Nvidia and Microsoft unveil RTX Spark, Windows PCs that run 125B-parameter models locally

In a fireside chat in San Francisco, Jensen Huang and Satya Nadella traced decades of collaboration and introduced the next phase: AI agents that run on the personal computer without cloud dependence. Huang noted that without Windows there would have been no GeForce, and that foundation is now being adapted for the agent era. Nadella praised the consistency of Huang's vision over the years, saying the current moment is a direct result of that long-term view.

The infrastructure behind the agents

Microsoft announced general availability of Microsoft Execution Containers (MXC), an OS-level framework that lets agents run in the background securely and under control. Pavan Davuluri, corporate vice president for Windows and Devices, said the goal is to make the desktop the safest place to run agents. Huang responded that MXC is poised to revolutionize how agents are built and deployed, much as DirectX once transformed application development.

Hardware that puts a petaflop in a backpack

RTX Spark pairs a Blackwell-series GPU with up to 6,144 cores and a Grace CPU with up to 20 cores, linked by a 600 GB/s interconnect. The result: up to one petaflop of FP4 performance and up to 128 GB of unified memory. According to the companies, that is enough to run a model such as Qwen 3.8 Flash Next — a 125-billion-parameter model with 51 billion n-gram — locally, with no usage limits and no data sent to the cloud. The system runs the full Nvidia CUDA stack, so developers can port models without rewriting code.

Laptops available for pre-order, desktops in November

Laptop pre-orders opened immediately, with availability starting October 16; compact desktops will follow in November. Microsoft's Surface Laptop Ultra was designed around RTX Spark and showcases the maximum specification. The OEM roster includes Acer, ASUS, Dell, HP, Lenovo, MSI and Gigabyte, spanning thin-and-light designs to compact workstations built for agents that operate continuously.