Microsoft’s AI accelerator was never designed to compete with Nvidia. It was designed to run Microsoft’s own inference workloads across its own fleet, and that scoping decision — made at the outset of the program — explains why the company has no intention of selling the part to anyone else.