NVIDIA unveiled RTX Spark, a consumer‑grade processor built on the GB10 Superchip and marketed as the first Windows 11 AI PC capable of running large language models locally. Co‑developed with Microsoft, the chip merges a Grace‑based 20‑core CPU with a Blackwell iGPU, targeting notebooks and compact desktops from major OEMs.

The silicon is fabricated on TSMC’s 3 nm EUV node and supports up to 128 GB of unified memory, enough for 200‑billion‑parameter models. Its iGPU houses 48 streaming multiprocessors—6,144 CUDA cores—delivering 1 petaFLOP of FP4 AI throughput and full DirectX 12 Ultimate features, including ray tracing and DLSS 4.5.

Microsoft’s Windows 11 AI PC layer provides an OS‑level translation that runs existing x86‑64 apps on Arm, while NVIDIA works with ISVs to ship Arm‑native versions of Photoshop, Premiere and other creator tools. Game studios receive DirectX 12 Ultimate support and the full RTX/DLSS stack, allowing 1440p AAA titles to run with ray tracing on thin‑and‑light laptops.

OEMs such as ASUS, Dell, HP, Lenovo, MSI and Microsoft Surface will ship RTX Spark laptops with 14‑ or 16‑inch OLED panels, 16:10 aspect ratios, Wi‑Fi 7 and up to 10 Gbps Ethernet, slated for fall 2026. By merging AI acceleration with high‑end graphics, NVIDIA aims to challenge Apple’s MacBook line and AMD’s Ryzen AI Max offerings in the premium notebook segment.