NVIDIA and Microsoft Launch RTX Spark for Windows AI Agents

NVIDIA and Microsoft announced a co-engineered hardware and software platform for AI agents on Windows PCs, called RTX Spark. The announcement was made at a Microsoft Windows event on October 7, 2026. Jensen Huang, NVIDIA's CEO, said the company was founded because of Windows, highlighting the significance of the partnership.

RTX Spark Hardware and Availability

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. The platform delivers one petaFLOP of FP4 AI performance and supports up to 128 GB of unified memory. It runs the full NVIDIA CUDA platform, the same software stack as DGX Station, and includes 5th-gen Tensor Cores with NVFP4, hardware-accelerated AV1 and 4:2:2 video encode/decode, DLSS, and RTX ray tracing.

Pre-orders for RTX Spark laptops start on October 7, 2026, with shipping on October 16, 2026. RTX Spark compact desktops will go on sale in November 2026. The compact desktop is designed for 24/7 operation, keeping AI agents always on. Microsoft's Surface Laptop Ultra is built around RTX Spark, offering up to 128 GB unified memory and up to one petaFLOP of AI compute.

Acer, ASUS, Dell, HP, Lenovo, Microsoft, MSI, and Gigabyte will ship RTX Spark devices.

Microsoft Execution Containers and Local AI

Microsoft announced general availability of Microsoft Execution Containers (MXC), which provide OS-level infrastructure for AI agents to run safely and persistently in the background. This infrastructure is designed to support the always-on nature of agents on Windows.

NVIDIA stated that RTX Spark can run the Qwen 3.8 Flash Next 125-billion-parameter model locally, without metering or sending data to the cloud. For gamers, NVIDIA claims RTX Spark can run AAA titles at 1440p over 100 fps with DLSS 5, Reflex, and G-SYNC.

DGX Station for Windows

NVIDIA also previewed DGX Station for Windows, the first deskside AI supercomputer for Windows. It uses the GB300 Grace Blackwell Ultra chip, with 748 GB of coherent memory and up to 20 petaFLOPS of FP4 performance. DGX Station for Windows can run trillion-parameter models locally. Previously, DGX Station ran only on Linux; now the same hardware is available to the Windows ecosystem.

Microsoft noted that Fortune 500 companies have standardized on Windows, and DGX Station reduces the need for a separate Linux environment. NVIDIA added that Linux AI toolchains remain accessible via WSL on DGX Station for Windows.