South Korea Plans to Beat Giants Like OpenAI and Google in the AI Race

South Korea has embarked on its largest sovereign artificial intelligence project to date. In August 2025, the government announced an initiative in which it invested 530 billion won, equivalent to approximately 8.775 billion Czech koruna. This funding is going to five selected local companies working on large foundation models. They include LG AI Research, SK Telecom, Naver Cloud, NC AI, and the startup Upstage. The goal is to create models tailored to the Korean language and culture that could compete with products such as OpenAI's ChatGPT or Google's Gemini.

The government, specifically the Ministry of Science and ICT, will review the companies' progress every six months. Those that fail will be eliminated, ultimately leaving just two leaders to spearhead the national effort to develop domestic artificial intelligence. This approach is intended to ensure that the country reduces its dependence on foreign technology, strengthens national security, and retains control over data in the AI era. According to available information, South Korea plans to invest a total of 65 trillion won, or about 1.092 trillion Czech koruna, in artificial intelligence infrastructure by 2027, including the world's largest planned data center with a capacity of 3 GW and the purchase of 15,000 advanced graphics processors.

LG AI Research and the Exaone Model

LG AI Research, the research division of the large LG Group conglomerate, offers the Exaone 4.0 model. This hybrid model combines broad language processing with advanced reasoning capabilities first introduced in the earlier Exaone Deep version. The latest version has 32 billion parameters and, according to benchmarks from Artificial Analysis, achieves solid results compared with its competitors.

The company focuses on access to real-world industry data in fields such as biotechnology, advanced materials, and manufacturing. Instead of chasing enormous scale, it concentrates on smart data processing before training so that the model delivers practical value. Honglak Lee, co-CEO, explained that the company improves its models through APIs, where user data is used for further training. LG is not trying to outperform global giants through hardware spending, but through efficiency—maximizing the performance of each chip and creating industry-specific models. This approach is intended to ensure that its artificial intelligence offers more than general-purpose models.

SK Telecom and the A.X Model

Telecommunications giant SK Telecom launched its personal artificial intelligence agent A. (pronounced A-dot) in late 2023 and introduced the new A.X model in July 2025. This model is based on Alibaba Cloud's open-source Qwen 2.5 and comes in two versions: a large one with 72 billion parameters and a lighter one with 7 billion.

A.X 4.0 processes Korean inputs 33 percent more efficiently than OpenAI's GPT-4o, highlighting its advantage in the local language. SK Telecom openly released its A.X 3.1 models this summer. The A. service includes features such as AI-generated call summaries and automatic notes, and had gained about 10 million subscribers by August 2025. The company's advantage lies in its access to telecommunications network data, including navigation and taxi-hailing data.

Taeyoon Kim, head of the foundation model office at SK Telecom, described the company's role as a bridge between research and real-world impact. With its infrastructure, large user base, and services such as A., it brings artificial intelligence into everyday life in areas including customer service, mobility, and manufacturing. SK Telecom is investing in infrastructure, including South Korea's largest GPU-as-a-service offering and a new hyperscale data center with AWS. It is also collaborating with Rebellions on chips, with the government and universities on data, and with MIT on applications in manufacturing, batteries, and semiconductors.

Naver Cloud and HyperCLOVA X

Naver Cloud, the cloud division of South Korea's leading internet company, introduced the HyperClova model in 2021. Two years later, it released the improved HyperCLOVA X version along with products such as the CLOVA X chatbot and the Cue search engine, which competes with Microsoft's Bing with Copilot and Google's AI Overview. This year, it added the multimodal HyperCLOVA X Think reasoning model.

According to a Naver Cloud spokesperson, the company sees the strength of large language models in connecting legacy systems and isolated services to make them more useful. The company prides itself on having a complete artificial intelligence stack—from the HyperCLOVA X model built from scratch to data centers, cloud services, applications, and consumer services. Like Google, but tailored to South Korea, it integrates artificial intelligence into search, shopping, maps, and finance.

Its advantage lies in real-world data—for example, AI Shopping Guide makes recommendations based on actual purchases. Other services include CLOVA Studio for creating custom generative artificial intelligence and CLOVA Carecall for making check-in calls to seniors living alone. The spokesperson emphasized that success against OpenAI and Google depends on having the perfect model recipe and the capital to scale, with the company focusing on sophistication rather than size.

Upstage and Solar Pro 2

Upstage is the only startup in the project. Its Solar Pro 2 model, launched in July, became the first Korean model designated as frontier-class by Artificial Analysis, placing it alongside OpenAI, Google, Meta, and Anthropic. With 31 billion parameters, it outperforms global models in Korean benchmarks despite being smaller than typical frontier models with 100 to 200 billion parameters.

Soon-il Kwon, executive vice president of Upstage, explained that the model is cost-effective and better suited to Koreans. The goal is to achieve 105% of the global performance standard in the Korean language. The company differentiates itself by focusing on real-world business impact, not just benchmarks. It is developing specialized models for finance, law, and medicine and supporting an ecosystem of artificial intelligence-focused startups.

Broader Context and Support

South Korea has adopted the AI Framework Act, the first comprehensive artificial intelligence law in the Asia-Pacific region, effective from January 2026. It offers tax credits of 25–35 percent for investments in equipment and up to 50 percent for research on strategic projects. The country has designated 30 key projects in fields such as robotics, automobiles, shipbuilding, semiconductors, and public services to boost productivity and technological self-sufficiency. This plan is intended to make South Korea an attractive destination for artificial intelligence investment and position it as a leader both in Asia and globally.

Source: techcrunch.com