GMKtec is launching the EVO-X3 on July 6, 2026, a $3,600 mini PC workstation featuring the AMD Ryzen AI Max+ 395 “Strix Halo” processor. The vertical, tower-style chassis abandons the previous flat-box design to accommodate a triple-fan cooling system, while bundling proprietary “OpenClaw” agent software for local-inference AI tasks.
The Shift from Compact Box to Thermal Tower
The transition from the EVO-X2 to the EVO-X3 represents a departure from traditional mini PC form factors. While the EVO-X2 struggled with thermal management and chassis-flex complaints, the EVO-X3 adopts a vertical, skyscraper-inspired architecture. By moving to a tower layout, GMKtec aims to solve the persistent fan noise and thermal throttling issues that plagued its predecessor. The unit occupies a footprint similar to an upright PlayStation 4, a configuration designed to maintain stable clock speeds for the Ryzen AI Max+ 395 during sustained professional workloads.
This design choice mirrors a broader trend in high-performance computing (HPC) where manufacturers prioritize airflow over extreme miniaturization. When dealing with a processor of this thermal design power (TDP), the physics of heat dissipation often dictate the chassis geometry. The move to a triple-fan array suggests that GMKtec is prioritizing longevity and sustained throughput over the “pocket-sized” marketing that defined the earlier EVO-X1 model.
Evaluating the Strix Halo Silicon
Despite the availability of newer silicon, GMKtec has opted to retain the AMD Ryzen AI Max+ 395. This processor remains a high-end choice due to its integration of CPU, GPU, and a dedicated NPU capable of 50 TOPS (trillions of operations per second). This exceeds the 40 TOPS baseline required for Microsoft’s Copilot+ certification, ensuring full compatibility with local AI acceleration features.
- Processor: AMD Ryzen AI Max+ 395 (Strix Halo)
- Memory: 128 GB LPDDR5X-8000 (Unified)
- Storage: Dual M.2 2280 PCIe Gen4x4 slots (up to 8 TB total)
- Inference Capability: Models up to 235 billion parameters
The decision to stick with the 395 chip over the newer 495 series has drawn scrutiny from enthusiasts, but it provides a known, stable target for the company’s proprietary “OpenClaw” software stack. By standardizing on the 395, GMKtec avoids the firmware volatility often associated with first-generation silicon, effectively treating the chip as a mature platform rather than a bleeding-edge experiment.
The Economics of Localized AI Inference
The most significant delta in the EVO-X3 launch is the pricing. At $3,600 for the base 128 GB/2 TB model, GMKtec has effectively quadrupled the entry price of its lineage since the 2024 launch of the $900 EVO-X1. This price hike reflects a shift toward the professional workstation market rather than the consumer enthusiast segment.
The value proposition rests on the “OpenClaw” agent suite. Unlike cloud-based solutions that incur per-token costs and necessitate data transmission, the EVO-X3 is designed for air-gapped or high-privacy environments. By utilizing 128 GB of unified memory, the machine can load large language models (LLMs) with high parameter counts directly into VRAM, bypassing the latency bottlenecks of network-attached storage or cloud APIs. For enterprise users concerned with intellectual property leakage, this creates a compelling, albeit expensive, alternative to external SaaS providers.
Market Positioning and Competitive Landscape
The EVO-X3 enters a crowded niche of “AI-first” workstations, competing directly with the MINIX ER939-AI Pro and the ONEXStation. All three devices target the same 128 GB memory ceiling, forcing the market to compete on thermal efficiency, software integration, and build quality. The competitive pressure to move away from “cheap-feeling” plastics is evident; GMKtec’s marketing emphasis on the steel-wrapped design is a direct reaction to the criticism the EVO-X2 received for its build quality.
As noted by hardware analysts, the integration of local AI agents is no longer a luxury feature but a requirement for high-end workstations. Developers looking for more information on the underlying hardware requirements for these local agents can reference the official AMD Ryzen AI documentation or evaluate the current state of local inference benchmarks via the IEEE Xplore digital library. The success of the EVO-X3 will ultimately depend on whether the proprietary OpenClaw software provides enough utility to justify the 300% price increase over the original EVO series.
The 30-Second Verdict
GMKtec is betting that the professional market is willing to pay a premium for local, high-memory AI inference capabilities. The transition to a vertical tower is a pragmatic engineering decision that addresses previous thermal failures. However, at a $3,600 starting price, the EVO-X3 is no longer a consumer-grade mini PC; it is a specialized tool for developers and data scientists who require 128 GB of memory and local model execution without the overhead of cloud infrastructure. Users can register for early access ahead of the July 6, 2026, shipping date to secure a $20 discount, though the overall cost remains a significant barrier to entry compared to the broader mini PC market.
