Nvidia and Microsoft say RTX Spark laptops will be available October 16, combining up to 128GB of unified memory with as much as one petaflop of AI compute. Nvidia is positioning the machines for local agents, creative software and development workloads that would normally exceed a conventional laptop’s memory.
Agent applications including GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi and WorkBuddy are receiving setup and security work for the platform. Nvidia says OpenShell and Microsoft Execution Containers can isolate agent activity and protect credentials. Hermes plans to add granular controls and container support by the end of October.
A new Nemotron 3.5 Bolt 74B model is scheduled for October 15. It uses a mixture-of-experts design with 9 billion active parameters and fits into 24GB through 2-bit quantization. Nvidia says it will run through llama.cpp, vLLM, Ollama, Unsloth and LM Studio, with weights, training data and recipes released openly.
The launch also includes hybrid routing between local and cloud models and broader support from CUDA and creative applications. Actual speed, battery use and model quality will depend on each laptop and workload, but the memory capacity makes substantially larger local models practical on a portable PC.