Jensen Huang walked onto the Computex stage in Taipei on Monday and announced that Nvidia is, finally, building the laptop chip everyone has been waiting a decade for. The RTX Spark superchip, internally referred to as N1X, fuses a 20-core Nvidia Grace CPU with a Blackwell GPU carrying 6,144 CUDA cores and fifth-generation Tensor Cores, linked over NVLink-C2C to 128GB of LPDDR5X unified memory running at up to 300 GB/s. It’s a data-center architecture compressed into something Nvidia says will ship in laptops as thin as 14 millimeters.

The cast list tells the real story. Dell, HP, Lenovo, Asus, and MSI are all building Spark machines for fall. Microsoft co-developed the platform and is putting it inside a new Surface Ultra. That’s the entire Wintel OEM bench, plus Microsoft itself, lining up behind an Arm-based Nvidia design.

For Intel and AMD, this is the x86 monopoly on premium Windows laptops cracking open in public. Qualcomm spent years trying to make Windows on Arm credible and never quite found a flagship moment. Nvidia gets one on day one, because it arrived carrying the only silicon brand consumers actually ask for by name.

The pitch isn’t subtle. Nvidia is selling Spark as the substrate for what it calls an agentic Windows: 120-billion-parameter models running locally, context windows up to 1 million tokens, the RTX 5070 laptop GPU’s gaming credibility bundled in. Prism-optimised updates are meant to keep x86 software running while the ecosystem migrates. The competitive frame Nvidia keeps gesturing at is Apple’s M5 MacBooks, where unified memory and a tight CPU-GPU-NPU stack have quietly reset expectations for what a premium laptop feels like.

Arm, of course, is the licensor underneath all of this, finally collecting on a Windows beachhead it has chased since the Surface RT.

What’s striking is the institutional choreography. Microsoft didn’t pick a partner; it picked the partner with leverage in the AI capex cycle and bet its halo laptop on the relationship. The Wintel arrangement that defined personal computing for thirty years didn’t end with an announcement. It ended with a stage in Taipei and five OEMs nodding along.

Sources