🧠 Food for thought

Implications, context, and why it matters.

LG and Nvidia tie-up fits South Korea’s AI push

• The work sits inside South Korea’s government-backed “Proprietary AI Foundation Model” project, a multiyear public-private effort with about 530 billion won (US$358 million) behind it to build globally competitive AI by 2027 1.
• EXAONE 4.5 has 33 billion parameters and serves as a step toward adding vision to LG’s larger flagship model, K-EXAONE. That model ranked first across all categories in the project’s first evaluation round in January 1.
• Nvidia also brings the NeMo framework, a software toolkit for building and training AI models, plus Nemotron datasets to help create Korean-specific foundation models under the project 2.
• EXAONE 4.5 comes with a non-commercial license, so it is limited to research and education. Commercial use needs a separate agreement 3.

Sovereign AI is creating specialized rivals backed by the state

• LG’s effort is part of a wider sovereign AI push that pairs domestic infrastructure with locally grounded models to cut reliance on foreign technology 2.
• That push includes a planned 120,000-graphics processing unit (GPU) data center in Paju, plus national deployments of more than a quarter-million Nvidia GPUs across sovereign clouds and AI factories 4, 2.
• State-backed models can target narrow strengths such as EXAONE’s Korean context plus industrial document analysis for contracts, technical drawings, and financial statements 5.
• For companies, the approach offers a way to compete through models tuned for regional languages or enterprise work such as analyzing technical drawings 6.