GMKtec has thrown its hat into the local AI ring with the EVO-X3, a compact vertical workstation built around AMD’s Ryzen AI Max+ 395 processor. The device starts at $3,600 and will hit early access on June 22, 2026, ahead of a broader global launch.

What GMKtec Just Announced

The EVO-X3 breaks from the typical flat mini PC by standing vertically—a design that prioritizes cooling and desktop real estate. At its core is AMD’s Ryzen AI Max+ 395, a Strix Halo chip that combines 16 Zen 5 CPU cores, an RDNA 3.5 integrated GPU with 40 compute units, and a neural processing unit (NPU) capable of 50 TOPS (tera operations per second) for AI tasks. This puts it in a different class from the NPU-toting laptop chips seen in Copilot+ PCs; the Max+ 395 is designed to handle sustained, heavy AI workloads locally.

Pricing starts at $3,600, but GMKtec hasn’t yet detailed every configuration. The processor supports up to 128 GB of unified, quad-channel ECC memory, which both the CPU and GPU can access without the usual data shuffling over a PCIe bus. That architecture is tailor-made for large language model (LLM) inference and smaller training jobs that might otherwise require a discrete GPU like an NVIDIA RTX 5000 series card. The chassis includes multiple M.2 slots for NVMe storage, a healthy selection of high-speed USB ports including USB4, and at least one 2.5 GbE Ethernet port—GMKtec tends to load their machines with connectivity, and the EVO-X3 is no exception. Wi-Fi 7 and Bluetooth 6.0 are also onboard.

GMKtec says the system is “built for AI engineers, developers, and data scientists who need desktop-friendly access to frontier models without relying on the cloud.” That pitch matters: it’s a workstation, not a consumer toy, and the price reflects the component cost and target market. Early access opens June 22, with units shipping a few weeks later. Global availability is expected in late July or early August 2026, though the company hasn’t pinned down exact dates yet.

What This Actually Means for You

The EVO-X3 lands at an intersection that will matter differently depending on who you are.

For AI developers and researchers: This is a machine that can run quantized versions of models like Llama 3 70B or Mistral Large entirely in memory, with no GPU offload to the cloud. The unified memory model eliminates one of the biggest bottlenecks in local AI work. If you’re iterating on prompts, testing RAG pipelines, or doing lightweight fine-tuning with LoRA, the EVO-X3 will likely handle it at speeds that rival a $2,000-plus standalone GPU setup—but in a box that fits next to your monitor and sips power compared to a traditional tower.

For IT departments and SMBs: Data privacy regulations are pushing more workloads on-premises. An EVO-X3 could serve as a dedicated inference server for internal chatbots, document analysis, or code assistants. At $3,600 it’s cheaper than many entry-level GPU servers and far easier to deploy.

For power users and enthusiasts: If you’ve been eyeing a Mac Studio with Apple Silicon for its unified memory but want to stay in the Windows/ Linux ecosystem, the EVO-X3 is a direct answer. AMD’s NPU and GPU combo also deliver strong performance in creative apps like DaVinci Resolve or Topaz AI, making the machine a viable mini workstation for video and photo work that leans on AI filters.

For the typical home user: Honestly, this is overkill. A $3,600 mini PC makes no sense for email, web browsing, or even casual gaming. The EVO-X3 isn’t for you unless you’re actively experimenting with local AI models or need a compact, powerful compute node.

How We Got to a $3,600 Mini PC

GMKtec is no stranger to packing big silicon into tiny boxes. Over the past few years, the Shenzhen-based company has built a reputation for affordable mini PCs using mobile Ryzen and Intel chips. The jump to AMD’s Strix Halo platform, however, represents a radical step up in price and capability.

AMD announced the Ryzen AI Max+ 395 at CES 2026, targeting high-end laptops and small form factor workstations. The chip’s unified memory architecture and powerful integrated GPU were designed to challenge both discrete mobile GPUs and Apple’s M-series chips. OEMs like ASUS and HP have already announced laptops built around it, but GMKtec is among the first to put it in a standalone desktop chassis.

This move makes sense. The AI hardware landscape is fragmenting. On one side, NVIDIA dominates with CUDA and expensive GPUs. On the other, cloud providers bill GPUs by the hour. The EVO-X3 slots into a growing middle ground: devices powerful enough to run large models locally, using less electricity than a tower stuffed with GPUs, but without locking you into a proprietary cloud. Apple’s Mac Studio with M3/M4 Ultra already proved the concept for a unified memory workstation; AMD and its partners are now doing the same for Windows and Linux users.

GMKtec’s timing is sharp. June 2026 is just as Microsoft’s Copilot+ ecosystem matures and developers start to complain about the cost and latency of cloud AI. A local machine that can handle inference without breaking the bank—or the AC circuit—hits the zeitgeist.

What You Should Do Now

If you’re interested in the EVO-X3, the first concrete step is to sign up for GMKtec’s early access program. The company hasn’t set up a reservation system yet, but its social channels and mailing list are the place to watch. Early access typically means you get priority ordering and possibly a slight discount—GMKtec has done this with past models.

Before you click buy, think hard about your workload. Ask yourself:
- Do I need to run LLMs larger than 8B parameters regularly?
- Am I hitting VRAM limits on my current dGPU?
- Is data privacy a hard requirement that blocks me from using cloud APIs?
- Can I justify $3,600 against a build with a used RTX 3090/4090, considering power, noise, and space?

For many, the math will favor the EVO-X3 if the alternative is a louder, larger, more power-hungry PC. But if your AI work is sporadic or you’re fine with smaller models that run on a mid-tier GPU, you can wait.

Also, watch for third-party benchmarks. GMKtec will tout its own numbers, but independent reviews will reveal real-world token generation speeds, thermal throttling under sustained load, and noise levels. The vertical chassis should help with cooling, but until someone runs LM Studio or llama.cpp on it for an hour, we won’t know how the 395 responds to consistent pressure.

Finally, keep an eye on competing products. Minisforum, Beelink, and other mini PC vendors will almost certainly release Strix Halo machines soon. NVIDIA’s DIGITS project (a small ARM-based AI box) also targets this space, albeit at a higher price. Competition will only improve your options.

What Comes Next

The EVO-X3 isn’t a one-off. It lights the path for an entire category: affordable, compact, local AI workstations. AMD’s Strix Halo platform gives OEMs a blueprint, and once supply chains stabilize, we’ll likely see even lower-priced variants with 8-core CPUs and less memory.

For now, GMKtec’s bet is that enough developers and small businesses want to escape cloud dependency. If early sales meet expectations, expect a rapid iteration cycle—maybe an EVO-X3 Lite or an EVO-X3 Pro with more RAM. The company is also known for listening to its community, so the I/O and cooling of the EVO-X3 will likely evolve based on early feedback.

The bigger story is the march toward local AI as the default. The EVO-X3 may be the first of many devices that make running your own models as mundane as running a spreadsheet. And for once, the price of entry stops looking like an exclusive club fee and starts resembling a reasonable pro tool investment.