Pure muscle and localized agentic AI are about to ignite a PC revolution.
In a briefing ahead of Nvidia’s Computex keynote, we got our first look at RTX Spark, Nvidia’s ambitious new Arm-based laptop superchip. Nvidia is actively rethinking the PC paradigm, launching an all-in-one system-on-a-chip (SoC) that combines premium CPU processing power, a gaming-grade GPU, and the core AI architecture seen in Nvidia’s enterprise-grade DGX Spark. Designed from the ground up to support autonomous, on-device AI agents, it marks Nvidia’s biggest move into the consumer laptop space to date.
Whether your PC becomes a high-powered agent or not, these new chips will power some of the most capable laptops ever made. Delivering the petaflops of processing power required to run massive local AI models, the RTX Spark is built to crunch through heavy workloads that are simply too demanding for traditional PCs paired with standard discrete GPUs.
We’ll surely hear more about it during Computex, but here is everything we know so far about Nvidia’s disruptive new silicon.
In advanced developer circles, the rise of agentic AI has been revolutionary, leading to maxed-out workstation PCs running 24/7 to write code, hunt for system bugs, and coordinate cloud workflows.
This hasn’t translated into the same level of use in consumer systems, but it’s only a matter of time. Nvidia sees agentic AI as becoming the new primary interface for PC users, with conversational tasking replacing traditional app interaction via keyboards and screens. Touting massive benefits like absolute data privacy, tighter security, and eliminating the need to learn the unique interfaces of individual software tools, Nvidia is going all-in on localized AI.
Powering that shift? Nvidia’s RTX Spark.
If Spark sounds familiar, that’s because Nvidia launched the DGX Spark last year, a Linux machine aimed at developers, powered by Nvidia’s Grace Blackwell architecture. That same hardware is now coming to consumers in Windows with the RTX Spark, aimed at a more mainstream audience of consumers, gamers, and developers, but featuring a lot of the same hardware we saw in the DGX desktop.
Made with the latest 3nm process, the RTX Spark will combine a 20-core Grace CPU with Blackwell architecture featuring 6.144 CUDA cores. Though not strictly a GPU like dedicated GPUs in current gaming laptops, Nvidia notes that the CUDA core count is on par with the RTX 5070 laptop GPU and expects similar gaming performance. The chip fully supports the entire RTX technology stack, promising smooth 100-fps gaming at 1440p, 3D scene rendering, and up to 12K video editing.
In a massive departure from standard PC architecture, the RTX Spark features up to 128GB of LPDDR5X unified coherent memory (OEM configurations will scale from 16GB to 128GB). This massive pool of shared memory allows on-device AI models to run at scales that would completely choke a traditional PC.
Nvidia isn’t discussing detailed chip specs or direct performance numbers yet, but claims that the chip is engineered to meet and beat anything currently on the market.
Not only will these chips be powerful, but they’ll also be efficient, scaling dynamically from single-digit wattage for idle web browsing and productivity work, up to 80 watts under full gaming or local AI compilation workloads.
The other detail set to shake up the laptop industry is that Nvidia’s chips will run Windows on Arm rather than x86. Much like current Qualcomm chips that run Windows with a combination of native apps and emulated apps (via Prism emulation), the new RTX Spark systems will also run on Arm.
Nvidia says it has been working closely with Microsoft to optimize WoA for the new RTX Spark chips, but some of those benefits will carry over to Qualcomm systems as well. Nvidia has been further pushing software vendors to offer native Arm support, optimizing apps for the non-x86 side of things, but even pushing for better PRISM emulation with dedicated Advanced Vector Extensions 2 (AVX2) and anti-cheat support for gaming.
Though Nvidia didn’t share much about software, the briefing did include information about Adobe rearchitecting several key software tools for 100% GPU-accelerated processing on RTX Spark.
The bigger portion of the software conversation revolved around open source frameworks for agentic AI, with out-of-the-box support for tools like OpenClaw and Nous Research’s Hermes Agent. Windows is getting new kernel-level support for agentic tools, something we expect to hear a lot more about this week during Microsoft Build.
While on-device AI agents could be discussed at great length, Nvidia focused on three example use cases: creative work, developer tasks, and gaming and entertainment.
The demo included discussion of using AI for image generation and animation, and of employing Adobe and AI workflows to translate simple mood boards and text prompts into high-quality images and video. Complex animations will leverage tools like DLSS and OptiX.
Developers could set agents to the task of monitoring website bugs and proactively finding fixes, which the developer can then choose to implement. Monitoring can extend to code repositories or live web projects, with proposed fixes handled in a sandboxed QA environment to prevent accidents and agent overreach.
Finally, Nvidia discussed applying advanced tools like DLSS frame generation to video and 3D rendering. Gaming alongside in-game agents could help optimize hardware settings, manage livestream tasks, or even generate assets for game mods. New DLSS Ray Reconstruction is coming this August, and RTX Spark will support that too.
For all we don’t yet know about RTX Spark, we do know when it’s coming. Nearly every manufacturer in the PC space will be rolling out RTX Spark laptops and even a handful of RTX Spark mini PCs this fall. Asus, Dell, HP, Lenovo, Microsoft, and MSI were all name-checked. More than 30 premium laptops and 10 desktops will be announced with various configurations and features. We expect several to be unveiled at Computex this week.
While Nvidia didn’t specify the exact number of chip SKUs, there will be a tiered family of products ranging from mainstream to premium configurations, differentiated primarily by Grace core counts, the number of Blackwell CUDA cores, and memory capacity.
Pricing is not currently known. RTX Spark chips are for consumer models (no professional or business versions were announced), but the more premium models will likely align with top gaming and workstation hardware. Prices should be announced in the lead-up to the fall launch.