Phison Electronics and Hsinchu MacKay Memorial Hospital held a press conference on the 28th to unveil the results of their collaboration, officially launching an on-premise generative AI system. This marks the hospital as one of the first medical institutions in Taiwan to deploy and operate a large language model (LLM) trained and inferenced entirely in-house. The event featured a live demonstration of the system generating real-time patient consultation records, with plans to gradually roll it out across the entire MacKay healthcare network.

Unlike mainstream AI models that rely heavily on cloud infrastructure, this on-premise generative AI deployment eliminates cybersecurity risks and recurring cloud service fees. Phison Electronics CEO K.S. Pua emphasized that the system is built on the company’s proprietary aiDAPTIV+ platform. The local model outperforms cloud-based inference speeds not merely through hardware power, but through precise training tailored to real-world medical environments. This enables a more streamlined and focused model architecture that delivers faster responses and improved performance during actual use.

The system currently supports multiple clinical applications, including outpatient medical record assistance, nursing voice notes, meeting summarization, and early childhood development assessments. Physicians who tested the system reported that the AI can instantly generate consultation summaries, significantly reducing administrative workload and addressing the pressing issue of repetitive, knowledge-based labor in clinical settings. Dr. Tsung-Ying Hsiao, Chairman of MacKay Memorial Hospital, stated that the AI system is expected to recover its cost within a year, citing substantial savings in human resources and improved operational efficiency.

To address the talent gap in local AI model development and training, MacKay Nursing College has established 50 on-premise AI training workstations—the first initiative of its kind in Taiwan’s healthcare sector. These will be used to cultivate professionals with hands-on experience in AI model implementation and application, supporting future expansion of AI systems in clinical settings.

For Phison, this collaboration represents a major step in commercializing its aiDAPTIV+ platform. Originally known for its NAND Flash storage solutions, the company’s pivot into on-premise AI inference and training system integration signals a promising entry into edge computing, as well as medical and industrial AI application markets.

Resource: 群聯aiDAPTIV+打入醫療場域，攜新竹馬偕搶先布局