Surface Laptop Ultra and Surface RTX Spark Dev Box (2026): Microsoft's Most Powerful Surface Devices, Read From the Announcement

---

Bottom line: on October 7, 2026 Microsoft opened pre-orders for the Surface Laptop Ultra (from $2,599, shipping October 16) and the desk-bound Surface RTX Spark Dev Box ($5,999, shipping in November). Both use NVIDIA's RTX Spark Superchip with up to 128 GB of unified memory and are stated to run AI models beyond 120B parameters locally. The laptop has a 15-inch display, a chassis under 18 mm thick, and what Microsoft calls the world's first built-in magnetic USB-C charging on a laptop.

Surface devices have usually been framed as thin, light Windows machines. This announcement is aimed elsewhere: at people who want creative work, development and local AI inference in one box.

This article uses only what Microsoft's blog post and its footnotes state, covering what shipped, what it costs, and where the caveats sit.

What you will find here

• What the two devices contain: chip, memory, display, ports
• Prices, availability and shipping dates
• The fine print behind "1 petaflop" and "120B parameters"
• Who should consider them, and who should not

The two devices

Microsoft published the same day a broader Windows vision for hybrid intelligence, combining PC and cloud. These two machines are the concrete expression of it.

Surface Laptop Ultra: what is new

Chip and memory

The machine is designed around NVIDIA's RTX Spark Superchip.

• An NVIDIA Blackwell-generation RTX GPU with up to 6,144 cores
• An NVIDIA Grace CPU with up to 20 cores
• Up to 128 GB of unified memory

The memory is shared dynamically between CPU and GPU. Microsoft's footnote states that the maximum the GPU can address depends on configuration and workload and is less than the total.

With that, Microsoft says the machine runs AI models beyond 120B parameters locally, with up to 1 petaflop of AI performance. The footnote qualifies that figure as theoretical FP4 performance using the sparsity feature, so it is a specification rather than a measurement. Treat it as a catalog number and wait for independent testing before assuming real throughput.

For developers, the practical part is that the CUDA ecosystem and the full Windows AI stack are available, so local models can be paired with GitHub Copilot for agentic coding on the device itself.

Thermals and chassis

Fast chips in thin bodies usually throttle. Microsoft put a new thermal architecture in place, claiming up to 2.5x more thermal power capacity than its current Surface Laptops.

The chassis measures under 18 mm thick and under 4.5 lb, which is thin and light for this performance class.

Display

The 15-inch PixelSense Ultra touchscreen is described as the brightest, sharpest and most vivid display Microsoft has put on a Surface Laptop.

• Peak HDR brightness up to 2,000 nits, measured over a 10 percent window with HDR content
• Up to 25% brighter peak HDR than MacBook Pro M5
• Comparisons are against laptops commercially available as of October 2026

Two caveats matter. First, 2,000 nits is a peak figure, and the footnote states sustained and SDR brightness are lower. Second, the touchscreen is optimised for touch only and does not support pen input such as Surface Pen, which rules the machine out for anyone who draws on screen.

Ports and charging

The headline feature is Magnetic Connect, described as the world's first built-in magnetic USB-C charging on a laptop. The included cable attaches magnetically to the right-side USB-C port and releases easily, and the port still works for video, data and other USB-C devices.

The full port set is:

• Three USB-C ports, one of which is the Magnetic Connect port
• HDMI
• USB-A
• An SD card reader
• A 3.5 mm headphone jack

It drives up to three 4K displays, so footage can be imported without adapters. Storage is user-removable, and the machine comes in Platinum and Nightfall.

Surface RTX Spark Dev Box: AI compute on the desk

The Dev Box puts the same RTX Spark Superchip into a compact desktop for developers and AI engineers.

• Up to 1 petaflop of AI performance, again theoretical FP4 with sparsity
• 128 GB of unified memory
• Support for models beyond 120B parameters with large context windows
• Local inference, evaluation and experimentation at interactive speeds

Its selling point is being ready out of the box. As part of the Project Zenith family it ships with Visual Studio Code, Git, the GitHub CLI, GitHub Copilot, WSL, Python, Node and more already installed, with the explicit goal of spending less time configuring and more time building.

At $5,999, it is sold exclusively on Microsoft.com in the U.S. and begins shipping in November, which means buyers elsewhere should check purchasing and warranty routes first.

How agents are governed

This part is easy to skip and matters to developers.

• Microsoft Execution Containers provide containment for agents
• Upcoming Windows capabilities use Microsoft Entra to distinguish agent activity from user activity
• Microsoft Agent 365 controls extend to local agents on the device
• Microsoft Dev Box adds cloud capacity when local compute is not enough

Microsoft frames Windows as the OS for agents and Surface as the hardware for it. The aim is governed AI development without spending cloud tokens on every step, with permission boundaries for agents moving into the operating system itself.

Prices and availability

• Surface Laptop Ultra: from $2,599 MSRP, pre-order on Microsoft.com or through select retailers worldwide, shipping from October 16
• Commercial configurations are available for pre-order as well
• Trade in an eligible MacBook Pro for up to $1,000 cash back
• Surface RTX Spark Dev Box: $5,999, exclusively on Microsoft.com in the U.S., shipping from November

Japanese pricing and release dates were not part of this announcement.

Who it fits, and who it does not

It fits people who want to:

• Run 120B-class models locally instead of sending data to a cloud API
• Keep creative work and development in one machine
• Use CUDA-based tools and libraries unchanged
• Have a pre-configured AI workstation on a desk, which is the Dev Box case

It fits poorly if you:

• Draw with a pen, since the touchscreen does not support pen input
• Want the lightest possible laptop, despite the sub-4.5 lb figure
• Are optimising for budget, given the $2,599 starting point
• Mainly want Windows on Arm gaming, where support is growing but not universal

Questions readers ask

Q1. Can I buy it in Japan?

A. The Laptop Ultra is described as available through select retailers worldwide, so Japanese availability is possible. The Dev Box announcement is U.S. only. Japanese pricing and dates are not in this announcement.

Q2. What is the 128 GB of memory for?

A. It is unified memory shared between CPU and GPU, and Microsoft notes the GPU-addressable maximum is less than the total and depends on workload. It exists to give headroom to large local models.

Q3. What does "runs models beyond 120B parameters" mean in practice?

A. It means a model with more than 120 billion parameters can run locally. Actual speed depends on the configuration and on how the model is quantised. The 1 petaflop figure is theoretical FP4 with sparsity.

Q4. How does it compare with a MacBook Pro?

A. Microsoft claims up to 25% brighter peak HDR than MacBook Pro M5, comparing against laptops available as of October 2026. Speed and thermal comparisons need hands-on reviews.

Q5. Does it support a pen?

A. No. Microsoft states the touchscreen is optimised for touch only and does not support pen input such as Surface Pen.

Q6. Can it play games?

A. Titles supporting ray tracing, DLSS and NVIDIA Reflex work, alongside the growing Windows on Arm catalog. Not every x86 game runs, so check the ones you care about.

Q7. What ships on the Dev Box?

A. Visual Studio Code, Git, the GitHub CLI, GitHub Copilot, WSL, Python, Node and more, pre-installed as part of the Project Zenith family.

Q8. Can agent execution be restricted?

A. Microsoft has announced containment through Microsoft Execution Containers, Entra-based separation of agent and user activity, and extension of Microsoft Agent 365 controls to local agents.

Summary

The announcement moves Surface from thin-and-light Windows machines toward high-performance hardware for local AI. The Laptop Ultra starts at $2,599 and ships from October 16; the Dev Box is $5,999 and ships in November. Both centre on the RTX Spark Superchip with up to 128 GB of unified memory and support for models beyond 120B parameters run locally.

Read the numbers carefully: 1 petaflop is theoretical FP4 with sparsity, 2,000 nits is peak HDR over a small window, and pen input is not supported. With those understood, the machines make sense for anyone who wants local AI, creative work and development in one device.

For the environment around local workloads, see our guide to agent sandboxes, and for small models running on-device, our EmbeddingGemma 2 explainer.

Sources

• Microsoft Devices Blog: Pre-order our most powerful Surface devices ever (October 7, 2026, Brett Ostrum)
• Windows hybrid intelligence vision: Windows Experience Blog
• Trade-in programme: Microsoft Trade-in

The hero image is the Open Graph image from the announcement (Source: Microsoft). Diagrams were made by cldnavi.com. Figures come from Microsoft's published specifications and footnotes. Content reflects information available on October 7, 2026 and does not include Japanese pricing or release dates.