A Register podcast episode discusses two parallel shifts: AI vendors increasingly moving beyond Nvidia's GPUs towards alternative datacentre silicon, and Microsoft finally loosening long-standing restrictions in Windows 11. Register Systems Editor Tobias Mann explains why companies are turning to chipmakers such as Cerebras and Marvell to handle inference workloads, and why Waymo is developing custom silicon for its robotaxis rather than relying on off-the-shelf Intel chips. The trend signals that GPUs, while still dominant, no longer have exclusive control of datacentre compute as firms seek faster or more efficient alternatives for AI inference.

On the hardware side, Cerebras is highlighted for its unusually large, dinner-plate-sized chips packing 44GB of fast SRAM, which sidestep constrained high-bandwidth memory supplies and allow rapid token serving; the company has recently partnered with AWS and, last month, AMD to offload compute-heavy inference tasks. Separately, Register Microsoft Ecosystem Reporter Richard Speed discusses Windows 11 changes, including the ability to move the taskbar (five years after its original placement was fixed), greater control over context menus, and improved Task Manager visibility into neural processing unit (NPU) activity.