There is a big difference between having interest in a market, and being able to compete in it. There are also many trade-offs.

Nobody has designed yet a GPU architecture that works at all from 500W HPC clusters to sub 1W embedded/IoT systems, much less that works well to be a market leader in all segments. So AFAICT whether this is even possible is an open research problem. If this were possible, there would already be nvidia GPUs at least in some smartphones and IoT devices.