Dr. Jen-Yao Chung, VP of Business Development at Inventec, presented at OCP Global Summit 2025 on Inventec's approach to Physical AI deployment using a workload-centric edge cluster concept built on NVIDIA's three-computer architecture. Inventec, a 50-year-old Taiwan-based ODM with USD 20.1B annual revenue producing over 200 million devices per year, proposes a rack-based Physical AI Three-Computer DevKit that combines IGX edge inference nodes, OVX simulation servers, and Kubernetes-orchestrated cluster management for on-premises AI development. The solution targets use cases in robotics, smart manufacturing, medical AI, railroad safety monitoring, and precision agriculture, addressing scenarios where data cannot be sent to the cloud due to privacy, latency, or connectivity constraints.

Key Takeaways

• Inventec has built a Physical AI Three-Computer DevKit implementing NVIDIA's three-computer concept in a scalable Kubernetes-orchestrated rack cluster, combining Inventec Edge IGX Orin nodes, P90G6 OVX servers, an OCP DC-MHS compliant EdgePro management node, and NVIDIA switches.
• The cluster architecture is designed for on-premises deployments where cloud connectivity is not possible — factory floors, operational rooms, moving vehicles — supporting hardware-in-the-loop and software-in-the-loop testing with real robot hardware.
• Key verticals targeted by Inventec's edge AI cluster include robotic management and development, smart manufacturing assembly line optimization, medical AI for surgical imaging and diagnosis, railroad track monitoring, and AI-driven precision farming.
• Inventec produces over 200 million devices per year across cloud infrastructure (44M), computing (70M), and smart devices (88M), manufacturing for major North American and Chinese cloud service providers from factories spanning Taiwan, Southeast Asia, USA, and Europe.
• NVIDIA Holoscan Sensor Bridge enables GPU-Direct ingestion of high-bandwidth sensor data (cameras, lidar) from physical devices into the edge cluster, forming a critical link between the sensing edge and the IGX inference tier.
• Inventec is already deploying the three-computer cluster in its own contract manufacturing factories and piloting with medical and railroad customers, using the cluster as both a development kit and a production inferencing environment.