GMKtec Unveiled Desktop AI Supercomputer at IFA 2026
The EVO-X5 Pro aims to enable offline large language model execution for enterprise users.
Updated on Sept. 20, 2026 in Artificial Intelligence
Live Poll
Do you trust local AI devices more than cloud-based services for protecting your personal data?
GMKtec has announced the EVO-X5 Pro, a compact workstation designed to run 300-billion-parameter language models entirely offline. The device was unveiled at IFA 2026 and marks a shift toward local enterprise AI infrastructure.
Why it matters
By facilitating local execution of large-scale models, this system addresses the growing need for data sovereignty and reduced reliance on cloud infrastructure for high-performance AI workloads. It is built to support researchers and developers who require dedicated, high-memory capacity.
The system features an AMD Ryzen AI Max+ PRO 495 processor with 16 Zen 5 cores and an integrated NPU delivering 55 TOPS of performance. It supports up to 192GB of unified memory with a bandwidth of 273 GB/s, allowing up to 160GB to be dynamically allocated as graphics memory.
The players
GMKtec
A Shenzhen-based hardware manufacturer specializing in small-form-factor computing and enterprise workstations.
AMD
A semiconductor company that develops high-performance processors and architectures for the global data center and workstation markets.
The details
The EVO-X5 Pro achieves high-density AI processing by utilizing unified memory architecture, which allows the CPU and graphics components to share the same memory pool without redundant data transfers. It integrates enterprise-grade management through an RJ45 DASH—a remote management interface for IT hardware—and a dTPM 2.0 security chip for hardware-level encryption. The design is validated for 24/7 operation, providing the stability required for continuous AI model training and inference.
Timeline
September 19, 2026: The EVO-X5 Pro was officially unveiled at IFA 2026 in Berlin.
September 28, 2026: The device will launch for retail availability.
The Tech Race
This development marks a significant move to push high-parameter AI model execution from data centers to local desktop workstations. The system aims to compete in the enterprise space by integrating deeply with the AMD ROCm ecosystem, a critical step for hardware vendors seeking to capture the developer market.
Enterprise developers and research institutions will be the first to gain access to this local compute capability, which eliminates latency associated with cloud-based API calls. While retail availability begins September 28, 2026, compatibility with specific AI frameworks will depend on ongoing software support within the AMD ecosystem.
The takeaway
The transition to offline 300-billion-parameter model execution allows organizations to keep sensitive data within internal hardware stacks. Keep watch for the September 28, 2026 retail launch to see how the system performs against existing high-end enterprise benchmarks.
Further reading
For broader trends in local AI hardware, visit the Artificial Intelligence section.
Source note: This article includes information reported by The Berkshire Eagle.
Live Poll
Do you trust local AI devices more than cloud-based services for protecting your personal data?
