NVIDIA and Microsoft Launch RTX Spark and DGX Station for Windows

NVIDIA and Microsoft announced co-engineered hardware and software for AI agents on Windows PCs on 7 October 2026. The collaboration introduces the RTX Spark platform and the DGX Station for Windows to provide local compute for models that previously required cloud infrastructure. If you currently maintain separate Linux and Windows environments for AI workloads, this announcement allows you to unify your workflow on a single operating system.

The Shift to Agentic Computing on Windows

The collaboration aims to usher in the agentic era—a period where AI agents run locally on personal computers as persistent assistants. Jensen Huang and Satya Nadella outlined a strategy where hardware and software are co-engineered to ensure these agents can run safely and persistently in the background. This shift moves AI execution from centralized cloud clusters to the local desktop, prioritizing privacy and unmetered usage.

Microsoft and NVIDIA

Microsoft and NVIDIA are focusing on local execution to reduce the need for high-latency cloud calls. By bringing the full NVIDIA AI stack into the Windows ecosystem, the companies aim to turn the personal computer into a primary host for personal assistants. This approach relies on new system primitives to manage the lifecycle of these background processes.

RTX Spark Platform Specifications

NVIDIA’s RTX Spark platform provides the hardware foundation for local AI, combining an NVIDIA Blackwell RTX GPU with up to 6,144 cores and an up to 20-core NVIDIA Grace CPU. The components are connected via a 600 GB/s interconnect to facilitate high-speed data movement between the processor and the graphics unit. This configuration is designed to handle the high-throughput requirements of modern LLMs.

The hardware delivers 1 petaflop of FP4

The hardware delivers 1 petaflop of FP4—a specific precision format used for AI performance calculations—and supports up to 128GB of unified memory. (unified memory: a memory architecture where the central processing unit and graphics processing unit share a single pool of memory.) Unified memory—a memory architecture where the CPU and GPU share a single pool of memory—allows the platform to run large models that would otherwise exceed the limits of traditional consumer hardware. The architecture also includes 5th-generation Tensor Cores with NVFP4 support, hardware-accelerated AV1, and 4:2:2 video encode and decode capabilities.

Surface Laptop Ultra and the 125B Model Threshold

Microsoft’s Surface Laptop Ultra is built specifically around the RTX Spark hardware to push the boundaries of what a laptop can process. Pavan Davuluri stated that the device features up to 128GB of unified memory, which allows it to run models such as Qwen 3.8 Flash Next. This 125B model with a 51B n-gram provides intelligence comparable to many cloud models while running unmetered on the local machine.

DGX Station for Windows: Frontier Compute for the Enterprise

For enterprise-grade requirements, NVIDIA previewed the DGX Station for Windows, which is the first deskside AI supercomputer to bring GB300 Grace Blackwell-class infrastructure into the Windows environment. The system features a GB300 Grace Blackwell Ultra Desktop Superchip, providing 748GB of coherent memory—memory that is accessible by multiple components with consistent data views—and up to 20 petaFLOPS of FP4 AI compute. This provides the hardware necessary to run trillion-parameter scale models directly on a desk-side machine.

Microsoft Execution Containers (MXC) Security Infrastructure

The software foundation for this hardware is Microsoft Execution Containers (MXC). (Microsoft Execution Containers (MXC): operating system level infrastructure primitives that allow agents to run safely and be governed by the operating system.) These are OS-level infrastructure primitives that allow agents to run safely, be observed, and be governed by the operating system. Jensen Huang stated that while Windows and DirectX revolutionized how applications were built, MXC is going to revolutionize how agents are built and deployed. These containers provide the security layer necessary for persistent background agents to interact with the OS without compromising system integrity.

Availability and Ecosystem Support

Laptop preorders for the RTX Spark platform opened on 7 October 2026, with availability scheduled for 16 October 2026. Systems will be available from manufacturers including Acer, ASUS, Dell, HP, Lenovo, Microsoft, MSI, and Gigabyte. The compact desktop configurations, designed for 24/7 operation of AI agents, will be available for sale in November.

Enterprise AI developers using Windows

Enterprise AI developers who currently manage Linux workstations for heavy AI workloads can now run frontier-class models on their primary Windows machines. The DGX Station for Windows allows these users to move models, tools, and workflows without rewriting code, as it runs the full NVIDIA AI stack. These developers should evaluate the DGX Station for Windows as a replacement for the current Linux-only DGX Station to reduce the need to maintain dual environments.

Local AI researchers running 125B models

Local AI researchers who compare local execution against cloud GPU instances can now run the Qwen 3.8 Flash Next model locally. The RTX Spark hardware provides the necessary 128GB of unified memory to handle the 125B parameter model without sending data to the cloud. These researchers should check the pre-order status of the Surface Laptop Ultra to secure a machine with the required memory capacity.

Content creators using AV1

Content creators who rely on standard workstations for high-resolution video production can leverage the RTX Spark superchip in a compact desktop. The hardware includes 5th-generation Tensor Cores and hardware-accelerated AV1, which are standard for modern video production pipelines. These creators should verify that their current video tools are compatible with the new NVFP4 support on RTX Spark hardware to ensure maximum performance.

Who this is not for: Mobile device owners

Mobile device owners can ignore this news, as the RTX Spark and DGX Station for Windows are strictly for desktop and laptop form factors. Users seeking mobile AI hardware should search for mobile-specific hardware announcements from NVIDIA or Microsoft. Software-only AI users should also skip this, as the focus is on hardware-level primitives and system-level containers; these users should look for software-only LLM updates or API documentation instead.

Enterprise deployment walkthrough for Yusuf

Yusuf runs an AI development team at a corporation that uses Windows for all productivity tools but currently uses a remote Linux cluster for heavy model training. By purchasing a DGX Station for Windows, Yusuf can connect the device to his existing infrastructure and run trillion-parameter models locally. Yusuf no longer has to switch environments or move data between Linux and Windows. If Yusuf only needed to run smaller models like Qwen 3.8 Flash Next, he would have instead purchased a Surface Laptop Ultra with RTX Spark to keep the workload mobile.

Practical applications for RTX Spark and DGX Station

The hardware and software integration supports a variety of use cases ranging from hobbyist experimentation to enterprise-grade infrastructure.

What this replaces: The RTX Spark and DGX Station for Windows reduce

The RTX Spark and DGX Station for Windows reduce the need for cloud-based inference for large models. By running models locally, users can access unmetered usage and reduce the need to send sensitive data to the cloud. Additionally, the DGX Station for Windows reduces the need to maintain dual Linux and Windows environments, allowing developers to use their standard Windows productivity tools alongside heavy-duty AI workloads.

What is not yet known: NVIDIA and Microsoft did not specify

NVIDIA and Microsoft did not specify who made the final decision to launch the DGX Station for Windows or the specific internal reasoning for the timing. The source also did not clarify how the RTX Spark hardware interacts with WSL for existing Linux toolchains. Finally, the source did not provide independent verification of the 1 petaflop FP4 performance claim, which remains an NVIDIA-provided figure.

The ultimate test of this ecosystem

While NVIDIA and Microsoft focus on this integrated Windows ecosystem, AMD offers a competing path for local AI through its Strix Halo and Gorgon Halo architectures. However, NVIDIA’s partnership with Microsoft provides a unique advantage in the form of MXC security primitives. The ultimate test of this ecosystem will be whether the MXC infrastructure can maintain the security of persistent agents while allowing the high-speed memory access required by 125B parameter models.

This analysis is based on NVIDIA’s NVIDIA, Microsoft Kick Off a New Beginning for Windows PCs with RTX Spark and AI Agents.