The government-led artificial intelligence (AI) foundation model project is gaining momentum with the intensification of the competition among the four consortia in the second-round evaluation. This phase sees the inclusion of new specialized partners focusing on various sectors such as 3D AI, large language model (LLM) inference chips, and high-quality training data. Recently, the Ministry of Science and ICT selected a consortium led by Motif Technologies to proceed to the project’s second round after three teams – LG AI Research, SK Telecom, and Upstage – advanced from the initial evaluations in January.
The consortia are now actively seeking partners to enhance their technical capabilities before the next evaluation, transforming the model-building competition into a quest to establish a globally competitive sovereign AI stack that encompasses models to practical deployment. For instance, LG AI Research has partnered with AI infrastructure and solutions provider Elice to advance its model K-EXAONE across public and private sectors. Elice will utilize its modular data center infrastructure to offer stable application programming interfaces (APIs) for the AI model, providing managed AI platforms for public agencies and enterprises without operational hassle.
On the other hand, Upstage is strengthening its consortium by collaborating with AI semiconductor startup HyperAccel and physical AI startup RLWRLD. HyperAccel is developing an LLM processing unit (LPU) to enhance inference efficiency and reduce power consumption for LLM services. By optimizing inference acceleration for generative AI workloads and creating a full-stack software platform, the team aims to boost Korea’s AI infrastructure and achieve global cost competitiveness.
Additionally, Motif Technologies has welcomed 3D AI startup N.Light and AI training data platform Crowdworks to its consortium to construct a 300-billion-parameter LLM that will morph into vision-language-action models (VLAs). N.Light will focus on developing an AI-based 3D data pipeline to generate high-precision 3D computer-aided design models and large synthetic datasets for training VLA models. Crowdworks, as the core data provider, will specialize in developing datasets for intelligent reasoning and utilize its data preprocessing solution to make complex documents understandable for AI systems.
Overall, the collaboration between these consortia and their specialized partners highlights a significant advancement in the development of AI technology, aiming to create robust models that are not only large but also efficient and applicable to real-world scenarios.
About the Author
