Providing generic third party access is a rare case in business. Usually, data is used by third parties for a specific purpose that requires substantial domain knowledge, in which case an RDF description is probably redundant and useless.
There may be business models based on generic data but those businesses will face a serious bootstrapping problem as they will likely require either a critical mass of generic data providers or consumers, which are mutually dependent.
The bootstrapping will have to be grassroots, as it was for the original "unstructured" web, which succeeded due to the ease of publishing HTML documents, which is much simpler than publishing structured data.
One important use case is publishing data for re-use within an enterprise or other organisation - it doesn't necessarily need to always be public.
Cheers, Michael
Think flexible data mashups (where you don't need to code to the interface of a particular service because they all use same linked data principles).
I have seen people build a simple application and then increase its value by automatically pulling in information from other large source of linked data such as DBPedia and GeoNames.
Fully agree that we need simple and understandable demos that show the value of linked data (assuming for a moment that linked data have business value :).