この記事はまだ翻訳されていないため、英語で表示しています。
Microsoft Surface RTX Spark Dev Box Explained: The $5,999 AI PC and Its Memory Catch
Surface RTX Spark Dev Box: an NVIDIA RTX Spark N1X desktop with 128 GB of unified memory for local AI at $5,999.99. What it does and the memory catch.
Microsoft made a Surface with no screen and no keyboard, and it costs $5,999.99. It is called the Surface RTX Spark Dev Box, a small desktop for people who build software and AI. The strange part: Microsoft's own laptop with the same memory and the same kind of chip costs $100 less.
This is not a hands-on test. Facts come from Microsoft and published reports. I found no independent speed tests of this box at the time of writing.
▶ WATCH ON YOUTUBEWatch the video
▶ WATCH ON YOUTUBEFull video
Key facts
| Surface RTX Spark Dev Box | |
|---|---|
| Price | $5,999.99, one configuration |
| Chip | NVIDIA RTX Spark N1X: 20-core CPU, 6,144-core GPU |
| Memory | 128 GB unified, shared by CPU and GPU |
| Storage | 2 TB. Neowin reports the drive is removable |
| Size (via Neowin) | 12.85 x 6.77 x 1.69 in, 4.4 lb |
| Cooling | Anodized aluminium body with 1,000 air vents, 100 W thermal envelope. The case works as the heatsink |
| Ports | 2x USB-C, USB-A, HDMI, Ethernet, headphone jack |
| Software | Windows 11 Pro with VS Code, GitHub Copilot, Git, Python, Node.js and WSL 2 set up |
| Availability | Microsoft.com only, US only. Pre-orders from October 7, 2026. The store lists November 25 |
Monitor and keyboard are sold separately. Microsoft announced it at Build in June. Satya Nadella summed up the idea: max the compute, max the memory.
Why 128 GB of unified memory matters
Most graphics cards have a small pool of memory of their own. Here the processor and the graphics chip share one big pool. Large AI models have to fit in memory before they can run, so this is the feature that matters.
Microsoft and NVIDIA call it up to one petaflop of AI computing. That is a big number on paper, not a daily speed. The 1,000 vents are a small joke about 1,000 teraflops.
What it is for
The pitch is running AI on your own desk instead of paying for the cloud.
- Every call to an online model can cost money. Do that all day with many agents and the bill grows.
- Microsoft says the box runs models above 120 billion parameters.
- In the stage demo, about 3.4 million tokens were used on the machine itself, with no per-use bill.
- Microsoft says it can hand up to 20% of GitHub Copilot work to local models.
- The Linux tools are set up to use the graphics chip from day one.
So the idea: keep everyday work on this box and save the cloud for the hardest problems.
How the price compares
| Machine | Memory | Price |
|---|---|---|
| Surface RTX Spark Dev Box | 128 GB, 2 TB | $5,999.99 |
| Surface Laptop Ultra, same chip tier | 128 GB, 1 TB | $5,899.99 |
| AMD Ryzen AI Halo | 128 GB | $3,999 list price |
The laptop costs less and includes a screen, a keyboard and a battery. The Dev Box gives you double the storage. Some store listings for the AMD box have been higher than list, so that gap can shrink.
For the laptop, see Surface Laptop Ultra explained and RTX Spark laptops compared.
The four catches
- No independent speed tests yet. The big AI claim is still Microsoft's claim.
- The GPU cannot use all 128 GB. A Microsoft footnote says the amount the GPU can address is less than the total. You can see it in Microsoft's own June demo: the screen shows 88.5 of 95.9 GB in use by the GPU. That was months before launch, and the amount can change with your setup.
- Reviews of the same chip in the laptop are mixed. Tom's Hardware saw it beat a MacBook Pro in a local AI test, 8 minutes to 10. Notebookcheck found a cheaper MacBook Pro ahead in its own tests. Those tests were on the laptop, so treat them as a clue.
- It is an expensive way to skip a cloud bill. You need to use a lot of AI before $6,000 pays for itself.
Why it matters
Microsoft is not selling this to everyday users. It is betting that developers want their AI at home, not rented by the hour. If that bet works, more computers will be designed around memory and not only speed.
Should you buy it?
- You run AI all day: it could make sense once real tests arrive.
- You want a powerful computer: the laptop or a cheaper box is probably smarter.
- You want to try local AI first: a Mac you already own can run useful models. See my Ollama guide.
Links
- Full video: https://youtu.be/d610jqGBLZg
- Product page: https://www.microsoft.com/en-us/surface/devices/surface-rtx-spark-dev-box
- Microsoft, June blog: https://blogs.windows.com/devices/2026/06/02/building-the-next-generation-of-devices-for-developers-surface-rtx-spark-dev-box/
- Neowin: https://www.neowin.net/news/microsoft-surface-rtx-spark-dev-box-is-a-6000-powerhouse/
- Notebookcheck (laptop test): https://www.notebookcheck.net/RTX-Spark-vs-M5-Pro-2-200-more-for-worse-performance.1426153.0.html
Would you pay almost $6,000 for a computer with no screen?
Not a hands-on test. Not sponsored by or affiliated with Microsoft or NVIDIA.


