AI · Oct 7, 2026
Microsoft releases new Nvidia-chip AI PCs with revamped Windows 11NVIDIA, Microsoft Kick Off a New Beginning for Windows PCs With RTX Spark and AI Agents
Oct 7, 2026, 11:45 AM · NVIDIA Blog

NVIDIA and Microsoft are betting that AI agents will live on the PC, not just in the cloud, and the security plumbing matters more than the petaflops in that bet.
Why it matters
At a Microsoft event in San Francisco on Wednesday, Jensen Huang and Satya Nadella laid out a joint push to run AI agents on Windows PCs. NVIDIA opened preorders for RTX Spark laptops, which pair a Blackwell RTX GPU with a Grace CPU, up to 128GB of unified memory and up to a petaflop of FP4 AI performance. Laptops ship October 16 and compact desktops go on sale in November, from makers including Acer, ASUS, Dell, HP, Lenovo, Microsoft, MSI and Gigabyte. Microsoft's new Surface Laptop Ultra is built around the chip.
Microsoft also announced general availability of Microsoft Execution Containers, or MXC, OS-level infrastructure meant to let agents run safely and persistently in the background under operating system control. NVIDIA previewed DGX Station for Windows, a deskside machine with 748GB of coherent memory that it says can run models up to trillion-parameter scale locally.
From the desk
We think local AI is one of the more quietly important shifts in this industry. Running capable models on a machine someone owns means work that does not get metered per token, and data that does not have to leave the building. NVIDIA says RTX Spark can run large open models without sending data to the cloud. For lawyers, clinicians, small businesses and anyone handling sensitive files, that is a real benefit, not a spec-sheet flex.
But the most consequential announcement on stage was not a chip. It was MXC. Agents that run persistently in the background, acting on a person's files and applications, are a new class of software risk. Nadella's line that the desktop needed to become the most secure place for agents to execute is the right instinct. Whether MXC actually contains misbehaving or hijacked agents is something we will only learn once security researchers get their hands on it.
There is a cost question too. A machine with 128GB of unified memory and a petaflop of AI compute is not a mainstream laptop, and NVIDIA's post does not list prices. Local AI at this level risks becoming a premium tier, while most people stay on metered cloud services. The compact desktops designed for always-on, 24/7 agents also point toward a future where personal computers are constantly doing work on our behalf, which raises the stakes on what those agents are allowed to touch.
DGX Station for Windows addresses a real annoyance: enterprise AI developers have juggled Linux for heavy workloads and Windows for everything else. Folding them together, with Linux tools still available through WSL, is a practical win for the Fortune 500 shops NVIDIA is targeting.
Our read: the hardware is impressive and the local-first direction is good for privacy. I'm watching MXC more closely than the benchmarks, along with real-world pricing once the laptops ship.
Context
NVIDIA's history is tied to Windows; Huang said at the event that without Windows there would be no GeForce. The companies framed this launch as the start of an agent era on the PC, with Microsoft building agent security and governance primitives directly into the operating system.
Who feels it
- Developers
- The same CUDA stack across RTX Spark and DGX Station means models and tools can move between machines without rewrites.
- Enterprises
- DGX Station for Windows removes the need to split heavy AI work onto separate Linux environments.
- Security teams
- MXC becomes the key control point for background agents and will need independent testing.
- Consumers
- Local, unmetered AI is a privacy gain, though the high-memory systems are likely to sit at the premium end.
What to watch
- Pricing and reviews of RTX Spark laptops after the October 16 ship date
- Independent security research on Microsoft Execution Containers
- Availability date and pricing for DGX Station for Windows
- Which local models developers actually run well on 128GB systems