> ## Content Index
> Fetch the complete content index at: https://globalfeed.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# Microsoft opens preorders for Surface RTX Spark, a compact desktop built for local AI
- URL: https://globalfeed.ai/en/microsoft-opens-preorders-for-surface-rtx-spark-a-compact-desktop-built-for-local-ai/
- Published: 2026-10-08T17:43:06.000Z
- Updated: 2026-10-08T17:43:06.000Z
- Description: Microsoft has opened U.S. preorders for the Surface RTX Spark Dev Box, a compact developer desktop built on NVIDIA's RTX Spark chip, with shipments starting in November. With 128 GB of unified memory, the company says it can run models with more than 120 billion parameters locally, and it ships with
- Author: GlobalFeed Editor
- Tags: Microsoft, local models, hardware, developer tools, x-surface, dil-en, elle

**Microsoft** opened preorders for the **Surface RTX Spark Dev Box** at its event in San Francisco on October 7, 2026\. According to Microsoft's announcement, the device is sold for now only in the U.S., exclusively on Microsoft.com, and shipments begin in November. First introduced at Microsoft Build in June, the computer is built on NVIDIA's RTX Spark chip. Current prices are on Microsoft's site.

The Dev Box is a small desktop unit; the monitor and keyboard are sold separately. Microsoft designed it for developers, AI engineers and technical teams who want to run, test and fine-tune large language models and build AI agents at their desks without sending the work to the cloud. According to the June announcement, the device can run models with more than 120 billion parameters locally at interactive speeds, with a context of 1 million tokens (the amount of text a model can read at once). Microsoft says this lets teams reserve cloud resources for the workloads that truly need them.

## The RTX Spark chip and 128 GB of unified memory

RTX Spark is what NVIDIA calls a superchip: a processor and a graphics processor combined in a single package. According to NVIDIA's blog, it pairs an NVIDIA Grace CPU with up to 20 cores and a Blackwell RTX GPU with up to 6,144 cores. The Dev Box has 128 GB of unified memory that is shared dynamically between the CPU and the GPU. Because a model has to fit in the memory the GPU can reach in order to run locally, memory capacity is a deciding feature for this kind of machine. Microsoft says the device delivers up to 1 petaflop of AI compute.

According to Microsoft's product page, the chassis is made of anodized aluminum with 1,000 air vents, and the body works as a heatsink as part of a 100-watt thermal design. The company says this helps performance hold steady during long training runs. Ports include two USB-C, one USB-A, HDMI, Ethernet and a headphone jack.

## A developer setup ready out of the box

The Dev Box ships with a developer-tuned Windows 11 Pro and belongs to the Project Zenith family, Microsoft's line of devices that are ready for coding from first boot. According to the June announcement, Developer Mode is turned on and PowerShell 7 is the default shell. Visual Studio Code, Git, GitHub CLI, GitHub Copilot, Python and Node.js come preinstalled. WSL 2, which runs Linux tools inside Windows, is configured with GPU passthrough and CUDA support. NVIDIA says RTX Spark runs the full CUDA platform, which means AI tools written for NVIDIA hardware can be used on this device as well.

On the model side, Microsoft highlights three tools: the AI Toolkit for VS Code for converting, fine-tuning and evaluating models; Windows ML with NVIDIA's TensorRT-RTX for local inference; and Microsoft Foundry for moving local prototypes into production. For security, the device comes with Secured-core PC architecture, BitLocker encryption and Microsoft Defender, and organizations can manage it with Entra ID and Intune.

## A local workspace for AI agents

In the October announcement, Microsoft positions Windows as the operating system for AI agents. According to the company, local models on the Dev Box can run GitHub Copilot tasks that investigate builds and propose fixes, while Microsoft Dev Box, the company's cloud-based developer environment service, can add capacity when more power is needed. Microsoft Execution Containers keep agents isolated from the rest of the system. Microsoft also said upcoming Windows capabilities will use Entra to tell agent activity apart from user activity and will extend Agent 365 controls to local agents on the device.

On the same day, Microsoft also opened preorders for the Surface Laptop Ultra, a 15-inch laptop that uses the same RTX Spark chip; according to the company, the laptop goes on sale on October 16.

The performance figures shared come from Microsoft and NVIDIA. According to the company's footnote, the 1 petaflop figure is a theoretical peak calculated at FP4 precision using the sparsity feature; real-world speed will vary with the model and settings. Microsoft also notes that the amount of memory the GPU can address is less than 128 GB and depends on system configuration and workload. The announcements do not say at which quantization level the 120-billion-parameter models were run. For now, the device is sold only in the U.S. For a fuller picture, follow evaluations from independent researchers as well.

Sources: [Microsoft: Pre-order our most powerful Surface devices ever](https://blogs.windows.com/devices/2026/10/07/pre-order-our-most-powerful-surface-devices-ever/?ref=globalfeed.ai), [Microsoft: Introducing Surface RTX Spark Dev Box](https://blogs.windows.com/devices/2026/06/02/building-the-next-generation-of-devices-for-developers-surface-rtx-spark-dev-box/?ref=globalfeed.ai), [Microsoft: Surface RTX Spark Dev Box product page](https://www.microsoft.com/devbox?ref=globalfeed.ai), [NVIDIA: RTX Spark at the Microsoft Windows event](https://blogs.nvidia.com/blog/local-ai-rtx-spark-microsoft-windows-event/?ref=globalfeed.ai).