NVIDIA and Microsoft Launch RTX Spark and Windows AI Agent Infrastructure

NVIDIA and Microsoft co-engineered hardware and software to bring AI agents to Windows PCs, unveiling the RTX Spark platform and Microsoft Execution Containers (MXC) at a San Francisco event. Jensen Huang and Satya Nadella emphasized that Windows is becoming the secure foundation for the agentic era, with MXC providing OS-level infrastructure to run agents safely and persistently in the background.

RTX Spark Hardware and Availability

RTX Spark integrates the full NVIDIA AI stack into Windows laptops and compact desktops, combining an NVIDIA Blackwell RTX GPU with up to 6,144 cores and an up to 20-core NVIDIA Grace CPU connected at 600 GB/s. The system delivers one petaflop of FP4 AI performance and up to 128GB of unified memory, enabling local execution of models like Qwen 3.8 Flash Next without cloud dependency.

• Laptops: Preorders open today, with availability starting October 16.
• Compact Desktops: Available for sale in November, designed for 24/7 always-on agent operation.
• Partners: Systems will come from Acer, ASUS, Dell, HP, Lenovo, Microsoft, MSI, and Gigabyte.

Surface Laptop Ultra and Developer Capabilities

Microsoft introduced the Surface Laptop Ultra, built around RTX Spark, featuring up to 128 gigs of unified memory and up to a petaflop of AI compute. This allows developers to run models that do not fit on traditional machines while maintaining compatibility with the NVIDIA CUDA platform.

• Developers: Can move models and workflows without rewriting code, using the same stack from RTX Spark to DGX Station.
• Creators: Benefit from 5th-generation Tensor Cores with NVFP4 support, hardware-accelerated AV1, and DLSS.
• Gamers: Can run AAA titles at 1440p over 100 frames per second using DLSS 5 and Reflex.

DGX Station for Windows

The event previewed NVIDIA DGX Station for Windows, the first deskside AI supercomputer bringing GB300 Grace Blackwell-class infrastructure to the Windows ecosystem. This addresses the historical split between Linux for AI workloads and Windows for productivity tools.

• Performance: Delivers 748GB of coherent memory and up to 20 petaFLOPS of FP4 AI compute.
• Capability: Supports models up to a trillion-parameter scale locally.
• Integration: Allows enterprise developers to build and run always-on AI agents directly within Windows applications, while retaining access to Linux toolchains via WSL.