AI firms adopt custom chips while Microsoft finally restores taskbar flexibility in Windows 11

Recent discussions reveal a shift in datacenter AI hardware away from Nvidia’s dominant GPUs toward specialized silicon from companies like Cerebras, whose new CS-4 rack system packs three modular chips with doubled clock speeds, and Marvell, which is now offering custom XPU IP alongside Broadcom for Google’s TPUs. Google’s cloud business is diversifying its accelerator portfolio, while Waymo has built a dedicated machine-learning ASIC to replace FPGA-based solutions in its autonomous-vehicle fleet, aiming for lower latency and higher compute density.

On the software side, Microsoft has introduced a movable taskbar, adjustable Start menu sizes, and a revamped context menu in the Windows 11 Release Preview, responding to user demand after five years of criticism. The update also adds a new Task Manager tab that reports NPU utilization, helping users identify inefficient workloads. These moves reflect broader industry trends of hardware diversification and user-focused refinements in operating-system design.