At GTC Taipei on June 1, 2026, NVIDIA unveiled the RTX Spark a superchip built from the ground up to power the next generation of Windows AI PCs. Jensen Huang’s line at the keynote said it best: “For forty years, you launched apps. Click. Type. With RTX Spark and Microsoft Windows, you ask and the PC does the work.”
That is not marketing fluff. The hardware backs it up.
Table of contents
What Is the NVIDIA RTX Spark Superchip?
The RTX Spark is a single chip that pairs a Blackwell RTX GPU (6,144 CUDA cores, fifth-generation Tensor Cores with FP4 precision) with a 20-core NVIDIA Grace CPU, all connected through NVIDIA’s NVLink-C2C chip-to-chip interconnect. The CPU design was built in collaboration with MediaTek, contributing to what NVIDIA claims is the best power efficiency in any PC chip to date.
Think of it as Grace Blackwell the same architecture behind NVIDIA’s data centre supercomputers shrunk into something that fits inside a laptop bag.
Key specs at a glance:
| Specification | Details |
|---|---|
| GPU Architecture | NVIDIA Blackwell RTX |
| CUDA Cores | 6,144 |
| Tensor Core Generation | 5th Gen (FP4 Precision) |
| CPU | 20-core NVIDIA Grace (MediaTek collaboration) |
| Interconnect | NVIDIA NVLink-C2C |
| AI Performance | Up to 1 Petaflop |
| Unified Memory | Up to 128GB |
| Process Node | TSMC 3nm |
| Transistor Count | 70 Billion |
1 Petaflop. 128GB Memory. In a Laptop.
Let’s sit with those numbers for a second. 1 petaflop of AI compute and 128GB of unified memory on a device you can carry on a flight. For context, your phone has somewhere between 8GB and 16GB of RAM. The RTX Spark, in its top config, holds the equivalent of several large AI models entirely in memory with zero cloud dependency.
That means 120-billion-parameter LLMs run locally. With up to 1 million tokens of context. No subscription. No latency. No one snooping on your prompts.
The power draw scales from single-digit watts all the way up to 80W which explains how NVIDIA and its hardware partners are promising all-day battery life in machines as slim as 14mm thick and as light as 3 pounds.
Personal AI Agents: The Actual Point of All This
The RTX Spark is not just a faster chip for existing workflows. It is purpose-built for on-device AI agents software that can reason, execute multi-step tasks, search your local files, write code, and work across applications without sending your data to a server.
NVIDIA and Microsoft are collaborating to bring this vision to Windows natively. That includes:
- New Windows security primitives identity, containment, and policy controls for agents running on your primary device
- NVIDIA OpenShell a runtime that lets you define what agents can and cannot access, routes queries intelligently between local models and cloud, and can mask personal information before anything leaves your machine
Agent platforms like Hermes Agent (Nous Research) and OpenClaw are already building native Windows apps on this foundation.
Dillon Rolnick, CEO of Nous Research, put it plainly: “You realize you’re buying a full-fledged assistant, not a typical laptop.”
What Can RTX Spark Actually Do?
Beyond agents, the full CUDA and RTX ecosystem runs natively. Here is what the hardware unlocks:
| Task | RTX Spark Capability |
|---|---|
| 3D Rendering | 90GB+ scenes via OptiX and DLSS |
| Video Editing | 12K 4:2:2 with Blackwell decoder |
| AI Video Generation | 4K via ComfyUI with 4x Frame Generation |
| LLM Inference | 120B parameter models, 1M token context |
| Gaming | 1440p, 100+ fps, ray tracing + DLSS + Reflex |
| Adobe Photoshop AI | Up to 2x faster with TensorRT |
| Adobe Premiere AI | Real-time editing and color correction |
Adobe is rearchitecting both Photoshop and Premiere from scratch for RTX Spark. Firefly-powered Generative Fill, GPU-accelerated compositing, live filters, HDR all of it rebuilt to tap the Blackwell GPU and TensorRT pipeline. Expect up to 2x faster performance across creative workflows when these updates roll out alongside device availability.
Blackmagic Design, OTOY Octane, ComfyUI, llama.cpp, Blender, and over 100 software providers are already on board. So are game studios including KRAFTON, Riot Games, NetEase, Remedy Entertainment, and XBOX.
Who Is Making RTX Spark Laptops and Desktops?
Major PC manufacturers are releasing RTX Spark-powered devices in fall 2026:
| Manufacturer | Product |
|---|---|
| Dell | XPS 16 Creator Edition |
| Microsoft | Surface Laptop Ultra |
| ASUS | TBA (fall 2026) |
| HP | OmniBook (one of the thinnest RTX Spark laptops) |
| Lenovo | TBA (fall 2026) |
| MSI | Compact desktop and laptop (TBA) |
| Acer | To follow post-launch |
| GIGABYTE | To follow post-launch |
Designs will be available in 14-inch to 16-inch screen sizes, with precision-machined aluminum chassis, tandem OLED displays with G-SYNC technology, and all-day battery life. Compact desktop variants are also in development for creators who want the same power at their desk.
What About India? Can You Buy One?
RTX Spark laptops are set to launch globally in fall 2026. Amazon.in listings are not live yet at the time of this writing, but you can track availability here:
Note: The above link will go live with products once devices ship. Bookmark it and check back in October-November 2026.
In the meantime, if you are in the market for a high-performance AI laptop with an NVIDIA GPU right now, current-generation options are available:
Should You Wait for RTX Spark?
If you are a video editor, 3D artist, AI developer, or serious gamer who wants a single portable device that handles it all yes, waiting makes sense. The combination of 128GB unified memory, 1 petaflop of AI compute, and native Windows agent support is a genuinely new category. Nothing shipping today comes close on paper.
If you need a laptop right now for everyday productivity or gaming, current RTX 40 and RTX 50 series laptops are solid performers. But RTX Spark is in a different class.
The Bottom Line
The PC has had the same interaction model for four decades. You open software. You click. You type. RTX Spark changes the premise. You ask and a local AI agent with access to 128GB of memory and 1 petaflop of compute gets it done privately, on your device, without pinging a server.
That is a real shift. NVIDIA, Microsoft, Adobe, and a large chunk of the software industry are all betting on it simultaneously. Fall 2026 cannot come fast enough.
Track RTX Spark laptop availability : Click here