Nvidia and Microsoft bring AI agents and RTX Spark hardware to Windows PCs

Nvidia and Microsoft used a joint event in San Francisco on Wednesday to lay out new hardware and software built to run AI agents directly on Windows computers. Nvidia CEO Jensen Huang and Microsoft CEO Satya Nadella described the plans during a fireside chat hosted by Sriram Krishnan at the Windows AI and Surface event.

Microsoft’s contribution centers on changes to the Windows operating system itself. The company announced general availability of Microsoft Execution Containers, called MXC, which Pavan Davuluri, Microsoft’s executive vice president of Windows and Devices, described as infrastructure that lets AI agents run safely and persistently in the background under operating system control. Davuluri said the containers work alongside Microsoft Security and Agent 365 to let agents on Windows be secured, observed and governed. Nadella said Microsoft needed to make the Windows desktop “the most secure place for agents to execute.”

On the hardware side, Nvidia detailed RTX Spark, a chip and system design meant to bring the company’s AI software stack into Windows laptops and small desktops. Preorders for RTX Spark laptops opened Wednesday, with laptops shipping October 16 and compact desktops going on sale in November, according to Nvidia. RTX Spark combines a Blackwell-generation RTX GPU with up to 6,144 cores and a Grace CPU with up to 20 cores, linked at 600 gigabytes per second, with up to 128GB of unified memory and roughly one petaflop of FP4 AI performance, a measure of computing speed for AI tasks. Nvidia says the chip can run large models, such as Qwen 3.8 Flash Next, directly on the machine without sending data to the cloud. Acer, Asus, Dell, HP, Lenovo, Microsoft, MSI and Gigabyte are building RTX Spark systems. Microsoft’s own Surface Laptop Ultra is built around the chip; Davuluri said its unified memory lets the laptop run AI models too large for conventional machines.

A supercomputer aimed at Windows developers

Nvidia also previewed DGX Station for Windows, a deskside system that had previously run on Linux only. The new version uses Nvidia’s GB300 Grace Blackwell Ultra Desktop Superchip, with 748GB of coherent memory and up to 20 petaflops of FP4 compute, which Nvidia says is enough to run models with up to a trillion parameters locally. Davuluri said the move brings frontier-class AI compute to a desktop running Windows, closing a gap that had pushed some enterprise developers to keep separate machines for AI work and everyday productivity tools. Linux-based AI tools remain available on the new system through Microsoft’s WSL compatibility layer. Nvidia has not announced pricing or a firm release date for DGX Station for Windows and is instead asking developers to sign up to be notified when it becomes available.

Huang framed the announcements as a continuation of Nvidia’s long relationship with Windows, saying the operating system’s existence helped make Nvidia’s GeForce graphics cards possible in the first place. He compared the potential effect of MXC on agent development to what Windows and its DirectX graphics software did for applications in earlier decades.

What happens next is mostly a matter of shipping dates: RTX Spark laptops reach buyers October 16, desktop versions follow in November, and Nvidia has yet to say when DGX Station for Windows will be available to order.