Kiwon Tech (CEO Dong-cheol Kim), an international standard email security company, announced on the 18th that it signed a three-party strategic business agreement (MOU) with HyperExcel (CEO Joo-young Kim), a semiconductor company specializing in AI acceleration, and Jeongwonensys (CEO Deok-hee Han), an IT system integration company, to build a next-generation AI ecosystem at the 2nd floor conference room of Jeongwonensys in Gangnam-gu, Seoul.
The core of this agreement is to build a GPU+LPU hybrid system centered on Kiwon Tech's generative AI-based civil complaint analysis solution, "Minwon-e," and utilizing HyperExcel's world's first LLM-specific semiconductor, LPU. By adding Jeongwonensys's public sector distribution capabilities, the company will complete a next-generation AI infrastructure capable of reliably processing large-scale voice and text data, simultaneously promoting technological innovation and supply chain expansion in the public complaint processing market.
Founded in 2023, HyperExcel specializes in generative AI acceleration semiconductors and server solutions. It is developing a semiconductor called the LPU (LLM Processing Unit) specifically designed for LLM inference. LPU is a next-generation AI semiconductor that aims to deliver over 10x the price-performance ratio of GPUs and five times the power efficiency. The government is also focusing on it as a key candidate technology for achieving "AI semiconductor independence."
Jeongwonensys, a leading IT company in Korea, has been active in the IT industry since 1978. It possesses IT consulting capabilities and technical support expertise, encompassing everything from system infrastructure construction to integration and migration. As a distributor for Hewlett-Packard Korea, it supplies hardware and solutions, including servers and storage. Recently, it secured a KRW 23.8 billion contract to supply infrastructure equipment for the national AI project.
The GPU+LPU hybrid system, built through collaboration between three companies, enables simultaneous, high-speed processing of large-scale voice consultation data and text data, previously limited by GPU limitations. HyperExcel's LPU handles LLM inference tasks, while the GPU handles speech recognition and natural language processing, dramatically improving the processing performance of the entire system while simultaneously securing power and cost efficiency.
Through this collaboration, Kiwon Tech has expanded its scope, leveraging its software expertise, to include providing integrated hardware-software optimization solutions. By optimizing "Minwon-e" for operation on HyperExcel's LPU-based platform, processing speeds are expected to increase by 30-50% compared to previous versions, while significantly reducing operating costs.
“LPU is a next-generation AI semiconductor optimized for LLM workloads, and we are very pleased that its value can be proven by being applied to actual public services such as Minwon-e,” said Kim Joo-young, CEO of HyperExcel. “Through this collaboration, we will demonstrate the excellence of domestic AI semiconductor technology and present a new standard for AI transformation in the public sector.”
“We are proud to provide a stable operational foundation for innovative AI solutions based on the IT infrastructure construction know-how we have accumulated since our founding and our recent achievements,” said Han Deok-hee, CEO of Jeongwonensis. “We will lead public digital innovation by building a fully self-sufficient AI infrastructure based on domestic technology.”
Kim Dong-cheol, CEO of Giwon Tech, said, “This three-way collaboration is an opportunity to achieve true technological independence through the combination of international standard-based technology and domestically produced AI semiconductors,” and added, “We will do our best to ensure that Minwon e becomes the new standard for public service innovation.”
- See more related articles
You must be logged in to post a comment.