NVIDIA RTX Spark laptop preorders opened today, giving buyers an October 16 availability date for PCs designed to run AI models locally. RTX Spark supports a maximum of 128GB in unified memory, and NVIDIA and Microsoft are jointly developing the Windows hardware and software needed for AI agents. NVIDIA's announcement did not include pricing for RTX Spark or Surface Laptop Ultra.
Microsoft's Surface Laptop Ultra uses RTX Spark as its computing platform. NVIDIA lists seven other system makers: Acer, ASUS, Dell, Gigabyte, HP, Lenovo and MSI. Compact desktop versions are scheduled to go on sale in November and use the same RTX Spark superchip, which NVIDIA intends for agents that operate 24/7.
Read more: Nvidia Announces RTX Spark Processor Family for Laptops and Desktops at Computex 2026
That desktop schedule follows an earlier October availability target for compact RTX Spark systems from Lenovo and Acer. Microsoft introduced Surface Laptop Ultra at Computex 2026 with a shipment target of later in 2026, describing a Windows on Arm machine whose software compatibility depends on native Arm applications, Prism emulation and anti-cheat support.
RTX Spark delivers one petaflop of FP4 AI performance. Its NVIDIA Grace CPU can have as many as 20 cores. The platform runs NVIDIA CUDA and supports local execution of the 125B Qwen 3.8 Flash Next model, which NVIDIA says avoids usage metering and sending data to the cloud.
Gaming support includes Reflex and DLSS 5, while NVIDIA claims RTX Spark can handle AAA games at 1440p. For creators, the platform offers AV1 acceleration in hardware and RTX ray tracing. Developers can also move their models, tools and workflows from RTX Spark to DGX Station without rewriting them.
Microsoft Execution Containers (MXC) also became generally available today, providing infrastructure controlled by the operating system for agents that persist in the background. Microsoft presents Agent 365 and Microsoft Security, together with MXC, as safeguards for those agents.
NVIDIA also previewed NVIDIA DGX Station for Windows today, bringing a system that previously required Linux to Windows. Its coherent memory capacity is 748GB, and NVIDIA targets it at developers, researchers and enterprises running model tuning, inference or continuously operating agents locally. Linux AI tools remain accessible through WSL.
