Surface RTX Spark Dev Box: Is a $5,999 AI Box Worth It?

Microsoft opened pre-orders for the Surface RTX Spark Dev Box on October 7, 2026, giving developers a Windows-first desktop built specifically for running large AI models locally. At $5,999 it is not an impulse buy, but the spec sheet explains the price: NVIDIA's RTX Spark platform, 128GB of unified memory, and Microsoft's claim of support for models above 120 billion parameters, all in a compact chassis that sits beside a monitor.

• Price and availability: $5,999 MSRP, pre-orders exclusively on Microsoft.com in the US, shipping begins in November per Microsoft.
• Platform: NVIDIA RTX Spark N1X with 20 CPU cores and an integrated Blackwell RTX GPU with 6,144 cores, per ServeTheHome and Liliputing.
• Memory and AI: 128GB unified LPDDR5X; up to 1 petaflop of AI performance (theoretical FP4 with sparsity, per Microsoft).
• Chassis: 326.5 x 172 x 42.9 mm, 2.0 kg, with a 2TB PCIe Gen5 SSD, Wi-Fi 7 and Bluetooth 5.4.

What Is the Surface RTX Spark Dev Box?

Microsoft calls the Dev Box a developer-first machine in its Project Zenith family, designed so builders can iterate on models locally and save cloud capacity for the jobs that really need it. It ships with Windows 11 Pro and a ready-made toolkit: Visual Studio Code, Git, GitHub CLI, GitHub Copilot, WSL, Python and Node are all preinstalled, so the setup time before your first local inference run should be short.

The Dev Box was first shown earlier this year. The news this week is that you can actually order one, alongside the Surface Laptop Ultra, which uses the same RTX Spark platform in a portable form.

RTX Spark Dev Box Specs and Ports

ServeTheHome lists the full hardware: the 20-core N1X processor, the 6,144-core Blackwell GPU sharing a single 128GB pool of LPDDR5X, and a 2TB Gen5 SSD. On the back you get HDMI 2.1b, two 10Gbps USB-C ports, a 5Gbps USB-A port, a 3.5mm audio jack and an RJ45 Ethernet jack, which STH believes is 10GbE but notes the speed is unconfirmed.

Unified memory is the key spec for local LLM work. Because the CPU and GPU share one large pool, a quantized model in the 70B to 120B parameter range can fit in memory without being split across a discrete graphics card's smaller VRAM. Our guide to how much memory local AI needs walks through the math by model size.

How Does It Compare With DGX Spark?

ServeTheHome describes the Dev Box as essentially Microsoft's Windows-centric take on NVIDIA's Linux-based DGX Spark class of developer boxes. The notable difference is networking: the Dev Box drops the high-speed ConnectX-7 QSFP ports that let DGX Spark units be clustered, which STH says is part of why it can sit at a lower price. If your workflow lives in Windows and WSL rather than a Linux-only stack, that trade will feel like the right one.

We covered the Linux side of this category with the 64GB DGX Spark configuration, and ASUS showed an RTX Spark mini PC of its own at Computex, so this platform is quickly becoming a real product category rather than a single device.

Who Should Consider the Dev Box?

This mini PC makes the most sense for AI engineers, researchers and small teams who prototype with open-weight models daily and are tired of waiting on cloud queues or paying per-hour GPU bills. Hobbyists running smaller 7B to 30B models will be well served by far cheaper hardware. For everyone in between, the Dev Box is a strong signal that serious local AI is moving onto the desk. Find more compact AI hardware in our mini computers coverage.

Sources: Microsoft Windows Blog — October 7, 2026; ServeTheHome — October 7, 2026; Liliputing — October 7, 2026.