NVIDIA's RTX Spark superchip merges CPU, GPU, and 128GB memory for AI PCs

At Computex 2026, NVIDIA CEO Jensen Huang introduced the RTX Spark superchip, describing it as a pivotal moment for personal computing. The chip integrates NVIDIA's Grace CPU and Blackwell GPU into a single package, paired with up to 128GB of unified memory. This architecture allows seamless sharing of data between processing units, eliminating bottlenecks common in traditional PC designs. Huang emphasized that the RTX Spark is not just an incremental upgrade but a fundamental reinvention of the PC for the age of artificial intelligence. The chip is optimized for both on-device AI inference and creative workflows, enabling tasks like real-time language model execution, 3D rendering, and high-fidelity gaming without relying on cloud servers. Targeting thin-and-light laptops, it aims to bring workstation-class performance to mobile users.

Beyond hardware specs, Huang highlighted the RTX Spark's ecosystem compatibility. It supports NVIDIA's full software stack, including CUDA, TensorRT, and the Omniverse platform. Developers can run large language models locally, while gamers benefit from ray tracing and DLSS optimizations tailored for the Blackwell GPU. The unified memory pool simplifies programming, as developers no longer need to manually manage separate CPU and GPU memory. Huang positioned the chip as a direct response to rising demand for local AI computing, citing privacy and latency benefits over cloud-based alternatives. With availability slated for late 2026, the RTX Spark could redefine expectations for what portable PCs can accomplish, blending enterprise-grade capabilities into consumer-grade devices.