How Zuid -Korea is planning to open the best, Google, others with handgrown ai | Techcrunch

From technical giants to startups, South Korean players develop large language models that are tailored to their own language and culture, ready to compete with global heavyweights such as OpenAi and Google.

Last monthThe nation launched its most ambitious sovereign AI initiative to date and promised ₩ 530 billion (around $ 390 million), to five local companies that build large-scale fundamental models.

The movement underlines Seoul’s wish to reduce trust in foreign AI technologies, hoping to strengthen national security and to keep stricter control over data in the AI ​​era.

The organizations chosen by the Ministry of Science and ICT were competing LG AI research” SK -Telecom” Retractor” NC AIand the startup Inform.

Every six months the government will assess the progress of the first cohort, cut underperformers and continue to finance the front runners until only two remain to lead the sovereign AI drive of the country.

Each player brings a different benefit on the AI ​​race of Zuid -Korea. Techcrunch spoke with several of the selected companies about how they intend to take OpenAi, Google, Anthropic and the rest on their home grass. NC AI refused to comment.

LG AI Research: Exaone

LG AI Research, the R&D unit of the South Korean Giant LG group, offers Exaone 4.0A hybrid reasoning AI model. The latest version combines broad language processing with the advanced reasoning functions that were first introduced in the earlier Exaone Deep model of the company.

Exaone 4.0 (32b) already scores reasonably well against competitors Artificial analysisIntelligence Index Benchmark (just like that of Upstage Solar Pro2). But it is planning to improve and go up by its deep access to data from the real industry, ranging from biotech to advanced materials and production.

It links that data with a focus on refining the data before feeding on the models to train. Instead of having pure scale, LG wants to make the entire process more intelligent, so that the AI ​​can deliver a real, practical value that goes beyond what models can offer for general purposes. “This is our fundamental approach,” Co-Head Honglak Lee told Techcrunch.

LG improves its models via well-known tactics: they offer through APIs and then with the help of the real-world data generated by users of those services to train the model to improve.

“As the LG models improve, our partners can provide better services, which in turn generate a greater economic value and even richer data,” he said.

Instead of having to chase massive GPU clusters, LG AI research focuses on efficiency, he gets the most out of every chip and creating industry-specific models, he said. The goal is not to surpass the global giants, but to outsmart them with good performing but more efficient AI.

The Telco giant SK Telecom (SKT) from Zuid -Korea launched his Personal AI agent A. (pronounced A-DOT) Service far back in the end of 2023 and has just rolled out its new large language model, AX, in July.

Built on top of the Chinese Open Source model of Alibaba Cloud, Qwen 2.5″ AX 4.0 Comes in two models, a large version of 72 billion parameter and a lighter 7B version.

SK says AX 4.0 Korean inputs process around 33% more efficiently than GPT-4O, which underlines its lead over the local language. (OpenAi’s GPT 5.0 -comparison data are not available.) SKT also opened from sourced ITS AX 3.1 models Earlier this summer. In the meantime, the A.-service functions such as AI offers calls and automatically generated notes. From August 2025 it has already put on around 10 million subscribers.

SK’s Edge is the versatility, because it has access to information from his telecom network, ranging from navigation to taxi-hailing.

“The role of SK Telecom is to act as a bridge between advanced model research and impact in practice. With our telecom infrastructure, extensive user base and proven service such as AI, we bring AI directly into daily life, whether in customer service, mobility or production,” Taeyoon Kim, tells Tchcrunch.

SK Telecom also invests in AI infrastructure, with the help of GPUAAS, the largest GPU-based service in South Korea and building a new hyperscale AI data center with AWS. Whatever it is missing, it works together to obtain.

“We are building a full-stack ecosystem with Korean AI-chip maker-Rebellies, securing trusted data partnerships through work with the government and universities and promoting a worldwide research network,” said Kim. “This includes projects such as our collaboration with MIT (MGAIC), which apply foundation models to advanced production and battery and semiconductor innovation.”

Return Cloud: Hyperclova X

Naver Cloud, the cloud services arm of the leading internet company of South Korea, introduced its large language model, Hyperclova in 2021. Two years later it revealed an improved version, Hyperclova X, together with new products powered by the technology: Clovded Bing and Google, AI-Drivend. Overview. This year it also revealed its multimodal reasoning AI model, Hyperclove X Think.

Afterwards Cloud believes that the true power of LLMS is to serve as “connectors” that link legacy systems and Siled Services to improve the usefulness, according to a retirement spokesperson.

Afterwards it stands out as the only company in Korea – and one of the few in the world – that can really claim to have an ‘AI full pile’. It built up its Hyperclova X model completely and has the huge data centers, cloud services, AI platforms, applications and consumer services that bring the technology to life, the spokesperson explained.

Just like Google – but tailored to South Korea – Naver will enclose his AI into core services such as searching, shopping, cards and finances. The advantage is Real-World data. For example, it is ai -shop guide, offers recommendations based on what people actually want to buy. Other services include Clova Studio, with which companies can build tailor-made generative AI, and Clova Carecall, an AI-driven check-in service aimed at seniors who live alone.

The Naver spokesperson says that the files of global AI giants such as OpenAi and Google depends on two things: perfecting his “recipe” for models and securing the capital to scale them. Nevertheless, the company emphasizes, instead of having to chase the size, the emphasis on refinement, with the argument that its AI is already competing on similar scales worldwide.

Upstage’s Solar Pro 2

Upstage is the only startup that participates in the project. The Solar Pro 2 -model, which was launched last July, was the first Korean model that was recognized as a frontier model through Artificial analysisPlacing in the ring with OpenAi, Google, Meta and Anthropic, according to Soon-Il Kwon, Executive Vice President at Upstage.

Although most border models have 100 billion to 200 billion parameters, Solar Pro 2 with only 31 billion-beter for South Koreans and Kwon is more, Kwon told Techcrunch.

“Solar Pro 2 has performed better than global models on large Korean benchmarks. With this project, Upstage wants to achieve Korean language performance of 105% of the worldwide standard,” said Kwon.

Upstage is intended to distinguish itself by concentrating on real business impact, not just benchmarks, he said. It therefore develops specialized models for industries such as finance, rights and medicine, while insisting on building a Korean AI ecosystem led by “AI-Native” startups.