Jensen Huang put a different kind of Windows machine on stage at Computex 2026, and it does not run on x86. The NVIDIA RTX Spark platform places a Blackwell GPU alongside a 20-core Arm processor in a single integrated Superchip, then hands that to Microsoft to build an agentic, AI-first Windows experience around. With more than 30 laptops and roughly ten desktops promised for Fall 2026, this is the most serious push Windows-on-Arm has seen, and it lands squarely in the AI PC conversation South African buyers have been having all year.
Quick Answer
RTX Spark is a Windows-on-Arm consumer platform announced by NVIDIA and Microsoft on 31 May 2026 at Computex. The RTX Spark Superchip integrates up to 20 Arm CPU cores, a Blackwell GPU with 6,144 CUDA cores, and 128GB of unified memory in a single package, connected over NVLink-C2C. Devices from Dell, HP, Lenovo, Asus, MSI and Microsoft are slated for Fall 2026.
What the RTX Spark Superchip Actually Is
The headline is the unified design. Instead of a separate CPU, GPU and pools of RAM, RTX Spark fuses an Arm processor and a Blackwell graphics core onto one Superchip with a single 128GB memory pool both sides can address. NVIDIA's NVLink-C2C interconnect links the two so the GPU can reach memory at up to 300 GB/s without copying data back and forth across a slow bus.
That architecture is borrowed straight from NVIDIA's data-centre Grace-Blackwell thinking and shrunk into a consumer envelope. The practical upshot is a machine that treats large AI models as a normal workload rather than an exotic one, because the GPU is not boxed into a small dedicated VRAM budget.
The Numbers That Matter
NVIDIA rates the platform at peak FP4 AI throughput of one petaflop. On the unified 128GB pool, the company says a Spark machine can run 120-billion-parameter language models locally with up to a million tokens of context, render 90GB-plus 3D scenes, edit 12K 4:2:2 video, and generate 4K AI video. Those are workstation-class claims in a laptop chassis.
Gaming is part of the pitch too. NVIDIA quotes "100 FPS at 1440p," leaning on DLSS 4.5 and Multi Frame Generation to get there. Treat that as a best-case figure tied to upscaling rather than native rendering, but it signals that the Blackwell GPU here is not a token inclusion.
Why unified memory changes the AI story
On a conventional gaming laptop, your model has to fit in the GPU's dedicated VRAM, often 8GB or 12GB, or it spills to system RAM and crawls. With 128GB shared, that wall largely disappears for consumer-scale models. A developer can hold a very large model resident and still leave room for the operating system and apps, which is the single most interesting thing about the platform for anyone running local AI.
Windows on Arm: The Software Question
The hardware is only half the story. Windows-on-Arm has improved steadily, and Microsoft's Prism translation layer now runs most x86 apps acceptably, but native Arm builds of heavy creative and engineering software still lag. Anyone moving from an x86 machine should check that the specific tools they depend on, especially niche plugins, drivers and older games with kernel-level anti-cheat, run cleanly before committing.
Microsoft is positioning RTX Spark as the flagship for an "agentic" Windows where local AI handles tasks in the background. That is a forward-looking bet, and the first wave of devices will be where it gets tested in the real world.
The roadmap beyond this first chip
NVIDIA did not announce one generation and go quiet. The Computex 2026 presentation outlined a three-generation arc: the current RTX Spark chip, followed by a second generation moving to the Vera CPU and Rubin GPU on faster LPDDR6 memory, and a third generation named Rosa Feynman. That roadmap is deliberately long-horizon, signalling that Spark is a committed platform investment rather than a single product bet. For buyers considering early adoption, the multi-generation commitment reduces the risk that software ecosystems and partner support dry up after the launch wave.
Who Should Care, and When
For now this is a platform to track rather than rush. The launch lineup spans Dell, HP, Lenovo, Asus, MSI and a Microsoft Surface Ultra, with Acer and Gigabyte to follow, so choice will be wide by the time stock settles. If your work is local AI inference, large-context coding agents or heavy video, the unified-memory design is genuinely compelling. If you mainly game or run mainstream x86 software, a current Blackwell-based desktop or laptop remains the safer pick today, and you can compare what is shipping now in the AI PC range.
Buyers weighing a graphics upgrade in the meantime can see which Blackwell cards are moving locally on the GPU best sellers list.
Frequently Asked Questions
When will RTX Spark devices actually go on sale?
NVIDIA and Microsoft have targeted Fall 2026 for the first laptops and compact desktops from Dell, HP, Lenovo, Asus, MSI and Microsoft, with Acer and Gigabyte models following after. SA availability and Rand pricing will trail the global launch, as is usual.
Is RTX Spark the same as DGX Spark?
No. DGX Spark is NVIDIA's small AI developer box. RTX Spark is the consumer Windows-on-Arm platform built on the same Grace-Blackwell philosophy but aimed at mainstream laptops and desktops running everyday Windows.
Will my current Windows software run on it?
Native Arm apps run best, and Microsoft's translation layer handles most x86 programs, but performance and compatibility vary. Check your essential applications, drivers and any anti-cheat-protected games for Arm support before buying.
How much VRAM does it have for AI work?
There is no separate VRAM figure in the traditional sense. The GPU shares the full 128GB unified memory pool, which is why it can hold very large local models that would not fit on a standard discrete GPU.
Is it good for gaming?
It can game. NVIDIA cites around 100 FPS at 1440p using DLSS 4.5 and frame generation. That is promising, but Windows-on-Arm game compatibility is still maturing, so gaming is a secondary strength rather than the main reason to buy one at launch.
RTX Spark is the platform to watch, but you can build a capable AI machine today. Explore current AI-ready systems in the Evetech AI PC range and get ahead of the Arm wave.