Jensen Huang took the stage at Computex 2026 in Taipei today and introduced what might be Nvidia’s boldest product move in years: the RTX Spark Superchip.
What’s inside
RTX Spark pairs a Blackwell GPU with a brand-new ARM-based processor called N1X, co-developed with MediaTek and Microsoft. The specs read more like a workstation than a laptop: 128 GB unified memory, one petaflop of AI performance, and a 20-core CPU. All of this is designed to run in laptops and desktops — not data centers.
More than 30 laptops and 10 desktops from Dell, HP, and other partners are expected to ship this fall. No pricing yet.
Why this matters
Nvidia is doing two things at once here. First, it’s entering the PC processor market — a direct challenge to Apple Silicon, Qualcomm’s Snapdragon X, and Intel. Second, Huang is positioning the PC itself as an AI platform. His quote: ‘This reinvention of the computer is as big of a deal as the reinvention of the phone into what we now know as the smartphone.’
Sounds like marketing, but the numbers behind it are real. 128 GB of unified memory means you can run large language models locally. One petaflop of AI performance on a laptop was science fiction two years ago.
The roadmap
Nvidia also outlined what comes next: after the current Blackwell-based Spark, there’s a Rubin generation with LPDDR6 memory, followed by something called ‘Rosa Feynman.’ Whoever names things at Nvidia is clearly having fun.
What this means for local AI
If you’re working with Claude Code, local LLMs, or AI-powered workflows, RTX Spark could be the inflection point where local AI processing becomes genuinely practical. 128 GB unified memory is enough for models that currently live only in the cloud. Whether Nvidia can deliver on the price-performance promise is the open question — we’ll find out this fall.
Huang emphasized again at Computex that Nvidia sees itself as an infrastructure company — not just a GPU maker. RTX Spark is the clearest evidence of that shift yet.
Sources: