Microsoft and Nvidia have launched the first wave of RTX Spark laptops, and the bigger story is how much Windows had to change to run on them. The Verge AI reports from a San Francisco launch event and a follow-up Nvidia briefing that these machines look like ordinary Windows laptops, but they run on Arm-based chips instead of the x64 architecture Intel and AMD use. RTX Spark pairs that Arm CPU with a Blackwell GPU and unified memory, and it’s built around one idea: running serious AI models locally on a laptop.
That goal forced Microsoft to rework several layers of Windows. Here’s what changed.
7 Things Microsoft Changed for RTX Spark
- Two clicks to a working AI agent New RTX Spark owners get a fresh Get Started experience that lets them “find the agent they want, get started with that agent, install it, and get to that agent in only two clicks,” according to Peter Dawoud, principal program manager at Microsoft. Agents install and update through the Windows Store, so you don’t have to chase down updates or dependencies yourself.
- Three launch agents, with more coming Hermes, Perplexity, and OpenClaw are the first officially supported agents. Microsoft says it’s working to add more, and that likely includes Muse, which Meta recently said is coming to Windows.
- Prism emulation tuned for Nvidia’s cores Apps without a native Arm version run through Microsoft’s Prism emulator. “Emulation isn’t generic. Each platform needs tuning,” Dawoud says. Microsoft optimized x64 translation for RTX Spark’s specific microarchitecture and ships compatibility updates every month.
- Gaming-focused CPU work Part of that tuning targets games. The CPU often sets frame pacing, which decides how smooth a game feels, while the GPU handles rendering. Getting emulation right on the CPU side matters a lot here.
- Unified memory up to 128GB Microsoft’s Surface Laptop Ultra ships with up to 128GB of unified memory, enough to load very large models. Windows holds back a little of that space so the system stays responsive. Microsoft also tuned memory for power, battery life, and multitasking, like gaming while a video renders in the background.
- A GPU memory slider Nvidia’s Windows app includes a slider that controls how much unified memory goes to the GPU. That’s most useful on 24GB and 32GB models, where you’ll want to move more memory toward the GPU for certain apps and games.
- Anti-cheat support out of the box Fortnite, Marvel Rivals, Arc Raiders, Counter-Strike 2, and Dota 2 all work at launch, along with other titles that use anti-cheat engines. Activision’s Ricochet, used in Call of Duty, arrives sometime in 2027, and EA has committed to bringing its Javelin anti-cheat on board too.
How It Stacks Up
Windows on Arm struggled for more than a decade, mostly because emulation was weak and native apps were scarce. Qualcomm’s Snapdragon X laptops changed that a couple of years ago by pulling developers toward Arm-native builds. Nvidia now gets that groundwork for free. What it adds is a much stronger GPU and a clear focus on local AI.
Microsoft and Nvidia say these aren’t gaming laptops. Still, they can run current AAA titles, and the companies spent “many, many, many, many months” making sure RTX Spark delivers a real RTX gaming experience.
The Caveats
The setup isn’t perfect yet. In The Verge’s testing, Hermes mostly took the promised two clicks, but OpenClaw kept reporting that a local gateway hadn’t installed properly. It took a full uninstall and reinstall to fix.
Speed is the bigger issue. Even on RTX Spark, large local models still answer prompts much more slowly than cloud models. The reviewer expects to use local agents mostly for overnight or background jobs. Beyond coding, the real value of local AI on these machines still needs a lot more testing.
The article doesn’t list pricing. Configurations range from 24GB up to 128GB of unified memory.
Why It Matters
What stands out is that Microsoft treats AI agents as core Windows software, installed and updated like any other app. If that two-click flow holds up, local agents stop being a hobbyist project and turn into a standard laptop feature.
The open question is whether local models get fast enough to compete with the cloud for everyday work. For now, RTX Spark looks like the strongest bet yet that they will, and its success will shape how fast the rest of the PC industry follows. You’ll find the full hands-on report at The Verge AI.