NVIDIA RTX Spark is a new chip platform — built with Microsoft — that lets Windows PCs run AI agents directly on the device instead of sending every request to the cloud. It pairs a 20-core NVIDIA Grace CPU with a Blackwell RTX GPU, sharing up to 128GB of memory so a single PC can run AI models with roughly 120 billion parameters locally. Laptops and desktops from ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI are expected to ship with it starting this fall.
Key Takeaways
- RTX Spark is NVIDIA and Microsoft’s joint platform for Windows PCs built to run AI agents locally, combining a 20-core Grace CPU and a Blackwell RTX GPU with up to 128GB of shared memory.
- It can run AI models with up to roughly 120 billion parameters and context windows as large as 1 million tokens, entirely on the device, with no cloud round-trip required for every request.
- A new runtime called OpenShell lets you set rules for what local agents may do, and can route a request to an on-device model or a cloud model depending on your privacy settings.
- Laptops and desktops from ASUS, Dell, HP, Lenovo, Microsoft Surface, MSI, Acer, and GIGABYTE are expected starting fall 2026, alongside support from Adobe and more than 100 other software makers.
- RTX Spark is a different, heavier-duty product than Microsoft’s existing Copilot+ PCs, which rely on a smaller chip called an NPU for lighter on-device AI tasks.
What Exactly Is RTX Spark?
RTX Spark isn’t a graphics card you install, and it isn’t a typical laptop chip either. It’s a system-on-a-chip — one piece of silicon that combines a CPU, a GPU, and shared memory, the same basic idea behind the chips in modern smartphones, just scaled up for a Windows PC built to do heavy AI work. NVIDIA designed it together with Microsoft, with MediaTek contributing to the custom CPU design.
Under the hood, RTX Spark has a 20-core NVIDIA Grace CPU (built on the Arm architecture used in phones and some cloud servers, rather than the x86 design in most Windows PCs) and a Blackwell RTX GPU with 6,144 CUDA cores and fifth-generation Tensor Cores. The CPU and GPU share up to 128GB of unified memory — one shared pool of memory, connected by a fast 600GB/s link called NVLink-C2C — instead of the separate system memory and video memory that split a typical laptop’s resources. NVIDIA puts the chip’s AI performance at up to 1 petaflop, a measure of roughly one quadrillion calculations per second.
That unified memory design is the real headline for AI work. Large AI models are memory-hungry, and a traditional laptop has to copy data back and forth between separate CPU and GPU memory pools, which slows things down. Sharing one big pool means RTX Spark can load and run a much larger model locally than a typical gaming laptop’s GPU memory would normally allow.
Why Does Running AI Locally Even Matter?
Most AI assistants today — ChatGPT, Gemini, Copilot — do their actual thinking on a server in a data center somewhere, which means your questions, documents, and files have to travel over the internet to get an answer. Running a model locally keeps that processing on your own machine: nothing has to leave your PC, there’s no round-trip delay to a server, and it can keep working without an internet connection.
NVIDIA and Microsoft are pairing RTX Spark with a new runtime called OpenShell, which lets you define what an AI agent is allowed to do on your PC, and decides whether a given request gets handled by a local model or sent to a cloud model based on your own privacy settings — including the ability to mask personal details before anything does go to the cloud. Two open-source agent projects, OpenClaw and Hermes Agent, are already building new Windows apps on top of this stack, and NVIDIA says future agent experiences will be reachable straight from the Windows taskbar.
What Can an RTX Spark PC Actually Do?
For everyday tasks
Local agents on RTX Spark are designed to work across the apps already open on your PC, not just inside a chatbot window. NVIDIA says they can carry out tasks inside Windows applications, semantically search your own files (finding a document by what it’s about, not just its filename), and even write small plug-ins or mini-apps on the fly.
For creators
NVIDIA is pitching RTX Spark at creative work too: rendering 3D scenes over 90GB in size, editing 12K video, and running local AI image and video generation. Adobe is reportedly rebuilding Premiere and Photoshop around the platform, targeting up to twice the AI and graphics performance, with Substance 3D Painter and Stager running natively.
For gamers
RTX Spark PCs are also built to game, with NVIDIA targeting AAA titles at 1440p and over 100 frames per second with ray tracing, DLSS, and Reflex enabled. DLSS 4.5 Ray Reconstruction is coming to Blender and dozens of games, and RTX Video with 4x Frame Generation is headed to the AI image tool ComfyUI.
Which Laptops and Desktops Will Have It?
NVIDIA says RTX Spark laptops are coming from ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI at launch, with Acer and GIGABYTE models to follow. The laptops are expected to run 14 to 16 inches, as slim as 14mm and as light as three pounds, with tandem OLED displays and NVIDIA G-SYNC for smooth gaming. Compact RTX Spark desktops are coming too, and for businesses, NVIDIA’s DGX Station for Windows packs the same idea into a deskside AI workstation. More than 100 Windows software makers are said to be supporting the platform at launch.
RTX Spark vs. Copilot+ PCs: What’s the Difference?
Microsoft’s existing Copilot+ PCs already run some AI features on the device, using a small, efficient chip called an NPU (Neural Processing Unit) for lighter jobs like background blur in video calls or searching your recent activity. RTX Spark is a much bigger step up: far more memory, far more raw compute, and a GPU built for the kind of AI workloads — running a full 120-billion-parameter model, for instance — that a typical Copilot+ PC’s NPU was never designed to handle. Think of Copilot+ as AI-assisted, and RTX Spark as built to run serious AI workloads entirely on the machine in front of you.
When Can You Buy One, and What Will It Cost?
NVIDIA and Microsoft announced RTX Spark together, with Microsoft following up with developer-focused details on Windows agent features at its Build conference. Laptops and desktops from the launch OEMs are expected in fall 2026. Neither company has published official pricing yet, and early coverage of NVIDIA’s related “Spark” hardware has pointed to a premium price well above a typical laptop — this is a high-end platform aimed at power users and creators rather than an everyday budget machine. Real-world pricing, battery life, and performance will need independent testing once devices actually ship, so treat NVIDIA’s own performance claims as a starting point rather than the final word.
Frequently Asked Questions
What is NVIDIA RTX Spark?
RTX Spark is a system-on-a-chip that NVIDIA built with Microsoft for Windows PCs, combining a 20-core Grace CPU and a Blackwell RTX GPU with up to 128GB of shared memory. It’s designed to run AI agents and large AI models directly on the device instead of relying on the cloud.
How is RTX Spark different from a regular Windows laptop chip?
A typical laptop has a separate CPU and GPU with their own separate memory pools. RTX Spark’s CPU and GPU share one large pool of up to 128GB of unified memory over a fast NVLink-C2C connection, which lets it load and run much bigger AI models locally than a standard laptop GPU can.
Do I need an RTX Spark PC to use AI agents on Windows?
No. Windows already offers cloud-based AI features through Copilot, and Copilot+ PCs handle lighter on-device AI tasks with a smaller NPU chip. RTX Spark is for people who want to run much larger AI models, or AI agents with deeper system access, directly on their own machine.
How much will an RTX Spark PC cost?
NVIDIA and Microsoft haven’t published official pricing. Early coverage of NVIDIA’s related Spark hardware line suggests a premium price well above a typical laptop, positioning RTX Spark as a high-end option for power users and creators rather than a budget AI PC.
Which laptop brands will sell RTX Spark PCs?
ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI are expected to launch RTX Spark laptops first, with Acer and GIGABYTE models following. NVIDIA also offers a DGX Station for Windows for business and enterprise use.
RTX Spark is part of a broader shift toward agentic AI that we’ve been tracking closely — see our rundowns on agentic AI browsers and the best AI agents of 2026 for how the software side of this shift is playing out, and our roundup of recent AI model and tool updates for everything else that’s shipped this quarter.
Sources: NVIDIA Newsroom, Tom’s Hardware.
