Advantech shows AI Factory Brain and Physical AI with NVIDIA at GTC Berlin

AI Factory Brain: goal-driven orchestration

Advantech (TWSE: 2395) is exhibiting at Booth #7018 at NVIDIA GTC Berlin 2026, held October 20 to 22 in Berlin. The showcase centers on AI Factory Brain and Physical AI, and on how the company works with NVIDIA to link AI orchestration, real-time perception and autonomous action in industrial settings.

AI Factory Brain combines Central, Simulation and Edge AI Agents. It is supported by Advantech SKYRack server and systems, the NVIDIA Factory Operations Blueprint (FOX) and NVIDIA NemoClaw. Advantech describes this as a shift from reactive, application-specific AI toward goal-driven orchestration, in which several agents work toward shared operational objectives.

The company says the system is already deployed across 20 production lines in its own factories. Frank Huang, Senior Director of Manufacturing Center at Advantech, described how it works: you set a target such as increasing output by ten percent, and the system analyzes bottlenecks, coordinates agents and delivers an improvement plan for review. He said output per employee in Q3 was 13.2% higher than in Q2, including an 8.91% output gain delivered by five AI agents in actual operation. These figures come from Advantech itself and are not independently verified.

Physical AI at the edge

For machines that must perceive and act in real time, Advantech combines edge AI computing, multi-sensor perception, AI Agent capabilities and functional safety. Two platforms are on show:

• MIC-742-AT, powered by NVIDIA Jetson Thor, which the company says delivers up to 2,070 TFLOPS (FP4). It integrates NVIDIA Holoscan Sensor Bridge to synchronize camera and sensor data for robotic perception, and NemoClaw lets AI Agents interpret information and automate actions.
• MIC-735-IT, powered by NVIDIA IGX Thor, which adds functional safety. An outside-in safety demonstration detects people entering defined safety zones and triggers immediate responses, intended to support safer human-robot interaction.

Sim-to-Real robotics workflow

Advantech also presents a Sim-to-Real architecture that links AI computing, NVIDIA Omniverse-based simulation, robotics control and multi-sensor integration. The aim, according to the company, is to help industrial customers validate robotic behavior more efficiently and speed up adoption of AI-enabled automation.

On the development side, the NVIDIA-Certified AIR-420 Edge AI HPC can hold up to two NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition GPUs. Advantech lists digital twin simulation, synthetic data generation, virtual environment creation and AI model post-training as use cases. For deployment, the AFE-A702 robotics controller, powered by NVIDIA Jetson Thor, supports multiple GMSL 2D and 3D cameras from ecosystem partners including Intel RealSense and Stereolabs. Together with Holoscan Sensor Bridge and Advantech Robotic Suite, it is meant to provide a scalable path from simulation into industrial environments.

What it means for industrial customers

The announcement is a product showcase rather than a launch with pricing or availability details, and the source gives none. Its main substance is the claim that coordinated AI agents can run across a whole factory, backed by results from Advantech's own lines. Linda Tsai, President of the Intelligent Systems Sector and COO, said AI Factory Brain delivers "goal-driven orchestration at the factory level" while Physical AI extends that intelligence to the edge, where machines act autonomously.

For manufacturers, the pitch is a connected stack from simulation to edge hardware to factory-level orchestration, built on NVIDIA technology. Advantech invites partners and customers to see the demonstrations at its booth in Berlin.