Particle.news
Download on the App Store

Nvidia and Microsoft Unveil RTX Spark Superchip for Windows AI PCs

The move could shift data‑center AI workloads onto local Windows machines and reshape how users run large models.

Overview

  • The RTX Spark was announced this week at Computex in Taipei and Nvidia and partners say laptops and mini PCs with the chip will begin shipping in fall 2026.
  • Nvidia describes the chip as a unified Arm-based design pairing a 20-core Grace CPU with a Blackwell RTX GPU linked by NVLink and offering up to 128 GB of shared memory and as much as 1 petaflop of FP4 AI throughput.
  • Major PC makers including Microsoft, Dell, Lenovo, Asus, HP and MSI have committed to RTX Spark systems and Microsoft confirmed a Surface Laptop Ultra and a consumer-available Surface RTX Spark Dev Box preconfigured for developers.
  • Key questions remain unanswered publicly, including final pricing, independent performance benchmarks, real-world battery and thermal tradeoffs, broad Windows-on-Arm app compatibility and supply/manufacturing scale.
  • If the platform delivers on unified memory and local-model performance, it could cut cloud GPU bills, boost on-device privacy and force responses from Intel, AMD, Qualcomm and Apple while raising demand for app and developer tooling optimized for local AI.