Nvidia unveils RTX Spark Superchip for laptops and desktop PCs at Computex 2026 – new platform promises to turn Windows into an agentic AI OS with Arm CPU, Blackwell GPU, and 128GB unified memory

News
By published

Over 30 laptops and 10 desktops coming this fall with "the most efficent platform ever built"

A representation of the RTX Spark platform
(Image credit: Nvidia)

Nvidia is transforming Windows into an agentic AI platform at Computex 2026. During his keynote, CEO Jensen Huang revealed the RTX Spark: a Windows on Arm platform for laptops powered by the company's RTX Spark Superchip. The company boldly claims that this platform is “the most efficient ever built,” and it’s throwing its full weight into building a first-class Windows on Arm experience for what it envisions as the next frontier of personal computing.

Nvidia says AI agents are already shaping a new mode of interaction with PCs. Instead of relying on the same mouse and keyboard inputs that have defined personal computing for 40 years, the company sees AI agents as a new interface that will let users command their systems and find information with natural language.

And once those agents have their marching orders, they’ll need to set goals, call tools, evaluate the quality of their work, and refine it, potentially using local and cloud AI models to achieve those ends. Agents might also continue working on long-running tasks even when the user is away from their system or overnight. That all requires powerful, efficient hardware and lots of local memory.

Latest Videos From

To power all this AI reasoning in the new era of computing it envisions, Nvidia is unleashing the RTX Spark Superchip, a Windows on Arm platform more powerful and capable any other on the market, and a roadmap for the Spark family outlining the next three generations of technology.

At full strength, this chip offers up to 20 Arm CPU cores, a Blackwell GPU with 6,144 CUDA cores, 128GB of LPDDR5X RAM, and up to 300 GB/s of memory bandwidth. That powerful CPU and GPU, connected over NVLink C2C, and the large memory pool give AI agents and 120-billion-parameter models plenty of power and space for long-running tasks with context lengths stretching to a million tokens, according to Nvidia.

RTX Spark will power high-end laptops from partners including Dell, HP, Lenovo, Asus, and MSI -- and notably, a new Surface Ultra laptop from Microsoft. Nvidia says it’s worked with those partners to create “the most extraordinary laptops [they’ve] ever built,” with tandem OLED G-Sync displays, “all-day” battery life, premium aluminum chassis with large glass touchpads.

Nvidia says that the incredible efficiency of the RTX Spark platform “transforms what a high-performance laptop looks like,” so buyers in the promised agentic AI age will no longer need to choose between high performance or thin chassis with long battery life. RTX Spark PCs will also deliver similar performance whether plugged in or unplugged, as we’ve come to expect from other Windows on Arm and Apple Silicon-powered systems.

RTX Spark will also bring this agentic Windows on Arm experience to compact, powerful desktops in the vein of the DGX Spark. In total, Nvidia expects over 30 laptops and “10 or so” desktops to lead the charge when the platform launches.

In addition to its agentic AI chops, Nvidia positions the RTX Spark Superchip as a creative and gaming powerhouse. The company promises the platform is good for “100 FPS 1440p gaming,” potentially enabled by DLSS 4.5 upscaling and Multi Frame Generation. And its large memory pool means creators can work with massive 3D projects and ultra-high resolution video files like 12K 4:2:2 content without running out of resources.

To further the RTX Spark platform’s creative chops, Nvidia says it’s working with Adobe to rebuild the core of Photoshop, transforming it into a 100% GPU-accelerated application for RTX Spark. Those updates will enable new generative workflows, high-dynamic-range editing, and more natural brushing for artists.

And Premiere is also getting a core overhaul that’s claimed to enable faster and more sophisticated AI workflows, editing, color, and effects. Adobe will also expose Model Context Protocol controls for AI agents to harness its products.

In partnership with Microsoft, Nvidia is also helping to transform Windows into an agentic platform with its OpenShell framework and a “new set of security primitives” that form a set of guardrails, ensuring that local agents and models only have access to the tools and data the user grants them access to. Nvidia says Microsoft will reveal more details of this agentic AI transformation at its upcoming Build conference.

RTX Spark systems will begin arriving in the fall of 2026, and we can't wait to dig into them to see whether Nvidia's backing will truly transform the Windows on Arm experience for the agentic AI era - or just make for a really great PC platform. Stay tuned.

Google Preferred Source

Follow Tom's Hardware on Google News, or add us as a preferred source, to get our latest news, analysis, & reviews in your feeds.

Jeffrey Kampman
Jeffrey Kampman
Senior Analyst, Graphics

As the Senior Analyst, Graphics at Tom's Hardware, Jeff Kampman covers everything that has to do with graphics cards, gaming performance, and more. From integrated graphics processors to discrete graphics cards to the hyperscale installations powering our AI future, if it's got a GPU in it, Jeff is on it. 

51 Comments Comment from the forums
  • Notton
    I suppose the follow up products will be called Melt, Smoke, and Fire?
    Reply
  • dva852
    As reported by Videocardz, RTX Spark system won't have dGPU capability. This affirms that gaming, or at least high-end gaming, isn't a focus. Nvidia will likely target "good enough" gaming for the mainstream, and push hard on agentic AI as the main selling point.

    That won't happen until the OS (Windows) has viable agentic AI capability. That'll likely come with Windows 12 next year. From now till then, will be time for the hardware to ramp-up, and for the software stack to cohere.

    For AI enthusiasts, Win12 isn't really in the equation. Spark just needs to get to a "good enough" price point (read: without the launch premium, and preferably with better RAM pricing).
    Reply
  • thestryker
    Seems like nvidia is pushing this like it's an Apple product: light on details and highlighting the premium design. I'm really curious who exactly this is for because the GPU is clearly power limited if they're saying similar performance to a laptop 5070 which has 25% fewer cores. They've also not promised anything with regards to Linux drivers which means Windows first.

    From what I'm seeing they're going to be selling very expensive devices that do a bit of everything. The only thing they're really offering is the higher memory pool with a more performant GPU than is in Strix Halo. Outside of this they'll be outclassed by many devices that are much cheaper.
    Reply
  • dva852
    thestryker said:
    Seems like nvidia is pushing this like it's an Apple product: light on details and highlighting the premium design.
    I saw the keynote. There isn't much detail, because OEMs will have their own announcements for their particular line-ups. AFAIK, Nvidia itself isn't selling reference devices.
    thestryker said:
    I'm really curious who exactly this is for because the GPU is clearly power limited if they're saying similar performance to a laptop 5070 which has 25% fewer cores.
    You can see the main selling points here, https://nvidia.com/en-us/products/rtx-spark
    . Built for agents and AI: This one is obvious. AI enthusiasts are onboard. It isn't a mainstream appeal at this point, however.

    . All-day battery life: TBD.

    . Creator crafted: There are many workflows that relies on CUDA. Photoshop's Generative Fill, Premiere Pro's Generative Extend, ComfyUI with its many visual workflow tools, etc. Good for professionals & AI enthusiasts. Not so much for mainstream consumers.

    . Game ready: This is obviously a big one for consumers. Spark does have a leg up on Qualcomm's ill-fated Snapdragon X, in that N1X is a much more powerful chip. Spark will run the full RTX gaming stack, CUDA, DLSS 4.5, Reflex, G-Sync, etc. And Microsoft's Prism emulator will have had two full years to iron out compatibility issues. How well these points stack up will be borne out in upcoming reviews, which will no doubt be focused on gaming performance.

    My take from the mainstream Windows user's (consumer) perspective is that it's an iffy proposition for the here and now. Much of the appeal is geared toward creators, not consumers. Agentic AI is not yet a thing for consumers, and may well not ever be a thing.

    Snapdragon X flopped because it didn't really have a good selling point. The power-efficiency angle wasn't enough, and the AI angle did a face plant when Microsoft failed with its "40 TOPS AI PC" initiative. Microsoft may well fail again with agentic AI.

    In Spark's corner this time is it'll have both the professionals and AI enthusiasts onboard for its compelling AI uses. The only x86 equivalent is Strix Halo, a single product at $3K-4K. Spark is superior, if only for CUDA, and it'll be a range of devices, not just one. Then there's the AI wave that pushes every AI-related products in its wake. Lastly, it'll have the might of Nvidia behind it, the world's largest company.

    I think Spark's chances are better than even. If not this gen, then the next.
    thestryker said:
    They've also not promised anything with regards to Linux drivers which means Windows first.
    I don't know about laptops, but for desktops sold as workstations, I'm confident there'll be Linux support OOB. I can't imagine AI developers having to use Windows as a work platform.
    Reply
  • hotaru251

    new platform promises to turn Windows into an agentic AI OS

    and who asked for this?
    only MS & nvidia.

    Consumers don't really care about it & would rather MS fix the OS before anything else.
    Reply
  • thestryker
    dva852 said:
    I don't know about laptops, but for desktops sold as workstations, I'm confident there'll be Linux support OOB. I can't imagine AI developers having to use Windows as a work platform.
    If this was the case then one would think they'd have mentioned support. This is a Windows on Arm product so while I agree I can't imagine it being limited to Windows it is what it is.
    dva852 said:
    There isn't much detail, because OEMs will have their own announcements for their particular line-ups.
    The OEMs don't make the SKUs, nvidia does, and they said nothing about them.
    dva852 said:
    You can see the main selling points here, https://nvidia.com/en-us/products/rtx-spark
    A laptop with an RTX 5070 will be cheaper and faster (barring some heavy wattage limit) at every performance metric listed unless VRAM limited. As you go up the stack until you get to the RTX 5090 laptops this is likely to remain true with the performance gap widening along the way.
    dva852 said:
    The only x86 equivalent is Strix Halo, a single product at $3K-4K.
    There are a lot of Strix Halo based products, but not many in laptop form factors. Only the 128GB DRAM models are in the $3-4k range and RTX Spark is also pretty much guaranteed to be in the same place with that capacity.
    dva852 said:
    I think Spark's chances are better than even. If not this gen, then the next.
    I think the biggest thing going for it at this point is nvidia's promise of at least 3 generations of RTX Spark. We don't know what nvidia's expectations for sales are, but even if this one is low volume there's more on the horizon as long as they don't reverse course.
    Reply
  • usertests
    dva852 said:
    As reported by Videocardz, RTX Spark system won't have dGPU capability. This affirms that gaming, or at least high-end gaming, isn't a focus.
    You mean no proper I/O for an eGPU, or vendor system pairing it with a dGPU internally, or literally no capability to utilize any dGPU? Because I know ARM systems can use typical dGPUs, Jeff Geerling and others have shown that. (Probably even RISC-V at this point.)

    I wouldn't expect it to be a focus given the whole point of laying down $3k-5k for one of these is to utilize up to 128 GB shared memory for AI, and the gaming performance (on paper at least) shouldn't be horrible.
    Reply
  • Jabberwocky79
    A minor footnote that could have bigger implications for my workflow - I'm super not-thrilled with the idea of "partnering" with Adobe for a complete rebuild of Photoshop - just another way of locking Photoshop users into hardware they wouldn't otherwise want.
    Reply
  • usertests
    Jabberwocky79 said:
    A minor footnote that could have bigger implications for my workflow - I'm super not-thrilled with the idea of "partnering" with Adobe for a complete rebuild of Photoshop - just another way of locking Photoshop users into hardware they wouldn't otherwise want.
    Is Photoshop going to drop support for x86 or not use the GPU/NPU to support new features on x86?

    Worst case scenario, they are optimizing for Blackwell/Nvidia generally, and AMD will be left in the cold. And in the long run, Intel could replace higher-end graphics with Nvidia tiles, which could become a minimum requirement for anything meaningful and leave Intel Arc/Xe/UHD in the dust.
    Reply
  • ET3D
    I'm altenating between puking (about agentic OS and AI stuff) and excitement (about an efficient gaming and productivity platform). It would be interesting to see how this turns out and what price it's sold for.
    Reply