TAIPEI (Taiwan News) — Electronics manufacturer Pegatron on Tuesday opened its second Taiwan R&D center in the Asia New Bay Area of Kaohsiung's Qianzhen District to support its manufacturing capacity in Southeast Asia.
At the opening ceremony, Company Chair Tung Tzu-hsien (童子賢) said that Pegatron will invest about NT$700 million (US$21.5 million) over the next three years in the facility for talent development and pilot projects in AI, 5G, and smart manufacturing. He added the company also plans to expand the facility to meet growing orders, per CNA.
The company will first utilize the Kaohsiung facility to develop, validate, and test smart manufacturing models for its factories in Southeast Asia, before implementing these models in plants across the region.
The center features an AI-powered computing hub and a demonstration area for smart manufacturing equipment, Tung said. The facility is also equipped with high-performance GPU servers to support generative AI, model training and the testing of AI agents and digital twin technologies.
Pegatron Co-CEO Cheng Kuang-chih (鄭光志) said the company’s Texas plant is setting up power infrastructure and production lines. The facility is set to begin mass production in the first quarter of next year, focusing on Nvidia’s GB300 and B300 series AI servers. The plant will also produce components for electric vehicles and communication devices.
Kaohsiung Mayor Chen Chi-mai (陳其邁), who also attended the ceremony, said the city boasts a complete industrial supply chain covering semiconductors, steel, petrochemicals, metal components, aerospace, and medical equipment. He added that with the logistical advantages of Kaohsiung Port, the city is well-suited for companies to set up operations.
Since December 2024, the city government has partnered with companies, including Nvidia, Foxconn, and Pegatron, to develop AI models that highlight Taiwanese culture and language. Chen noted that most large language models are trained on global datasets and hopes that a Taiwan-focused AI platform could better represent local data.
