Windows laptops start running large models locally: RTX Spark packs 128GB unified memory and a petaflop into the chassis

The Windows PC is being repositioned as a persistent platform for agents, with new execution containers at the operating-system level and new local compute capable of carrying large models. Enterprise developers previously either moved to Linux to reach AI compute or stayed on Windows with limited hardware options for heavy model development and multi-agent workloads. Microsoft announced general availability of Microsoft Execution Containers (MXC), OS-level infrastructure that lets agents run safely and persistently in the background under operating system control.

RTX Spark brings the full NVIDIA AI stack into laptops and compact desktops, so models, tools and workflows move between RTX Spark and DGX Station without rewriting. That class of local model capability previously depended on the cloud or rented clusters, with data sent off the machine. The part pairs a Blackwell RTX GPU with up to 6,144 cores and an up to 20-core Grace CPU connected at 600 GB/s, delivering one petaflop of FP4 AI performance and up to 128GB unified memory, enough to run a 125B model such as Qwen 3.8 Flash Next.

DGX Station enters the Windows ecosystem for the first time, putting GB300 Grace Blackwell-class infrastructure on a desk. DGX Station previously ran on Linux, which meant enterprise developers maintained two separate environments. The machine runs on the GB300 Grace Blackwell Ultra Desktop Superchip with 748GB of coherent memory and up to 20 petaFLOPS of FP4 AI compute, enough for trillion-parameter-scale models locally, with Linux AI toolchains still reachable through WSL.

The report covers hardware and system updates Microsoft announced at its Windows AI and Surface event in San Francisco: RTX Spark laptops on preorder now and available October 16, compact desktops on sale in November, and DGX Station for Windows as a preview. Systems will come from Acer, ASUS, Dell, HP, Lenovo, Microsoft, MSI and Gigabyte, ranging from slim laptops to compact desktops built for always-on agents. The intended readers are developers, researchers and enterprise teams who need to run large models and multi-agent workloads locally.

All compute, memory and model-running capabilities come from vendor statements at a launch event, and the material gives no third-party benchmark or independent review. MXC is described as letting agents run safely and persistently under operating system control, but the material does not detail its isolation mechanism or governance. DGX Station for Windows is only a preview, and the material gives no availability date.