NVIDIA unveils RTX Spark AI‑focused PC SoC delivering 1 PFLOPS

NVIDIA announced its RTX Spark platform, a system‑on‑chip designed for the emerging AI Agent PC market. The chip combines a Grace‑based Arm CPU, a Blackwell GPU with 6,144 CUDA cores, NVLink‑C2C interconnects delivering up to 600 GB/s bandwidth, and 128 GB of unified memory, all within an 80 W TDP envelope. NVIDIA claims the solution provides up to 1 PFLOPS of local AI compute, targeting developers of large language models and content‑creation tools. The platform was showcased at COMPUTEX and introduced by CEO Jensen Huang at GTC 2026, positioning RTX Spark as the successor to previous AI‑enhanced PC concepts such as Copilot+ PCs.

Analysts compare RTX Spark with the existing RTX 50 series laptops, noting that while RTX Spark offers larger unified memory and stronger AI performance, its memory bandwidth (≈300 GB/s) trails the RTX 50’s 600‑800 GB/s. The launch is seen as NVIDIA’s attempt to move beyond the saturated gaming‑laptop market into high‑end, AI‑centric consumer devices, potentially opening a new market segment alongside traditional x86‑based machines.