Surface RTX Spark Dev Box opens for preorder at $5,999

Microsoft opened preorders for the Surface RTX Spark Dev Box on October 7, 2026, priced at $5,999 in the United States. The compact desktop pairs an Nvidia RTX Spark chip with 128GB of memory and comes set up for coding. It is Microsoft's bid for developers who want to run large AI models on their own desk instead of in the cloud.

What the Dev Box is

The Dev Box is a small desktop with no screen, keyboard or battery. Microsoft first showed it in June, XDA Developers notes. Microsoft's product page calls it "the powerhouse for local AI and agent development." It claims up to one petaflop of AI compute, or a quadrillion calculations per second. Pulse 2.0 says that figure uses FP4, a low-precision number format, with sparsity.

Spec	Surface RTX Spark Dev Box
Price	$5,999 (Microsoft's page shows "From $5,999.99")
Chip	Nvidia RTX Spark: 20-core CPU, Blackwell GPU with 6,144 CUDA cores (XDA)
Memory	128GB unified, shared by CPU and GPU
Storage	2TB SSD (XDA)
AI compute	Up to 1 petaflop
Thermal envelope	100W
Ports	Two USB-C, USB-A, HDMI, Ethernet, headphone jack
Operating system	Windows 11 Pro
Where to buy	Microsoft.com, United States only

Unified memory means the processor and the graphics chip share one pool of memory. That matters for AI, because a model runs fast only when it fits in memory the GPU can reach. With 128GB shared, the Dev Box can hold models that would not fit on most graphics cards.

The chassis is anodized aluminum and doubles as a heat sink, Microsoft says. The Desk Brief adds that the body is 3D-printed with 1,000 vents.

The software comes preinstalled

Microsoft lists Visual Studio Code, GitHub Copilot in the CLI and Windows Terminal, PowerShell 7, Intelligent Terminal and Coreutils for Windows. It also ships WSL, the Windows Subsystem for Linux, which runs Linux tools on Windows. Pulse 2.0 adds Git, the GitHub CLI, Python and Node to the list. Microsoft promises "a ready-to-code Windows experience, pre-configured out-of-the-box so you can build from day one," XDA reports.

The machine is part of Project Zenith, Microsoft's Windows mode built for local AI coding, according to Pulse 2.0 and XDA.

On models, Microsoft says the Dev Box runs coding models locally up to 120 billion parameters. Its page names Aion 1.0 Instruct for on-device text and agent tool calling. It also says GitHub Copilot's intelligent routing can offload up to 20% of workloads to MAI-1-Code-Flash and Nemotron-3-Super.

Price and shipping details conflict

The sources disagree on details. Microsoft's page shows "From $5,999.99." The Desk Brief says $5,999 was "the single price shown on Microsoft's launch-event slide, not a starting price." XDA rounds it to $6,000.

Shipping is less clear. XDA and Pulse 2.0 say shipments begin in November. The Desk Brief says Microsoft has announced no Dev Box ship date, and Microsoft's page gives none. "A preorder button is therefore not evidence that the desktop will ship alongside the laptops," writes The Desk Brief's Ryan Billings.

The laptop that launched beside it

Microsoft also opened preorders for the Surface Laptop Ultra, which uses the same Nvidia chip. It starts at $2,599 and is available from October 16, 2026, according to Pulse 2.0. It has a 15-inch touchscreen with 2,000 nits of peak HDR brightness and up to 128GB of memory, in a body under 18mm thick.

How it compares on price

The Desk Brief lists rival machines for local AI work:

Machine	Price	Memory
Surface RTX Spark Dev Box	$5,999	128GB
Nvidia DGX Spark	$6,950	128GB
Nvidia DGX Spark	$4,999	64GB
Apple Mac Studio M5 Max	from $2,499	up to 128GB
Apple Mac Studio M5 Ultra	from $5,499	up to 512GB

The 128GB DGX Spark's price rose in early October, The Desk Brief notes, citing ServeTheHome. Apple's new Mac Studio put more memory bandwidth behind local AI in September.

What this means for developers

• Do the cloud math first. Compare $5,999 with what your team spends each month on hosted models and cloud GPUs. The box pays off only if you run models locally most days.
• Memory decides what fits. Microsoft cites coding models up to 120 billion parameters. Check the quantized size of the models you actually use against 128GB, leaving room for the operating system.
• Test your toolchain under WSL. The Desk Brief lists GPU passthrough and CUDA support. If your stack is Linux-native, try it in WSL on an existing machine before you buy.
• Wait for a firm ship date. With sources split between November and no date at all, preorder only if a delay will not hurt your plans.
• Watch the agent controls. Pulse 2.0 says Microsoft plans agent controls through Microsoft Execution Containers, Microsoft Entra and Agent 365. Teams that let coding agents run on local machines should check what those controls cover once they ship.