South Korea’s Top Tech Players Race to Develop Localized AI Models, Challenging Global Giants
South Korean tech companies are making strides in the development of large language models tailored to their native language and culture, aiming to compete with global leaders such as OpenAI and Google.
In a significant move last month, the nation unveiled its most ambitious artificial intelligence initiative yet, pledging approximately $390 million to five local companies working on building large-scale foundational models. This endeavor signifies Seoul’s intention to reduce reliance on foreign AI technologies and bolster national security while maintaining tighter control over data in the AI era.
The selected organizations for this initiative include LG AI Research, SK Telecom, Naver Cloud, NC AI, and startup Upstage. Over the next six months, the government will evaluate their progress, cut underperforming companies, and continue funding the frontrunners until only two remain to lead South Korea’s sovereign AI drive.
Each company brings unique advantages to South Korea’s AI race. LG AI Research, the R&D unit of South Korean conglomerate LG Group, offers Exaone 4.0, a hybrid reasoning AI model that combines broad language processing with advanced reasoning features first introduced in their earlier Exaone Deep model.
Exaone 4.0 (32B) already demonstrates competitive performance against rivals on Artificial Analysis’s Intelligence Index benchmark, as does Upstage’s Solar Pro2. However, LG AI Research aims to improve and climb the ranks by leveraging its extensive access to real-world industry data across various sectors such as biotech and advanced materials manufacturing. They are focusing on refining this data before feeding it to their models for training purposes.
Instead of pursuing sheer scale, LG AI Research seeks to make the entire process more intelligent so that their AI can deliver practical value beyond what general-purpose models offer. Co-head Honglak Lee stated, “This is our fundamental approach.” LG plans to achieve this by offering its models through APIs and using the data generated by users of those services to train and improve the models.
SK Telecom (SKT), South Korea’s telco giant, launched its personal AI agent A. service in late 2023 and recently released its new large language model, A.X, in July. Built on Alibaba Cloud’s Qwen 2.5 open-source model, A.X 4.0 comes in two models: a 72-billion-parameter version and a lighter 7B variant.
According to SKT, A.X 4.0 processes Korean inputs approximately 33% more efficiently than GPT-4o did, highlighting its advantage with the local language. SKT also open-sourced its A.X 3.1 models earlier this summer and offers features like AI call summaries and auto-generated notes through its A. service. As of August 2025, it has amassed about 10 million subscribers.
SK Telecom’s edge lies in its versatility due to access to information from its telecom network, ranging from navigation to taxi-hailing services. The company aims to act as a bridge between cutting-edge model research and real-world impact by leveraging its telecom infrastructure, extensive user base, and proven services like A.
SK Telecom is also investing in AI infrastructure through GPUaaS, South Korea’s largest GPU-based service, and building a new hyperscale AI data center with AWS. To fill any gaps, they are partnering with Korean AI chipmaker Rebellions, securing trusted data partnerships through work with the government and universities, and fostering a global research network.
Naver Cloud, the cloud services arm of South Korea’s leading internet company, introduced its large language model, HyperClova, in 2021, followed by an upgraded version, HyperCLOVA X, in 2023, along with new products powered by the technology: CLOVA X, an AI chatbot, and Cue, a generative AI-driven search engine that competes with Microsoft’s CoPilot-enhanced Bing and Google’s AI Overview. This year they also unveiled HyperCLOVE X Think, their multimodal reasoning AI model.
Naver Cloud believes the true power of large language models (LLMs) lies in their ability to serve as connectors linking legacy systems and siloed services to improve usefulness. Naver stands out as Korea’s only company with an “AI full stack,” having built its HyperCLOVA X model from scratch and managing the massive data centers, cloud services, AI platforms, applications, and consumer services that bring the technology to life.
Similar to Google but tuned for South Korea, Naver is embedding its AI into core services like search, shopping, maps, and finance. The company’s advantage lies in its real-world data, such as its AI Shopping Guide offering recommendations based on actual consumer preferences. Other services include CLOVA Studio, which allows businesses to create custom generative AI, and CLOVA Carecall, an AI-powered check-in service designed for seniors living alone.
The Naver spokesperson emphasizes that overtaking global AI giants like OpenAI and Google depends on two factors: perfecting their “recipe” for models and securing the capital to scale them. However, instead of pursuing size, the company underscores sophistication, arguing its AI is already globally competitive at comparable scales.
Upstage is the only startup participating in the project. Launched last July, Upstage’s Solar Pro 2 model was the first Korean model recognized as a frontier model by Artificial Analysis, placing it alongside OpenAI, Google, Meta, and Anthropic, according to Soon-il Kwon, executive vice president at Upstage. While most frontier models have 100 billion to 200 billion parameters, Solar Pro 2 — with just 31 billion — outperforms global models on major Korean benchmarks.
Upstage aims to differentiate itself by focusing on real business impact rather than just benchmark performances. The startup is developing specialized models for industries like finance, law, and medicine while working to establish a Korean AI ecosystem led by “AI-native” startups.
