Microsoft's Surface Laptop Ultra with Nvidia RTX Spark goes on sale, top model already sold out

Microsoft put its Nvidia-powered Surface lineup on sale on October 7, with CEO Satya Nadella appearing alongside Nvidia CEO Jensen Huang to mark the move from announcement to retail availability. The Surface Laptop Ultra starts at $2,600, climbs to $3,700 for the mid-tier configuration, and tops out at $5,900 for the maximum build — which was already out of stock within hours of going live.

Both the laptop and a new developer-focused desktop, the Surface RTX Spark Dev Box, are built around Nvidia's RTX Spark chip (known internally as the GB10), first shown off in June. The chip pairs a 20-core Arm-based Nvidia Grace CPU with a Blackwell GPU carrying 6,144 CUDA cores and fifth-generation Tensor Cores, connected over NVLink C2C. It's a different architecture than the x86 chips that have powered most Windows PCs for decades, aimed squarely at running AI models locally rather than in the cloud.

Why local compute, not cloud

The pitch for both devices is the same: enough unified memory and AI throughput to run large models directly on the machine, without a round trip to a datacenter. The Dev Box packs 128GB of LPDDR5X memory, with up to 112GB assignable to the GPU, and Microsoft says it can run models above 120 billion parameters with a context window up to one million tokens — workloads that would otherwise require cloud GPU rental. At a 100-watt thermal envelope, it's positioned as a direct answer to Apple's Mac Studio for AI developers who want that scale of local inference without a server room.

Nadella framed the hardware push around a broader argument about where AI agents need to live. "You really do need to orchestrate, and you need to have memory outside of the model," he said, pointing to a Windows-level role in managing AI agents rather than leaving that entirely to the model itself.

Developer tooling bundled in

The Dev Box, starting at $6,000, ships configured for AI development out of the box, with VS Code, GitHub Copilot CLI, WSL, and PowerShell 7 preinstalled. That positions it less as a general-purpose desktop and more as a dedicated local-inference workstation — a category Nvidia and Microsoft are betting will grow as more development teams want to test and run large models without per-hour cloud GPU bills.

Competition and context

Dell is shipping a competing machine, the XPS 16 Creator Edition, also built on RTX Spark and available for preorder at Best Buy for $3,800 with October delivery, suggesting Nvidia is positioning the chip as a platform for multiple PC makers rather than a Microsoft exclusive. Microsoft is also sweetening the laptop's pitch with a trade-in offer of up to $1,000 off for customers switching from a MacBook Pro, a direct shot at the machine the Surface Laptop Ultra is most often compared to on price and build quality.

The sold-out top configuration within hours suggests real demand for local AI hardware at the high end, even at a price point that puts the top Surface Laptop Ultra above most previous-generation gaming laptops. Whether that demand holds at scale, once the novelty of local large-model inference wears off, is the question Microsoft and Nvidia are now racing to answer with actual sales data rather than announcement-day hype.