CES ended a few days ago, and I think that most processor producers like Intel or AMD are focusing mainly on their NPUs and AI processing locally. Although running everything related to AI on our servers is quite transparent for the user, if we start using this new hardware from now on I think we could reduce many costs and deliver better versions of our products to the user. Next week I am launching a solution for processing images with AI that runs completely locally, and I would like to know what you think of this approach. Do you think it makes sense, or do you see cloud processing as a better solution?