NVIDIA RTX Spark Lands - 128GB Unified Memory and Local 120B LLMs Come to Windows Laptops

NVIDIA has announced RTX Spark at Computex 2026, a new superchip architecture that promises to bring high-end AI capabilities to consumer laptops and desktops. The chip combines a 20-core Arm CPU with a Blackwell GPU featuring 6,144 CUDA cores, but the standout specification is 128 gigabytes of unified memory with 300 gigabytes per second of bandwidth. This memory configuration enables the chip to run 120 billion parameter language models with million-token context windows locally, whilst also handling demanding creative workloads like rendering 90-gigabyte 3D scenes and editing 12K video. NVIDIA claims these capabilities come without sacrificing traditional gaming performance, with systems still achieving triple-digit frame rates at 1440p resolution in slim laptop form factors with all-day battery life. The company has secured partnerships with major OEMs including Asus, Dell, HP, Lenovo, Microsoft, and MSI to ship RTX Spark systems in autumn 2026. Microsoft is contributing a new Windows security layer called OpenShell, designed to sandbox local AI agents safely. NVIDIA positions RTX Spark as "DGX Spark for the rest of us," referencing their enterprise AI platform whilst targeting consumer markets. The announcement signals a significant shift in the local AI landscape. If 128GB of unified memory becomes available at consumer price points, it could democratise access to large language models that currently require expensive workstations or cloud computing. This development blurs the traditional distinction between AI workstations and consumer laptops, potentially enabling widespread adoption of agentic AI systems that can operate independently on personal devices.