NVIDIA and Microsoft Reboot the Windows PC: RTX Spark Laptops Open for Preorder, MXC Gives AI Agents a Secure Home

What was announced

On Wednesday, October 7, Microsoft held a Windows AI and Surface event at Dogpatch Studios in San Francisco. NVIDIA founder and CEO Jensen Huang and Microsoft CEO Satya Nadella took part in a fireside chat hosted by Sriram Krishnan, a former senior White House policy advisor on AI.

The two companies said they are co-engineering hardware and software so that AI agents can run on Windows PCs. Huang traced NVIDIA's decades-long relationship with Windows and said, "If not for Windows there would be no GeForce." He argued that once an agent lives on your own computer, it becomes your personal assistant, and that the PC is the ultimate tool. Nadella credited Huang with a consistent vision held as a secular trend, and said that is what brings the industry to this moment.

Microsoft's piece: MXC reaches general availability

Nadella opened the event. Pavan Davuluri, Microsoft's EVP of Windows and Devices, then walked through the Windows updates designed for the agentic era. The central item is Microsoft Execution Containers, or MXC, now generally available. It is OS-level infrastructure that lets agents run safely and persistently in the background, under operating system control. Davuluri said Microsoft is building these primitives directly into Windows so agents can be secured, observed and governed, and that MXC together with Microsoft Security and Agent 365 unlocks the full potential of agents on Windows. Nadella added that the desktop needed to become the most secure place for agents to execute. Huang compared the change to history: just as Windows and DirectX changed how applications were built, he said, MXC will change how agents are built and deployed.

In engineering terms, the security boundary moves up from each application's own sandbox into the operating system. A long-running agent reads files, calls applications and uses the network, so its risk surface is much larger than a single chat request. Putting isolation, audit and policy at the system layer gives enterprise IT a chance to govern agents with tools it already knows.

NVIDIA's piece: RTX Spark

On the hardware side, the lead product is RTX Spark. NVIDIA says it puts the full NVIDIA AI stack into Windows laptops and compact desktops. Laptop preorders opened on the day of the event, with availability on October 16. Compact desktops go on sale in November. RTX Spark combines a Blackwell RTX GPU with up to 6,144 cores and a Grace CPU with up to 20 cores, connected at 600 GB/s. It offers one petaflop of FP4 AI performance and up to 128GB of unified memory. NVIDIA says it can run models such as Qwen 3.8 Flash Next locally, described in the source as a 125B model with 51B n-gram that matches the intelligence of many cloud models, unmetered and without sending data to the cloud. These figures are NVIDIA's own claims. Independent testing has not yet confirmed them.

On the software side, RTX Spark runs the full CUDA platform. Developers can move models, tools and workflows from RTX Spark to DGX Station without rewriting. Creators get fifth-generation Tensor Cores with NVFP4 support, hardware-accelerated AV1, and 4:2:2 video encode and decode, plus DLSS and RTX ray tracing. For gamers, NVIDIA says AAA titles can exceed 100 frames per second at 1440p with DLSS 5, Reflex and G-SYNC. The compact desktop version is built for 24/7 operation, a dedicated local AI system that keeps agents running continuously.

The device ecosystem

Davuluri introduced Surface Laptop Ultra, built around RTX Spark.

He said that with up to 128GB of unified memory and up to a petaflop of AI compute, it can run models that do not fit on a traditional machine. Systems are coming from Acer, ASUS, Dell, HP, Lenovo, Microsoft, MSI and Gigabyte, with designs from slim laptops to compact desktops made for always-on agents.

DGX Station for Windows

The event also previewed NVIDIA DGX Station for Windows. NVIDIA calls it the first deskside AI supercomputer to bring GB300 Grace Blackwell-class AI infrastructure into the Windows ecosystem. It runs on the GB300 Grace Blackwell Ultra Desktop Superchip, with 748GB of coherent memory and up to 20 petaFLOPS of FP4 AI compute, enough to run models up to a trillion parameters locally. Davuluri said this brings capabilities that once required renting a cluster to a deskside machine.

Until now, DGX Station ran on Linux. Because the vast majority of Fortune 500 companies standardize on Windows, developers either moved to Linux for AI compute or stayed on Windows with limited hardware for heavy model work. Maintaining two environments cost time and resources. The new version lets teams build and run always-on agents that connect directly to the Windows applications and infrastructure they already use, and fine-tune and run inference on large models without leaving their main machine. Linux AI toolchains remain available through WSL. The product is a preview for now, and NVIDIA offers a sign-up for availability news.

Why it matters, and what to watch

For developers, one CUDA stack that stretches from a laptop to a deskside supercomputer lowers migration cost. For enterprise IT, MXC and Agent 365 offer a single governance entry point, and that often decides whether agents reach production more than raw compute does. For PC makers, eight brands joining at once shows that local AI is becoming a selling point.

The open questions are clear. First, the specifications and performance numbers are vendor claims without third-party tests. Second, the disclosed material gives no price, power or battery figures. Third, whether a local model can truly replace the cloud depends on the task. Fourth, always-on agents raise permission and audit problems that MXC must prove it can handle in real enterprise settings. The first shipments on October 16 and the compact desktops in November will be the first checkpoints for this story.