Overview
- Nvidia unveiled the RTX Spark superchip at Computex this week, describing an Arm 20‑core CPU paired with a Blackwell RTX GPU, a company‑claimed ~1 petaflop of AI performance, and up to 128GB of unified memory for loading large models locally.
- Microsoft showed a Surface Laptop Ultra and a Surface RTX Spark Dev Box configured for developers and said major PC makers are building dozens of RTX Spark systems with shipments expected later in 2026.
- On‑site hands‑on reviews at Computex reported strong performance for video editing, gaming, and local model inference, but independent benchmarks for sustained workloads are not yet available.
- Analysts and reporters flagged unanswered execution risks including battery life and power efficiency under steady agent workloads, Windows-on‑Arm app compatibility and emulation performance, final pricing, and supply constraints for high‑capacity memory.
- The move reframes the PC as a runtime for local AI and challenges Intel, AMD, Qualcomm, and Apple’s Mac ecosystem that already uses unified‑memory M‑series chips, leaving initial adoption likely concentrated among creators, developers, and AI enthusiasts willing to pay premium prices.