How South Korea plans to best OpenAI, Google, others with homegrown AI 

From technology giants to startups, South Korean players develop large linguistic models designed for their language and culture, and ready to compete with global heavy arms such as Openai and Google.

Last month, the nation launched the most ambitious Amnesty International initiative so far, as it pledged 530 billion dollars, (about 390 million dollars), to five local companies that build large -scale establishment models.

This step emphasizes Seoul’s desire to reduce dependence on foreign artificial intelligence techniques, in the hope of enhancing national security and maintaining more dominant control of data in the era of artificial intelligence.

The organizations chosen by the Ministry of Science, Information and Communications Technology for Competition were LG AI Research, SK Telecom, Naver Cloud, NC AI, and startup.

Every six months, the government will review the progress of the first regiment, reduce the weak performance, and continue to finance the applicants so that there are only two to lead the sovereign intelligence engine in the country.

Each player brings a different advantage for the South Korea artificial intelligence race. Techcrunch spoke to many selected companies about how they planned to take Openai, Google, Anthropic and the rest on home grass. NC AI refused to comment.

LG AI Research: Exaone

LG AI Research, the South Korean Giant LG LG Group, Exaone 4.0, and mixed logical AI model. The latest version mixes the wide language processing with advanced thinking features that were first presented in the company’s previous Exaone Deep model.

EXAONE 4.0 (32B) is really reasonably good results against competitors in the IQ index in artificial analysis (as in Solar Pro2 in UPSTAGE). But it plans to improve and raise the ranks through its deep access to the data of the real world industry that ranges from biotechnology to advanced materials and manufacturing.

It is the association of that data with a focus on improving data before feeding for the forms for training. Instead of chasing severe domain, LG wants to make the whole process smarter, so that Amnesty International can provide a real process value that exceeds what models can offer for general purposes. “This is our primary approach,” said Honglak Lee, co -said to Techcrunch.

LG improves its models through familiar tactics: submit them through application programming facades, then use the real world data created by users of these services to train the model to improve.

He said: “With the improvement of LG models, our partners can provide better services, which in turn generate a greater economic value and even richer data.”

However, instead of chasing the huge GPU groups, LG AI research focuses efficiency, gets the most beneficial from each slide, and creates industry models. The goal is not to overcome global giants, but to overcome them through high -performance artificial intelligence, but more efficient.

TelCo Telecom (SKT) giant in South Korea (SKT) launched A. (Spotric A-DOT) service service in late 2023 and just provided a new large linguistic model, AX, in July.

AX 4.0 is designed at the top of the Chinese source style of Alibaba Cloud, QWEN 2.5 and AX 4.0 in two models, a huge version of 72 billion of the teacher and a lighter 7B version.

SK says that AX 4.0 treats Korean inputs about 33 % more efficiently than GPT-4O, confirming the edge of its local language. (The GPT 5.0 comparative data from Openai.) SKT also opens the AX 3.1 models earlier this summer. Meanwhile, A. A. A.A. service provides advantages such as artificial intelligence summons and notes that were automatically created. As of August 2025, it was already withdrawn in about 10 million subscribers.

SK’s Edge is its diversity, because it has access to information from its network that ranges from mobility to cars.

“The role of SK Telecom is to work as a bridge between advanced typical research and the effect of the real world. With our communications infrastructure, a large -scale user base and installed service like A. We come to Amnesty International directly to daily life, whether in customer service, movement or manufacturing,” said Taeyoon Kim, head of the Foundation Model Office at SK Telecom.

SK Telecom also invests in Amnesty International’s infrastructure, using GPUAAS, the largest service based on the South Korean graphics processing unit, and the construction of a new data for artificial intelligence data with AWS. Whatever you lack it, it is a partnership to get it.

Kim said: “We are building a complete environmental system with Korean artificial intelligence makers’ rebellions, and we secure reliable data partnerships by working with the government and universities, and strengthening a global research network,” Kim said. “This includes projects such as our cooperation with MIT (MGAIC), which applies the basic models to advanced manufacturing, batteries and semi -conductors.”

Naver Cloud: Hyperclova X

Naver Cloud, the cloud services arm of the leading Internet company in South Korea, presented its great linguistic model, Hyperclova, in 2021. Two years later, a copy of the upgraded, Hyperclova X revealed, along with new technology -backed products: Clova X, AI Chatbot, Cue, which is an inspection engine AI-RVARIVE. The multimedia intelligence model this year has also unveiled the Hyperclove X Think.

Naver Cloud believes that the real LLMS power is to serve as “conductors” linking old systems and horrific services to improve interest, according to a Naver spokesman.

Naver stands out as the only Korea company – and one of the few in the world – which can really claim to have a “complete staple of artificial intelligence”. The spokesman pointed out that the Hyperclova X model from zero point and runs huge data centers, cloud services, artificial intelligence platforms, applications and consumer services that bring technology to life.

Like Google – but it was seized for South Korea – Naver includes artificial intelligence in basic services such as research, shopping, maps and financing. Its advantage is the real world data. It is a guide to artificial intelligence, for example, it provides recommendations based on what people really want to buy. Other services include Clova Studio, which allows companies to build a dedicated obstetric, and Clova Carecall, which is a check -up service for the elderly who live alone.

Naver’s spokesperson says that better than the artificial intelligence giants such as Openai and Google depend on two things: mastery “described it” for models and securing capital to expand their scope. However, instead of chasing the size, the company emphasizes the development, on the pretext that artificial intelligence is already globally competing for similar standards.

UPStage’s Solar Pro 2

UPSTAGE is the only emerging company that competes in the project. The Solar Pro 2 model, which was launched last July, was the first Korean model recognized as a border model through artificial analysis, and put it in the ring with Openai, Google, Meta and Anthropic, according to Soish-Lil Kwon, Uppstage Executive Vice President.

While most of the boundaries have 100 billion to 200 billion teachers, Solar Pro 2-31 billion-works better for South Koreans and more effective.

“Solar Pro 2 has outperformed international models on major Korean standards. With this project, UPSTAGE aims to achieve Korean language by 105 % of the global standard,” said Coone.

He said that UPSTAGE aims to distinguish itself by focusing on the influence of real business, not only the criteria. Therefore, it is developing specialized models for industries such as financing, law and medicine, while pressing the construction of an ecosystem of Korean artificial intelligence led by startups “Amnesty International”.