Nvidia unveiled the RTX Spark chip at Computex 2026, pairing its Grace ARM CPU (up to 20 cores) with a Blackwell GPU housing 6,144 CUDA cores, mirroring the RTX 5070. Fabricated on TSMC’s 3 nm process, the die integrates up to 128 GB of LPDDR5X unified memory and stems from the GB10 silicon used in the DGX Spark mini‑PC line.

CEO Jensen Huang emphasized that the chip’s primary goal is to run AI agents locally, the next evolution beyond large language models. Those agents demand heavy compute and large memory, making the 128 GB pool essential. Nvidia positions RTX Spark for slim, high‑performance laptops slated for release this fall.

Nvidia has secured OEM partners including Acer, Asus, Dell, Gigabyte, HP, Lenovo, MSI and Microsoft, which will debut the Surface Laptop Ultra with the new SoC. While pricing remains undisclosed, the DGX Spark reference system sells for $3,500‑$4,700, hinting that premium laptops equipped with OLED displays could command a steep price tag.

By delivering a full‑stack, AI‑ready SoC in consumer notebooks, Nvidia aims to shift workloads from cloud servers to the edge, forcing rivals to reconsider laptop architectures. The RTX Spark thus represents a decisive step toward on‑device generative AI, reshaping performance expectations for future portable PCs.