AMD’s most ambitious mobile processor has quietly landed on store shelves in an unconventional form — a compact desktop workstation designed to put large language models and AI workloads on your desk, no cloud required. As of this week, Micro Center began selling a mini PC built around the Ryzen AI Max+ 395, a chip AMD calls “Strix Halo,” priced at $3,999 and loaded with 128GB of unified memory. The system ships with Windows 11 Pro or Linux support, marking one of the first concrete steps in AMD’s plan to bring x86-based local AI acceleration to developers and power users.

This isn’t a laptop. It’s a small-form-factor workstation that transforms a chip originally envisioned for premium notebooks into a desktop AI sandbox. The listing confirms that AMD and its partners are serious about challenging the idea that serious local AI inference requires a multi-thousand-dollar GPU or a cloud subscription.

The critical specs inside that $3,999 price tag

According to the Micro Center product page, the system is built around the Ryzen AI Max+ 395, the top Strix Halo silicon. That processor combines 16 Zen 5 CPU cores with a large integrated GPU based on AMD’s RDNA 3.5 architecture, plus a dedicated neural processing unit (NPU) for on-device AI. The headline figure, however, is the unified memory pool: 128GB of what is almost certainly LPDDR5X RAM shared between the CPU and GPU. In a desktop context, that means the entire 128GB can be allocated as graphics memory for AI inference, enabling workloads that would choke even a high-end discrete GPU with 24GB of VRAM.

Other details from the listing are sparse. The chassis appears to be a small vertical or flat tower, roughly the size of an Intel NUC Extreme, with a custom motherboard. Connectivity typically includes multiple USB ports, HDMI or DisplayPort outputs, Ethernet, and Wi-Fi. Power is delivered through an external brick, a common compromise for mini PCs of this caliber. The system ships with Windows 11 Pro by default, but the product description explicitly notes Linux support, which will be critical for anyone who wants to tap into AMD’s ROCm open-source AI software stack.

At $3,999, this is not a casual purchase. But when you break down the component costs — a flagship mobile processor with an oversized iGPU, a custom compact motherboard, a robust cooling solution, a high-wattage power supply, and 128GB of the fastest mobile memory — the price begins to look less like a markup and more like the cost of early-adopter silicon. For comparison, an Apple Mac Studio with M2 Ultra and 128GB of unified memory starts at $5,399, and that machine cannot natively run Windows.

What this machine actually means for your workflow

The arrival of a Strix Halo desktop workstation changes the calculus for three distinct groups of Windows users.

For AI tinkerers and local LLM enthusiasts

If you’ve been running quantized large language models on a gaming GPU and hitting VRAM limits, this box is a potential game-changer. With 128GB of unified memory, you can load 70B-parameter models at full precision or run multiple models simultaneously without offloading to system RAM. The NPU and GPU are designed to work together via AMD’s AI Engine, so inference workloads can be shunted to the most efficient processor on the fly. On Windows, this will likely rely on Microsoft’s DirectML or ONNX Runtime; on Linux, ROCm gives direct access to the GPU for frameworks like PyTorch and TensorFlow. The 16 Zen 5 cores also mean you can run a local vector database, a chat interface, and the model itself on the same machine without hitting CPU bottlenecks.

For developers targeting Windows on Arm

Oddly enough, an x86 machine might be the fastest way to prototype Windows on Arm AI applications. Developers can use the Strix Halo workstation to train or fine-tune models on x86, then deploy them to Arm-based Copilot+ PCs via ONNX or other cross-platform formats. The large unified memory pool means fewer compromises when working with large datasets, and the Windows 11 Pro license means you’re in a familiar environment with full WSL2 support.

For IT administrators and professional workstations

This is not a machine you’d drop into a standard cubicle. But for a specialized data science or engineering role — the kind where a $5,000 NVIDIA RTX workstation is the norm — a compact, quiet, and relatively power-efficient AI box has appeal. It consumes far less power under load than a Threadripper system with a discrete GPU, it can sit on a desk rather than under it, and it eliminates the need to manage multiple drivers and power cables. IT departments that have standardized on Windows will also appreciate that the OS is pre-installed and that the machine can be managed through Intune or Group Policy like any other PC.

The long road from mobile chip to desktop AI workstation

Strix Halo was never supposed to debut in a desktop. AMD announced the Ryzen AI 300 series in mid-2024, positioning the chips for premium laptops that could rival Apple’s M-series silicon on both CPU performance and AI acceleration. Laptops using lower-tier Strix Halo processors — with fewer cores and less memory — have already been announced by Asus, HP, and Lenovo, but the top-spec Max+ 395 has been conspicuously absent from the mobile market. Instead, the first retail product using the chip is this desktop machine from an unnamed OEM, sold exclusively through Micro Center.

That channel choice makes sense when you consider the economics. A $3,999 laptop with 128GB of RAM would be an extraordinarily niche product, and the thermal challenges of cooling a 16-core chip and a large iGPU in a thin chassis are non-trivial. By putting the processor in a small desktop, the OEM can use a larger heatsink, a slower-spinning fan, and a more forgiving power budget. It also sidesteps the battery-life conversation entirely — this machine is designed to be plugged in at all times.

This isn’t the first time we’ve seen a mobile chip rehoused for desktop use. Intel’s NUC line has long used laptop-class processors, and Apple’s Mac Mini and Mac Studio follow the same principle. What’s new is the explicit targeting of AI workloads and the massive memory footprint. AMD has been shouting from the rooftops that the future of AI is hybrid — some workloads in the cloud, some on-device — and Strix Halo is its opening salvo in that war.

What you should do if you’re considering this machine

First, confirm availability. Micro Center’s product page shows limited stock at select store locations and for shipping. If you live near a Micro Center, it’s worth checking in-store, as the retailer often holds back stock for walk-in customers. If you’re outside the United States, you’ll likely have to wait for other OEMs to release their own Strix Halo desktops or laptops.

Before buying, clarify your OS choice. The system ships with Windows 11 Pro, but Linux support is official. If you plan to use ROCm for AI development, Linux is the path of least resistance; AMD’s Linux driver stack is more mature, and many AI frameworks are optimized for it. Windows is a viable option if you need commercial software, game development tools, or corporate compliance tools that aren’t available on Linux. The NPU on Windows can accelerate certain Windows Studio Effects and small models through DirectML, but the heavy-lifting GPU AI work will rely on third-party library support that is still maturing.

Consider the total cost. At $3,999, this machine is within striking distance of a decent custom desktop with an NVIDIA RTX 4090 and 64GB of system RAM. That system would have much faster raw GPU compute for tasks like gaming or 3D rendering, but it would be limited to 24GB of VRAM for AI workloads. The Strix Halo workstation’s killer feature is memory capacity, not raw GFLOPS. If you need to run large models locally without quantization, this is currently one of the few x86 options that can do it without resorting to exotic multi-GPU setups.

What comes next

The Micro Center listing is almost certainly a soft launch. Expect AMD to formally announce this machine — or a branded equivalent — in the coming weeks, possibly with multiple OEMs unveiling their own takes at Computex or later this year. Software support will be the make-or-break factor. AMD has promised that its Ryzen AI software stack will unify the NPU, GPU, and CPU under a single API, but that promise is still a work in progress. For now, early adopters will need to be comfortable with some driver rough edges and less plug-and-play AI demos than they’d get on a Mac.

But the directional signal is clear: the era of powerful, integrated-AI workstations running Windows has arrived. Whether you buy this exact box or wait for the next one, the local AI landscape just got a lot more interesting.