Nvidia just dropped what might be the most consequential personal computing announcement in years. At GTC Taipei 2026, running June 1 through June 4 alongside the annual Computex tech expo, CEO Jensen Huang unveiled the RTX Spark superchip, a platform designed to turn everyday Windows PCs into personal AI agent machines.
What RTX Spark actually is
The superchip pairs a Blackwell RTX GPU packing up to 6,144 CUDA cores with a 20-core Grace CPU. These two processors are connected via NVLink-C2C, the same high-bandwidth interconnect Nvidia uses in its server hardware.
The result: up to 1 petaflop of AI processing power in a consumer device. In English, that’s a quadrillion floating-point operations per second.
The platform supports up to 128 GB of unified memory, meaning the CPU and GPU share the same memory pool. This matters enormously for running large AI models locally because it eliminates the bottleneck of shuttling data between separate memory banks.
