Nvidia spent the last decade owning the data center. At Computex 2026 in Taipei on June 1, it made clear it now wants the laptop on your desk too. The company unveiled the RTX Spark Superchip, an Arm-based system-on-chip that fuses a 20-core CPU with a Blackwell RTX GPU and up to 128GB of unified memory, pushing as much as a petaflop of AI performance into a single PC. The reaction told the story: shares of AMD, Intel, and Qualcomm slid on the news. Nvidia is no longer content to sell the graphics card. It wants to be the whole computer.

What was actually announced

The RTX Spark pairs a high-performance Arm CPU with a Blackwell-generation GPU on one piece of silicon, linked by Nvidia's chip-to-chip interconnect so the two halves talk at very high bandwidth. The headline feature is the 128GB of unified memory, a single pool the CPU and GPU both address directly. Nvidia is positioning it as the engine for an "agentic" Windows, a PC where AI models run locally and continuously rather than living entirely in the cloud. Laptops and small desktops built on it are slated for the fall from ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI, with more vendors to follow.

Why unified memory is the real story

The spec that matters most is the least flashy one. In a conventional PC, the CPU and GPU each have their own memory, and running an AI workload means constantly copying data back and forth between them. That copying is slow, and it caps how large a model you can practically run on a personal machine. A single unified memory pool removes that wall. The processor and graphics engine reach into the same 128GB, so a much bigger model can sit in memory and run without the usual shuffling. This is the same architectural idea that makes Apple's chips punch above their weight, applied at a scale aimed squarely at AI. It is the difference between a PC that can dabble in local AI and one that can actually host it.

Why competitors flinched

The stock moves were not an overreaction. For years the PC silicon world had a comfortable division of labor: Intel and AMD supplied the CPU, Nvidia or AMD supplied the GPU, Qualcomm pushed Arm chips for thin-and-light machines. The RTX Spark collapses those boxes into one part that Nvidia controls end to end. If it succeeds, Nvidia is no longer a component vendor inside someone else's design; it is the platform, and everyone else is competing for whatever is left. The fact that Arm-chip specialist MediaTek collaborated on the CPU only underscores how serious the effort is. This is not a science project. It is a coordinated push to own a layer Nvidia historically did not.

Who this affects

For buyers, the promise is a laptop that can run capable AI models on-device, which means lower latency, more privacy, and features that keep working without a network. That is genuinely appealing. For the rest of the industry, it is a threat. AMD and Intel suddenly face a rival that controls the most desirable part of the AI stack and is now bundling a CPU with it. Qualcomm's pitch for Arm-on-Windows just got crowded by the most powerful name in AI hardware. And software makers will start optimizing for the platform that ships in volume, which tends to become self-reinforcing.

The catch worth naming

None of this is guaranteed. Nvidia has never had to win the messy consumer PC market, with its thin margins, picky OEMs, and demanding power and thermal limits. A petaflop on a slide is not the same as a quiet, cool, all-day laptop. Pricing is unknown, the software story for a new Arm Windows platform is never seamless, and the broader memory-pricing crunch squeezing the whole industry will not spare Nvidia. The data center taught Nvidia how to build dominance; the laptop will test whether it can do it in a market that does not already worship it.

There is also a longer-term question about lock-in. If the RTX Spark becomes the platform that developers optimize their local AI software for, that software starts to assume Nvidia's hardware, its memory architecture, and its tools, exactly the dynamic that made the company unassailable in the data center. Buyers may welcome a powerful AI laptop today without thinking about how hard it becomes to leave the ecosystem tomorrow. The history of computing is full of moments where a single vendor's integrated design became so convenient that the alternatives quietly withered. Whether that is good for the PC, which has long been an open, multi-vendor platform, is a different question from whether the chip is fast.

Our take

The RTX Spark is the clearest sign yet that the AI era is reorganizing the PC from the silicon up. The old boundaries between CPU, GPU, and memory were artifacts of an era when those parts came from different companies and rarely needed to cooperate. AI breaks that assumption, because the workload wants everything close together and addressing the same memory. Nvidia is simply the first to build the whole thing as one coherent product and aim it at consumers. Whether or not the first generation lands, the direction is set: the personal computer is being redesigned around running AI locally, and the company that already won the data center intends to win this layer too.

Trending on Tom's Hardware, analysis by GenZTech.