Pegatron NVIDIA Omniverse + Visual AI Cuts Defect Rates 67% and Construction Time 40%

Title

Pegatron NVIDIA Omniverse + Visual AI Cuts Defect Rates 67% and Construction Time 40%

Content

Pegatron, a Fortune 500 electronics manufacturing services provider operating more than 100,000 employees and 24 sites worldwide with around $35 billion in annual revenue, deployed two NVIDIA-built platforms across its factories and reduced defect rates by 67% while cutting labor costs 7% per assembly line and shortening new factory construction time by 40%, with the company also reporting a 400% acceleration in AI agent development over the past four years. According to Andrew Hsiao, Deputy General Manager of AI Development at Pegatron, "Pegatron uses NVIDIA Omniverse to assist in developing applications for planning virtual factories in advance. This can reduce the period of time to construct a new factory by 40%," while also as Associate VP of AI Development Hsiao added, "We'll use AI Blueprint VSS to create a Visual AI Agent that we use to monitor our operations and produce more insights. For example, the agent can summarize when a task is not good enough and will perform analysis of the root cause." Pegatron addressed two bottlenecks in parallel that had blocked scaling AI in its factories: under PEGAVERSE, the company built a digital twin platform on NVIDIA Omniverse that creates physically accurate simulated environments, where production-line plans are now simulated for cycle time and utilization before any capital is spent because OEE could previously only be evaluated after operations had started and money was already committed; six virtual factories now run in parallel with physical facilities. Under PEGA AI, the company built an AI factory on NVIDIA AI Enterprise and NVIDIA DGX that lets internal teams build, train, and deploy agents using large language models and NVIDIA VILA-based vision language models, absorbing sensor data from robots via NVIDIA Isaac Sim and camera infrastructure via NVIDIA Metropolis. The combined stack produced the PEGA Visual Analytics Agent, built on the NVIDIA AI Blueprint for Video Search and Summarization (VSS), which monitors manual assembly in real time and alerts operators to anomalies such as forgotten screws, while a separate agent trains glue-dispensing robots in PEGAVERSE simulation and transfers policies to physical machines via a data flywheel that adapts glue viscosity and room temperature in real time. The same architecture is already showing downstream results at Kinsus International Technology, a Pegatron IC substrate customer whose multimodal AI agent combining image analysis with manufacturing data lifted defect-analysis accuracy from 76% to nearly 95% and cut time-to-root-cause from days to near zero. Pegatron is now extending PEGA AI agents to downstream customers and presenting the combination of PEGAVERSE and PEGA AI as a reference architecture for industrial AI deployments.