The answer is contained within the Qualcomm White Paper which...

The answer is contained within the Qualcomm White Paper which you can access from the below link. In short their NPU is the Hexagon which they have been developing since 2015 a bit before the AKIDA 2.0. It is an accelerator and it is not programmable. Though they constantly refer across their literature to being ahead of their competitors in performance in the White Paper they restrict themselves to Intel's X86 as the competitor product they out perform.

So the simple answer is they are using Hexagon because they have invested probably very large sums of money to develop it since 2015 and it is better than what Intel has in the form of X86.

They do admit in the White Paper that future demand for GenAi on device will demand something even more powerful than Hexagon can deliver. Hexagon can deliver 45 TOPS currently compared with AKIDA 2.0 TENNS/ViT which can deliver 131 TOPS so who knows for the future.

The next AKIDA 3.0 is being specifically targeted at GenAi so it may deliver more than 131 TOPS.

https://www.qualcomm.com/content/dam/qcomm-martech/dm-assets/documents/Unlocking-on-device-generative-AI-with-an-NPU-and-heterogeneous-computing.pdf

My opinion only DYOR

Fact Finder