South Korea is making a significant push to establish its sovereignty in the realm of artificial intelligence, actively developing large language models (LLMs) tailored to its unique language and cultural nuances. This ambitious undertaking aims to rival global AI powerhouses like OpenAI and Google, reducing national reliance on foreign technologies, bolstering national security, and maintaining stricter control over data in the burgeoning AI era.

To spearhead this initiative, Seoul launched its most extensive sovereign AI project to date, committing ₩530 billion (approximately $390 million) to five local companies. These organizations, chosen by the Ministry of Science and ICT, include LG AI Research, SK Telecom, Naver Cloud, NC AI, and the startup Upstage. The government plans to rigorously review the progress of these companies every six months, continuing funding for the frontrunners and discontinuing support for underperformers until only two remain to lead the nation's AI drive. Each selected entity brings distinct advantages and strategies to this high-stakes competition.

LG AI Research: Exaone

LG AI Research, the dedicated R&D division of the LG Group, is advancing its Exaone 4.0, a sophisticated hybrid reasoning AI model. This latest iteration seamlessly integrates broad language processing capabilities with the advanced reasoning features initially seen in its Exaone Deep model. While Exaone 4.0 (32B) already demonstrates competitive performance against rivals on Artificial Analysis’s Intelligence Index benchmark, LG's strategy focuses on leveraging its extensive access to real-world industry data—spanning from biotech to advanced materials and manufacturing. This data is meticulously refined before being used for model training. Instead of merely pursuing larger model scales, LG prioritizes making the entire AI process more intelligent, aiming to deliver tangible, practical value that surpasses the offerings of general-purpose models. The company improves its models by offering them through APIs, then utilizing the real-world usage data to further refine and enhance the models, fostering a symbiotic relationship with its partners. Crucially, LG AI Research emphasizes efficiency over massive GPU clusters, striving to maximize performance from existing chips and develop industry-specific models, thereby outsmarting global giants through smart, efficient AI rather than sheer expenditure.

SK Telecom: A.X

SK Telecom (SKT), South Korea’s telecommunications giant, entered the AI landscape with its personal AI agent "A." service in late 2023, followed by the launch of its large language model, A.X, in July. Built upon Alibaba Cloud's open-source Qwen 2.5 model, A.X 4.0 is available in two versions: a robust 72-billion-parameter model and a lighter 7B version. SKT proudly states that A.X 4.0 processes Korean inputs approximately 33% more efficiently than GPT-4o, highlighting its significant local language advantage. Earlier this summer, SKT also open-sourced its A.X 3.1 models. The "A." service, which had already garnered around 10 million subscribers by August 2025, provides features such as AI call summaries and auto-generated notes. SKT's primary edge lies in its versatility, stemming from its direct access to extensive information across its telecom network, encompassing services from navigation to taxi-hailing. The company positions itself as a crucial bridge between cutting-edge AI research and practical, real-world applications, integrating AI into everyday life through customer service, mobility, and manufacturing. Furthermore, SK Telecom is heavily investing in AI infrastructure, utilizing GPUaaS—South Korea’s largest GPU-based service—and collaborating with AWS to build a new hyperscale AI data center. SKT actively pursues partnerships to complement its capabilities, forging a full-stack ecosystem with Korean AI chipmaker Rebellions, securing trusted data partnerships with government and academic institutions, and fostering a global research network, including a collaboration with MIT (MGAIC) focused on applying foundation models to advanced manufacturing, battery, and semiconductor innovation.

Naver Cloud: HyperCLOVA X

Naver Cloud, the cloud services division of South Korea's leading internet company, first introduced its large language model, HyperClova, in 2021. This was significantly upgraded in 2023 with the unveiling of HyperCLOVA X, alongside new AI-powered products such as CLOVA X, an AI chatbot, and Cue, a generative AI-driven search engine designed to compete with Microsoft’s CoPilot-enhanced Bing and Google’s AI Overview. This year also saw the introduction of its multimodal reasoning AI model, HyperCLOVA X Think. Naver Cloud views LLMs as vital "connectors" that bridge legacy systems and siloed services, thereby enhancing overall utility. Naver distinguishes itself as Korea's sole company—and one of a select few globally—that can credibly claim an "AI full stack." This means it developed its HyperCLOVA X model from scratch and also operates the vast data centers, cloud services, AI platforms, applications, and consumer services that bring the technology to fruition. Similar to Google, but specifically tuned for the South Korean market, Naver integrates its AI deeply into core services like search, shopping, maps, and finance. Its significant advantage is its wealth of real-world data, enabling services such as its AI Shopping Guide, which offers recommendations based on actual purchasing behaviors. Other offerings include CLOVA Studio for businesses to build custom generative AI, and CLOVA Carecall, an AI-powered check-in service for elderly individuals living alone. Naver believes that outperforming global AI giants hinges on perfecting its model "recipe" and securing the necessary capital for scaling, emphasizing sophistication over sheer size, and maintaining that its AI is already globally competitive at comparable scales.

Upstage: Solar Pro 2

Upstage represents the sole startup participating in this national AI initiative. Its Solar Pro 2 model, launched last July, gained recognition as the first Korean model to be designated a frontier model by Artificial Analysis, placing it in direct competition with offerings from OpenAI, Google, Meta, and Anthropic. While most frontier models boast parameters ranging from 100 billion to 200 billion, Solar Pro 2, with a significantly smaller 31 billion parameters, achieves superior performance for South Korean users and proves to be more cost-effective. Upstage asserts that Solar Pro 2 has outperformed global models on key Korean benchmarks, with the project aiming to elevate its Korean language performance to 105% of the global standard. Upstage differentiates itself by prioritizing tangible business impact over mere benchmark scores, developing specialized AI models for sectors such as finance, law, and medicine. Concurrently, it is actively fostering a Korean AI ecosystem, championed by "AI-native" startups.

These diverse approaches and technological strengths underline South Korea's determined effort to carve out a leading position in the global AI landscape, leveraging local expertise and strategic investments to build a robust sovereign AI ecosystem.