I think you are misunderstanding the article.

Windows for ARM is designed specifically for ARM, and it has the translation layer. The translation layer effectively allows it to function as if it’s running an x86 Windows install off the bat by offering the ability to run x86 applications on the ARM hardware. It’s not actually running an x86 OS.

The chipset is very powerful but it doesn’t require additional hardware to achieve this translation. The additional processing power built into these chips are NPUs (Neural Processing Units) which are designed to more effectively run ML/AI/LLM workloads. The translation system just works on the normal raw processing power of the machine, just the same as the M-series Macs.