SEOUL, July 24 (AJP) - NVIDIA will pour $300 million into a joint research laboratory with the Korea Advanced Institute of Science and Technology to develop artificial intelligence built specifically for the Korean language and South Korea's industries, in one of the largest corporate commitments to AI research at a Korean university.
The NVIDIA-KAIST Joint AI Research Lab will be housed at the Kim Jaechul Graduate School of AI on the Seoul campus of the Korea Advanced Institute of Science and Technology (KAIST), the university said Friday. The collaboration runs for an initial five years and includes AI computing resources worth $50 million a year, supplied through NVIDIA's cloud partners in South Korea.
KAIST's announcement on Friday came as NVIDIA moves to finalize a separate joint research lab with Seoul National University, a sign that the world's dominant AI chipmaker is anchoring its Asian research strategy in South Korea's two leading research universities at once.
The KAIST lab will focus on agentic AI, systems that go beyond answering questions to planning and carrying out multi-step tasks on their own, such as analyzing information, selecting tools, and executing work with little human supervision. Researchers will build agentic models and agent systems tuned to the Korean language and to domestic industries including manufacturing, semiconductors, biotechnology, and services.
The work will draw on NVIDIA's full-stack AI technology, the company's integrated platform spanning chips, computing systems, software, and models, along with Nemotron, a family of open AI models NVIDIA has released for companies and research institutions to adapt freely. The lab plans to use Nemotron as a base for models with stronger Korean-language capability.
Beyond infrastructure, the partnership targets talent. The lab will fund at least 10 KAIST researchers each year and provide each of them with an internship at NVIDIA. The company also plans to hire top Korean AI researchers into full-time research positions, linking funding, internships, and permanent jobs into a single pipeline intended to keep ambitious researchers building careers in South Korea.
"Korea is home to world-class AI researchers and one of the most advanced technology ecosystems in the world," said Bill Dally, NVIDIA's chief scientist and senior vice president of research. "The NVIDIA-KAIST joint lab will accelerate the development of AI models and agent systems for Korea's industries, language, and future, and serve as a foundation for pioneering next-generation AI research."
Kim Hyun-woo, an NVIDIA researcher who joins the Kim Jaechul Graduate School of AI as an assistant professor next month, will direct the lab, setting its research agenda and coordinating joint work between KAIST teams and NVIDIA's global research organization. Kim, currently a postdoctoral researcher at NVIDIA working with AI researcher Yejin Choi and NVIDIA vice president Jan Kautz, won the Outstanding Paper Award at the EMNLP 2023 natural language processing conference for research on knowledge distillation, a technique for transferring capabilities between AI models. He recently released Nemotron-Personas-Korea, the first large-scale Korean persona dataset, which topped Hugging Face's trending dataset rankings for 10 days.
KAIST president Bae Choong-sik called the partnership "the starting point of a strategic partnership combining world-class AI research capability with cutting-edge AI infrastructure," and said the university would build "a global AI research hub representing the Republic of Korea."
Research topics will include large AI models optimized for Korean, industry-specific models, next-generation AI with reasoning and planning capability, and cooperation and verification among multiple AI agents. Both sides said they intend research results to flow into companies, public institutions, and national AI systems rather than remain academic. The two organizations are still finalizing details of research selection, researcher support, and infrastructure provision, and said the lab's start date and operating plan will be announced once negotiations conclude.
The lab at Seoul National University (SNU) is taking shape on a separate track. The university and NVIDIA plan to sign a memorandum of understanding this month covering joint research in physical AI and AI semiconductors, Aju Press has learned, with 10 to 20 research teams to be selected for an initial term of about three years. The plan includes NVIDIA internships for top students, a dedicated collaboration space on campus for visiting NVIDIA researchers, and jointly developed curricula, with SNU courses to be shared worldwide through NVIDIA's Deep Learning Institute education platform. A joint steering committee and a symposium in September are expected to follow the signing.
The agreement follows months of contact between the two sides, capped by Jensen Huang's visit to the university in June, when the NVIDIA chief executive toured its AI research center and robotics lab and met with students.
NVIDIA has run joint research labs with major universities in Singapore and Hong Kong. The Seoul push is widely seen in the industry as an effort to make South Korea a core Asian research base, secure AI talent, and counterbalance China's AI industry.
Copyright ⓒ Aju Press All rights reserved.
