THE CRUNCH
Microsoft has done substantial behind-the-scenes work to prepare Windows for Nvidia's RTX Spark laptops, whose chips are Arm-based rather than the x64 architecture used by AMD and Intel. The platform pairs an Arm CPU with a Blackwell GPU and unified memory so large AI models can run locally, and Microsoft's biggest challenge was making that feel effortless. New RTX Spark machines get a Get Started experience that lets users find, install and open an AI agent in roughly two clicks, with agents distributed and updated through the Windows Store. Hermes, Perplexity and OpenClaw are the first three officially supported agents, and Microsoft says more are on the way.
The Verge's hands-on testing suggests the flow mostly works but is not flawless: setting up Hermes took about two clicks, while OpenClaw repeatedly complained about a misinstalled local gateway and needed a reinstall. There is also a practical caveat. Large local models respond noticeably more slowly than cloud models, so the reviewer expects to use them mainly for overnight or background tasks.
Emulation is the other piece of engineering. RTX Spark laptops rely on Microsoft's Prism emulator for apps without native Arm versions, and the tuning is chip-specific. "Emulation isn't generic. Each platform needs tuning," said Peter Dawoud, principal program manager at Microsoft, noting the x64 translation was optimised for RTX Spark's microarchitecture. Prism gets compatibility updates monthly, and gaming has received particular attention because the CPU governs frame pacing.
Memory has been tuned too. Microsoft's Surface Laptop Ultra can ship with up to 128GB of unified memory, enough to load very large models, though some headroom is reserved to keep the system responsive. A slider in Nvidia's Windows app lets owners of 24GB or 32GB machines manually shift memory between the GPU and the system for specific apps or games.