I have no doubt that they would prefer selling an end user tool rather than an API. There's much more money to be made in the first case.

Even if these interface wrappers can seem easy it also comes with time consuming integrations, custom features, marketing and sales efforts.

My bet is however that they will lose this bet because their efforts would be spread out too thinly.

As long as the open source or competitor models are close enough to the GPT variants, this won't work.