Microsoft wants your next AI workstation to sit on your desk, not in a data center. On October 7, at an event in San Francisco during the city's Tech Week, the company opened preorders for the Surface Laptop Ultra, a 15-inch notebook built around Nvidia's RTX Spark chip, with availability starting October 16. Pricing starts at $2,599.
The headline is the silicon. RTX Spark pairs an Nvidia Blackwell RTX GPU with up to 6,144 cores and an Nvidia Grace CPU with up to 20 cores, backed by as much as 128GB of unified memory. Microsoft says the machine can run AI models exceeding 120 billion parameters entirely on-device, delivering up to 1 petaflop of AI performance. Nvidia CEO Jensen Huang joined Microsoft CEO Satya Nadella on stage to introduce it — a signal of how seriously both companies take the local-AI PC category.
Because RTX Spark is Arm-based, the Laptop Ultra runs Windows on Arm, with existing x86 applications handled through Microsoft's Prism emulation layer. Microsoft claims the laptop outperforms Apple's MacBook Pro with M5 Pro chip in several AI workloads, and positions it against the $2,999 16-inch MacBook Pro for creators, developers, and engineers. The device carries a mini-LED touchscreen display — which Microsoft calls its brightest and sharpest Surface display yet — along with three USB-C ports, USB-A, HDMI, an SD card reader, a 3.5mm headphone jack, and a large haptic trackpad. A new Magnetic Connect system adds magnetic USB-C charging, which Microsoft describes as the first built-in magnetic USB-C charging on a laptop. Notably, the display does not support pen input.
Alongside the laptop, Microsoft opened preorders for the $5,999 Surface RTX Spark Dev Box, a compact desktop aimed at developers and AI engineers who want to experiment with models on their own hardware. It ships in November and is available exclusively through Microsoft's online store in the United States, bundled with developer tooling including VS Code, GitHub Copilot CLI, WSL, and PowerShell 7.
The hardware lands with a software story. Microsoft made Execution Containers generally available on Windows 11 — a policy-driven sandbox that lets IT teams restrict which files and networks AI agents can access while they run. Nadella framed it as the "harnessed layer" Windows needs for the agent era: orchestrating models, memory, context, and actions safely on the PC rather than sending every workload to the cloud. Copilot's hybrid approach splits work between local hardware and cloud models, keeping sensitive or cost-sensitive tasks on-device.
For IT decision-makers, the pitch is economic as much as technical: local inference is effectively free once the hardware is purchased, and keeping data on the endpoint sidesteps both cloud bills and data-residency headaches. The open question is software maturity — Windows on Arm emulation has improved markedly, but buyers should verify their critical applications run well under Prism before committing a fleet.