Nvidia has announced RTX Spark – a specialized chip for Windows-based PCs designed to enable on-device execution of AI agents and large language models. The project, developed in collaboration with Microsoft over approximately three years, is positioned as the foundation for personal AI systems that operate directly on the user's hardware without mandatory reliance on cloud infrastructure. Specified performance metrics include up to 1 petaflop of compute power and up to 128GB of unified memory, which Nvidia states will allow users to run agent-based and generative models locally on their PCs.

The hardware platform integrates a Blackwell RTX GPU with 6,144 CUDA cores and a 20-core Grace CPU developed with participation from MediaTek. The solution is optimized for Windows 11 and use cases involving AI agents, content creation, and gaming. Concurrently, Nvidia and Microsoft revealed joint efforts on infrastructure for agent systems within the OS, citing new security mechanisms and the NVIDIA OpenShell platform for deploying agents on end-user devices. Nvidia CEO Jensen Huang framed this development as a shift from the paradigm of "PC as a tool" to "PC as an assistant."

Adobe is optimizing Photoshop and Premiere for the RTX Spark architecture: Nvidia claims a twofold improvement in AI and graphics performance on the new platform, though independent verification of these figures is pending. In May, Photoshop 27.7 introduced an on-device model for the Remove tool, enabling object deletion without cloud data transmission, while Premiere received AI-powered features including Object Mask. The first RTX Spark-based PCs are expected in fall 2026, with launch partners including ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI; Acer and GIGABYTE are slated to join later. Notably, in March Nvidia announced infrastructure plans for space-based data centers.