Sinno was Amazon’s AI chip director, responsible for building Trainium and Inferentia which were custom ASICs that gave Amazon Web Services a credible alternative to Nvidia's pricey accelerators. Now he’s jumped ship to ARM, where he’s expected to lead the company’s push beyond just flogging IP and into the murky, cash-burning world of CPU manufacturing.

This is the clearest signal yet that ARM, under CEO Rene Haas, is gunning for a bigger slice of the AI silicon pie. Haas recently confirmed the firm was working on “full-end” solutions, a not-so-subtle hint that it plans to stop licensing all its best ideas to rivals and start cashing in directly.

ARM chips already power half of NVIDIA’s datacentre units, thanks to the Grace CPU architecture, and the company sees an opening to push harder into server racks. With Sinno on board and SoftBank happy to throw money at high-risk bets, ARM looks set to unveil its own silicon sooner rather than later.

The real challenge will be cracking a market still ruled by Intel and AMD who know a thing or two about defending their turf. But ARM isn’t starting from scratch. With NVIDIA already selling Grace CPUs using ARM cores, its architecture is proven. The question now is whether ARM can build and sell chips at scale without tripping over the very licensing model that made it rich in the first place.