Microsoft on October 7 began taking pre-orders for its first Nvidia-chip AI PCs, the Surface Laptop Ultra and the Surface RTX Spark Dev Box, alongside a Windows 11 update designed to run AI agents locally.
The launch is Microsoft’s most explicit bet yet that on-device AI processing can reshape how work gets done, promising to cut cloud-compute costs and latency by letting users run large models and autonomous agents directly on a laptop or desktop, not just in data centers.
The new machines are built around Nvidia’s RTX Spark superchip, a system-on-a-chip that combines a Blackwell RTX GPU with up to 6,144 CUDA cores and a 20-core Grace CPU on a single package. Microsoft’s Devices Blog announced that the Surface Laptop Ultra starts at $2,599 and will ship on October 16, while the desk-bound Surface RTX Spark Dev Box is priced at $5,999 and will begin shipping in November. The Dev Box is pre-order only in the U.S., while the laptop is available worldwide.
The hardware arrives as Microsoft refreshes Windows 11 with a “hybrid intelligence” architecture that mixes local and cloud AI. The Windows Experience Blog says Microsoft Execution Containers (MXC), a sandboxing technology for agents, is now generally available on Windows 11, allowing IT administrators to define which files and networks an agent can access. Nvidia contributes its own OpenShell runtime, which can disguise personal information in queries sent to cloud models and enforce user-defined privacy policies, according to Nvidia’s newsroom.
Satya Nadella, Microsoft’s chairman and CEO, posted on X that it was “great to partner with NVIDIA as we kick off this next chapter for Windows,” while Nvidia CEO Jensen Huang called the RTX Spark “the new PC—the personal AI computer.”
The Surface Laptop Ultra is a 15-inch device under 18 mm thick and weighing less than 4.5 lb, with a PixelSense Ultra touchscreen that Microsoft says delivers peak HDR of 2,000 nits—25% brighter than a MacBook Pro M5, based on published specs. The chassis houses up to 128 GB of unified memory shared between CPU and GPU, enough to run language models exceeding 120 billion parameters locally with up to 1 million tokens of context, according to Nvidia.
The RTX Spark platform delivers up to 1 petaflop of AI performance in FP4 precision when using sparsity, the two companies claim. The blog says Microsoft has also shrunk its own MAI Code 1.1 Flash model—a 137-billion-parameter coding model with 6.8 billion active parameters—to a 3-bit local version that preserves coding quality while reducing memory use by nearly 80%.
On the software side, the Windows Experience Blog lists agents that already support MXC, including OpenAI Codex, GitHub Copilot, OpenClaw, and Replit, with Anthropic Claude Code, Box, Perplexity, and others said to be adding support. Meta’s Muse agent is promised as a native Windows app with MXC integration “coming soon.”
Microsoft is offering up to $1,000 cash back for customers who trade in an eligible MacBook Pro during the pre-order period, which runs from October 7 through November 23. The trade-in is administered by Teladvance and requires the old device to power on and be free of liquid damage, with the final value determined after inspection.
No independent performance benchmarks or battery-life figures were included in Microsoft’s or Nvidia’s announcements. The two companies said RTX Spark-powered laptops from ASUS, Dell, HP, Lenovo, and MSI are also coming this fall, with models from Acer and Gigabyte to follow.