Nvidia RTX Spark brings local AI workloads to laptops

Nvidia’s RTX Spark laptops go on sale Oct. 16, targeting developers, creators and power users who want to run AI models and autonomous workflows without relying on cloud services. The platform combines an Arm-based Grace CPU, Blackwell RTX graphics and up to 128GB of unified memory through Nvidia’s N1X chip.

Shared memory gives the CPU and GPU access to the same pool, reducing data transfers and allowing larger models to run locally than conventional laptop graphics memory typically permits. On the 128GB model, up to 96GB can be allocated to GPU access. Configurations with the largest memory capacity are projected to cost about $6,000. Nvidia promises 99% of plugged-in performance when running on battery, though performance claims still await independent testing.

Microsoft is adapting Windows scheduling and memory management to support background agents alongside everyday applications. Microsoft Execution Containers restrict agents’ access to files and network resources, while separate agent identities limit permissions. Nvidia’s OpenShell adds controls intended to block unauthorized access and keep login credentials hidden from agents.

The platform also supports local creative tools, RTX gaming features and distributed inference across networked PCs through Nvidia’s Personal AI Router. Native Arm game support and anti-cheat compatibility remain developing areas. Microsoft confirms RTX Spark systems qualify as Copilot+ PCs, although some Windows AI features will arrive later.