NVIDIA and Microsoft bring AI agents to the Windows PC: control of the desktop is the new battleground

By FayerWayer · October 12, 2026.

NVIDIA and Microsoft have unveiled RTX Spark, a platform for Windows laptops and compact PCs designed to run AI agents locally. According to the article, the announcement was made by Jensen Huang and Satya Nadella in San Francisco. Pre-orders are already open and the laptops will hit the market on October 16. Compact desktops will launch in November.

According to the figures cited, 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 delivers up to 1 petaflop of AI performance at FP4 precision (a 4-bit numeric format that speeds up inference at the cost of precision) and up to 128 GB of unified memory, meaning memory shared between the CPU and GPU. Manufacturers mentioned include Acer, ASUS, Dell, HP, Lenovo, Microsoft, MSI and Gigabyte. With this hardware, models such as Qwen 3.8 Flash Next can be run locally without sending sensitive data to the cloud.

The second piece is Microsoft Execution Containers (MXC). Pavan Davuluri, executive vice president of Windows and Devices, described them as operating-system-level infrastructure that lets agents work persistently and under supervision in the background. They work alongside Microsoft Security and Agent 365. Nadella compared them to what DirectX meant for video game developers.

The third is the preview of the NVIDIA DGX Station for Windows. It integrates the GB300 Grace Blackwell Ultra superchip with 748 GB of coherent memory and up to 20 petaFLOPS of AI performance. According to the information, it will make it possible to train and deploy models of up to one trillion parameters without leaving Windows. WSL (the Linux subsystem within Windows) will remain available when needed. No date or price has been announced.

**Our take.** This story is not just about hardware. It is about where agents live and who sets their rules. Until now, a powerful agent meant the cloud and data leaving your machine. With 128 GB of unified memory, the PC becomes a reasonable place to run serious models. That is good news overall for privacy, cost and sovereignty, and it fits the trend of the open frontier advancing quickly.

But the strategic point is MXC. As models become commoditized, value shifts to the layer of permissions, isolation and auditing between the agent and your data. Microsoft wants to be that layer on the desktop, just as DirectX was the layer for games. Whoever controls it captures recurring value, and with it the power to decide what an agent can see and do. It is the same shift we see in regulation: the control point moves from the model to access and decision-making.

There are important caveats. All performance figures come from the companies, and there are no independent benchmarks. A petaflop at FP4 does not equal sustained performance on real-world tasks. That the desktop is "the safest place" to run agents, as Nadella said, remains a marketing aspiration until it is demonstrated through external audits. And a persistent background agent expands the attack surface rather than shrinking it, no matter how many containers there are.

In the short term, the bet favors NVIDIA and Microsoft, which strengthen their alliance against those relying solely on the cloud. It also challenges Linux as the default environment for AI development, something the DGX Station for Windows is clearly aiming for. In the long term, having capable, private agents under our own control is a necessary piece of a future in which AI works for people and not just for platforms. The key will be ensuring that control truly belongs to the user.