South Korea’s Plan to Challenge OpenAI and Google in AI
▼ Summary
– South Korea is investing ₩530 billion ($390 million) in five local companies to develop sovereign AI models, aiming to reduce reliance on foreign technology.
– The government will conduct progress reviews every six months, eventually narrowing the funding down to just two top-performing companies.
– LG AI Research is focusing on efficiency and industry-specific models, using real-world data from sectors like biotech and manufacturing to add practical value.
– SK Telecom leverages its telecom infrastructure and user base to integrate AI into everyday services, while partnering to build a full-stack AI ecosystem.
– Naver Cloud and startup Upstage are developing competitive models tailored for the Korean language and specific industries, emphasizing real-world application over sheer scale.
South Korea is making a decisive push to establish its own leadership in artificial intelligence, directly challenging the dominance of global leaders like OpenAI and Google. A significant government investment of ₩530 billion (approximately $390 million) has been allocated to five domestic companies tasked with developing large-scale foundational AI models. This strategic initiative aims to reduce reliance on foreign technology, bolster national security, and maintain greater control over sensitive data in an increasingly AI-driven world. The selected consortium includes LG AI Research, SK Telecom, Naver Cloud, NC AI, and the startup Upstage.
The government’s plan involves a rigorous, performance-based selection process. Progress for this first cohort will be evaluated every six months, with underperforming entities being cut from the program. Funding will continue for the frontrunners until only two companies remain to spearhead the nation’s sovereign AI ambitions. Each organization brings a distinct set of strengths to the table, positioning them to compete effectively on their home turf.
LG AI Research is advancing with its hybrid reasoning model, Exaone 4.0. This model integrates broad language capabilities with sophisticated reasoning features. Rather than pursuing sheer computational scale, LG’s strategy centers on intelligent data refinement and industry-specific applications. The company leverages its deep access to real-world data from sectors like biotech and advanced manufacturing. By focusing on efficiency and practical value, LG aims to create AI that delivers tangible benefits beyond what general-purpose models can achieve. The improvement cycle is fueled by user data generated through API services, creating a feedback loop that enhances model performance for partners.
SK Telecom has launched its new large language model, A.X, built upon Alibaba Cloud’s open-source Qwen 2.5. The model comes in two sizes, a substantial 72-billion-parameter version and a more nimble 7B variant. A key claimed advantage is its processing efficiency for the Korean language, which SK says is about 33% more efficient than GPT-4o. The company’s strength lies in its versatility, drawing on data from its extensive telecom network, which includes navigation and taxi-hailing services. SKT is building a comprehensive ecosystem through partnerships, investing in GPU-as-a-service infrastructure, and collaborating with entities like Rebellions and MIT to apply AI to fields like advanced manufacturing and semiconductors.
Naver Cloud, the cloud arm of South Korea’s dominant internet company, offers its HyperCLOVA X model. The company positions itself as having a unique “AI full stack,” having built its model from the ground up while also operating the data centers, cloud services, and consumer applications that bring it to life. Similar to Google, Naver is embedding AI across its core services, search, shopping, maps, and finance, but with a specific focus on the Korean market. Its advantage stems from vast amounts of real-world user data, which powers services like an AI Shopping Guide that provides recommendations based on actual consumer behavior. The company believes competing with global giants depends on perfecting its model “recipe” and securing the capital for scaling, while prioritizing sophistication over mere size.
The sole startup in the initiative, Upstage, is competing with its Solar Pro 2 model. Despite having just 31 billion parameters, far fewer than most frontier models, it has been recognized as a frontier model by Artificial Analysis. The company highlights that Solar Pro 2 outperforms global competitors on Korean benchmarks and offers superior cost-effectiveness. Upstage’s differentiation strategy involves a sharp focus on real business impact within specific sectors such as finance, law, and medicine. The startup is also dedicated to fostering a vibrant Korean AI ecosystem led by AI-native companies, aiming to achieve a Korean language performance that exceeds the global standard.
(Source: TechCrunch)
