South Korea launches national AI model in tech race with U.S. and China

CNBC

2 days ago

South Korea has tasked some of its biggest companies and promising startups to build a national foundational AI model using mainly domestic technology, in a rare move to keep the country apace with the U.S. and China.
The project will feature South Korean technologies from semiconductors to software, as Seoul looks to create a near self-sufficient AI industry and position itself as an alternative to China and the U.S.
The Ministry of Science and ICT (MSIT) for Korea announced that five consortia have been selected to develop the models. One is led by SK Telecom, a telecommunications giant in Korea and includes gaming firm Krafton and chip startup Rebellions among other companies.
There are other teams led by some of the country's other prominent firms including LG and Naver.
"We are going through an important juncture in terms of our technological development. So Korea, at the national level, is focusing on ensuring that we lay the technical foundation to have our competitiveness," Kim Taeyoon, head of the foundation model office at SK Telecom who also leads the company's consortium, told CNBC.
"Korea has many entities that would excel at creating a big AI industry. And we could clearly see the possibility that we are very capable of creating a good AI stack," Kim added.
A "stack" refers to various technologies that make up a product or other technology.
The initiative aims to draw on the strategic position of some of South Korea's firms and the technology they develop that are crucial to AI.
For example, SK Hynix makes high-bandwidth memory (HBM) which is critical to Nvidia's products. Samsung is also another major memory player. SK Telecom has been expanding its business into data centers. While Rebellions, which is part of SKT's consortium, is developing chips designed to handle AI workloads.
Samsung, meanwhile, has its own chip manufacturing business, also known as foundry.
"This means the country possesses the entire AI stack, from chips to cloud to AI models, and also benefits from a robust community of advanced AI researchers who are actively publishing papers and securing patents," Nick Patience, practice lead for AI at The Futurum Group, told CNBC.
Given the intricacies of technology supply chains, no one country can do it alone. The consortia will still rely on graphics processing units (GPUs) from American firm Nvidia which have become the gold standard for training AI models.
Meanwhile, SK Telecom will train the models it develops on its own Titan supercomputer, which is made up of Nvidia GPUs, as well as an AI data center the company is developing with Amazon.
SK Telecom is not new to the AI model game. The company launched a beta version of its first chatbot based on its own large language model in 2022 called "A." which is pronounced "A dot." Since then, it has developed more advanced versions of the model and chatbot.
SK Telecom's consortium plans to release its first model by the end of the year, Kim said. It will be initially focused on the market in South Korea, but could be used globally. The model will be open-source, meaning it will free for developers to use and build on, potentially with some licensing requirements.
Any AI models coming out of South Korea's project will face intense competition from players including OpenAI and Anthropic as well as many of the strong open-source offerings out of Chinese firms like Alibaba and DeepSeek.
Creating an AI model won't be a problem, given SK Telecom and other companies' already-proven track record in doing so.
The bigger challenge will be putting forward models that can compete with those coming out of frontier AI labs, which are pouring billions of dollars into research and development. Another issue will be getting traction among developers to build upon these models. That has what has made a success of other open-source models, like those from Alibaba.
SK Telecom's Kim said the goal is to create models that can rival these other companies.
"Our first goal is to create a very strong state-of-the-art open source model and we already have an example of those open source models which are on par in terms of performance with those large tech (players) like OpenAI or Anthropic," Kim told CNBC.
He added that there will be models of different sizes that can be used by different industries.
An open-source national AI model could also provide benefits by giving businesses across the country access to the latest technology without having to rely on a tech giant from abroad.
Meanwhile, South Korean AI models could be positioned as an alternative to U.S. and Chinese-developed systems.
"Beyond domestic benefits, a proven sovereign AI model presents significant export potential. Just as Korea excelled in memory chips, this could become a valuable product for other nations seeking alternatives to U.S. or Chinese systems, strengthening Korea's position in the global AI landscape," Patience said.
Underpinning this push from South Korea is the concept of "sovereign AI" that has gained traction with many nations.
This is the notion that AI models and services, which governments see as having strategic importance, should be built within a country and run on servers located domestically.
"All major nations are increasingly concerned about AI sovereignty as the US and China vie for AI dominance," The Futurum Group's Patience said.
"Given AI's growing influence on critical sectors like healthcare, finance, defense, and government, countries cannot afford to cede control of their digital intelligence to foreign entities."