RTX Spark launches to run local AI agents

What happened: NVIDIA said its RTX Spark laptops go on sale Friday, Oct 16, built around a superchip with up to 128GB of unified memory and 1 petaflop of FP4 AI compute, pitched as a gaming PC, creative studio and personal AI computer in one. Ahead of launch, its partners rolled out agent support, including a “Getting Started” screen that installs an agent app such as GitHub Copilot, Hermes Agent, OpenClaw, Perplexity, Pi or WorkBuddy. NVIDIA’s Nemotron 3.5 Bolt 74B-A9B, a 74B-parameter mixture-of-experts model built for local agents, releases Oct 15.

Context: NVIDIA PAIR, its personal AI router, left beta and adds llama.cpp support so networked PCs can pool their compute. Agent sandboxing uses NVIDIA OpenShell and Microsoft’s Execution Containers, while an open-source router called Switchyard shifts work between local and cloud models. On the gaming side, hundreds of titles ship at launch with DLSS 5, a 2x faster DLSS 5 model debuts in NBA 2K27, and the first native PyTorch-CUDA and Python builds for Windows on Arm arrive alongside.

Why it matters: It is a large hardware bet that consumer machines can run the agents that drive them, rather than leaning on the cloud. If local models keep closing the gap with frontier cloud models, the economics and privacy of everyday AI work shift toward the device.

(Original source: NVIDIA)