Minisforum AI X1 Pro in a Rack: A 2U Home for the Ryzen AI Mini PC

Minisforum's AI X1 Pro is the company's answer to the current wave of NPU-equipped mini PCs, built around a Ryzen AI chip with a dedicated neural processing block alongside the usual CPU and GPU cores. It's a different pitch than the MS-01 or MS-A2 workstation-class boxes Minisforum is better known for in homelab circles. Where those are built for virtualization and storage, the AI X1 Pro is built around sustained mixed CPU, GPU, and NPU load, which changes how you should think about mounting and cooling it in a rack.

The chassis is compact, closer to a thick mini PC than to a full workstation, but it runs warmer than a typical fanless box because all three compute blocks can be active at once. That makes airflow a bigger factor here than it is for something like a ThinkCentre Tiny sitting mostly idle.

The mount for this build:

Why this one needs 2U

It also means the usual homelab habit of tucking a mini PC into whatever gap is left on a shelf doesn't translate well to this device. A box that can run hot under a real workload deserves the same rack discipline you'd give a switch or a NAS, not the afterthought treatment a fanless thin client gets.

The AI X1 Pro's chassis is taller than the usual 1U mini PC mount can clear, mostly due to the cooling stack needed to keep an NPU-plus-GPU combo from throttling under a sustained load. Our cavity for the 19" Minisforum AI X1 Pro Rack Mount - 2U (Modular) is sized at 2U, roughly 88.9 mm under EIA-310-D, which clears the chassis height with margin for the top-venting most of these units use. Going to a shorter cavity isn't an option here the way it sometimes is with a slimmer mini PC.

Depth is a secondary concern but still worth checking against your rack before you order, particularly in a 10-inch mini rack where usable depth runs out faster than height does.

The weight is also a step up from a typical mini PC mount. The heatsink doing the work of keeping an NPU, GPU, and CPU all cool at once adds real mass, and the mount's screw bosses and support ribs are sized with that in mind rather than built around a featherweight fanless box. If you're used to handling a ThinkCentre Tiny or a NUC mount, this one will feel noticeably more substantial before it ever goes into the rack.

One detail worth checking before you order either size: confirm your port layout against the current revision of the chassis. Mini PC vendors in this category have a habit of shuffling the rear I/O between hardware revisions without changing the model name, and a port that lands an inch off from where the mount expects it can turn a clean install into a fight with a cable. A quick photo comparison against the listing before you order is a few minutes well spent.

Power and thermals

A Ryzen AI chip with an active NPU draws more than a typical fanless mini PC under mixed load, though Minisforum doesn't publish a single clean number that covers CPU, GPU, and NPU running together, so treat any figure you see as a rough guide rather than a hard ceiling. Local inference workloads in particular can hold the chip at a higher sustained draw than a typical desktop task would, since the NPU and GPU can both be active at once for longer stretches than a CPU-bound job usually runs.

The fans are audible under that kind of load, more so than a passive NAS or a fanless router, though quieter than something like a GB10-class AI workstation. In a rack, that noise moves away from your desk, which for most people is the whole point of mounting it in the first place. If the rack sits in a closet, a basement, or anywhere other than the room you work in, the noise question mostly disappears on its own.

Give it clear front and rear space the same way you would any actively cooled device. Don't block the intake with a cable bundle, and don't stack it directly below something that's already running hot.

Idle draw is a different and more forgiving picture. When the NPU isn't actively running an inference job and the chip is sitting at typical desktop load, the fan curve drops and the draw settles to something in line with a mid-range mini PC. The gap between that idle number and a sustained inference session is wide enough that it's worth sizing any UPS runtime math around the worst case rather than the number you'll see most of the time.

10-inch or 19-inch

The 10-inch mini rack version makes sense if the AI X1 Pro is joining a small homelab stack, a gateway and a switch alongside it, in a desktop enclosure. It's a common pattern for people experimenting with local inference who don't want a full server rack for one box.

The 19-inch modular version fits the same way any other 2U device slots into a standard cabinet, and the modular design leaves the door open to pairing it with a second unit later if your local AI workload grows past one box.

If you're deciding between the two chassis sizes for the first time, the honest answer is to match whatever rack you already have rather than buying a size specifically for this device. A single 2U box doesn't change the calculus between 10-inch and 19-inch the way a dense multi-switch build might. What it does change is how much room is left in that rack afterward, since two of your available units are now committed to one compute box rather than split across a gateway and a switch.

What this chip is actually good at

The NPU on this class of chip is built for efficient, sustained inference work rather than training from scratch. Running smaller local language models, image classification, or other inference-heavy tasks plays to its strengths without the power draw of a discrete GPU workstation. If your workload needs training large models or running the largest local LLMs, a GB10-class box is the more honest purchase, and this is a smaller, cheaper, and cooler step below that.

For most homelab users adding local AI to an existing stack, that's exactly the right amount of hardware. Pairing it with a NAS for model storage and a decent switch for throughput covers the realistic use case without overbuilding.

It's also worth being realistic about what the NPU does and doesn't offload. Not every piece of inference software is written to take advantage of a Ryzen AI NPU yet, and a fair amount of local AI tooling still falls back to the CPU or GPU cores regardless of what silicon is sitting idle on the die. Check whether the specific tools you plan to run actually have NPU support before assuming the chip you bought is the chip doing the work.

Cabling and placement

Route your network cable with a bit of slack rather than pulling it tight against the chassis. Active cooling on a box like this means more vibration over time than a fanless mini PC, and a loose connector from a tight cable run is an annoying thing to chase down months later.

Keep the rear I/O clear of a solid blank or another device's exhaust. A U of space above and below pays for itself the first time you run a long inference job and watch the fan curve hold steady instead of climbing.

Label the cable before you tuck it into the rack, not after. It's a small habit that saves real time the next time you're staring at a bundle of unlabeled Ethernet trying to figure out which run feeds which box. If you're running this box headless, as most rack installs end up, confirm remote management access before it goes in. A compute box that's two rack units deep in a closet is a lot less convenient to reach for a monitor-and-keyboard troubleshooting session than one sitting on a desk, so get SSH or whatever remote tooling you plan to rely on working first.

Wrap-up

The AI X1 Pro asks for the same basic rack discipline as any actively cooled compute box: get the height right, plan your circuit around sustained load rather than idle draw, and give it room to breathe. Whether you're running it in a 10-inch mini rack next to your network gear or in a 19-inch cabinet alongside a NAS, the mounting logic doesn't change.

None of this is complicated once you've done it once. The part that actually trips people up is skipping the height requirement because the box looks small enough for a 1U mount in photos, and then finding out the hard way that the chassis doesn't clear once the unit is in hand. Measure twice, order the 2U mount, and the rest of the install is routine.

Mounts in this build