A supercycle of alliances, policy, and memory will define South Korea’s AI push in 2025

South Korea's artificial intelligence (AI) push will reach a defining moment in 2025, as global alliances, national policy, and hardware demand converge to create remarkable momentum.

Nvidia and OpenAI have deepened their cooperation with South Korea, announcing plans for a large-scale AI infrastructure backed by the government and local conglomerates. At the same time, the government has pushed sovereign AI to the core of national strategy with a significant budget increase, and the National Assembly passed the Comprehensive AI Framework Act, scheduled to go into effect next year, making South Korea the first country to implement such a policy.

While companies expand the real-world applications of AI beyond the development of language models, the hardware sector is enjoying a memory supercycle driven by a surge in data center demand. This boom is creating supply bottlenecks not only for AI-specific high-bandwidth memory (HBM), but also for traditional memory semiconductors.

Deepening our relationship with a global AI giant

Nvidia is at the center of South Korea's efforts to secure sufficient AI accelerators. In October, NVIDIA CEO Jensen Hwang visited South Korea for the 2025 Asia-Pacific Economic Cooperation Conference and announced plans to supply 260,000 Nvidia Blackwell graphics processing units (GPUs) to the government, Samsung Electronics, SK Group, Hyundai Motor Group, and Naver.

Samsung, SK, and Hyundai plan to use Nvidia GPUs to enhance their manufacturing capabilities with AI factories and digital twin technology. A digital twin is a technology that mirrors real-world entities into a virtual environment. Naver expands the capacity of hyperscale data centers while powering physical AI applications. The government plans to deploy GPUs, develop sovereign AI foundation models, and build national computing centers to support AI development across businesses and industries.

OpenAI CEO Sam Altman also visited South Korea several times this year, resulting in Samsung and SK joining the Stargate Project, a $500 billion joint venture for U.S. AI infrastructure. As part of the partnership, Samsung Electronics and SK Hynix will each supply memory chips to meet OpenAI's chip needs – approximately 900,000 wafers worth of high-performance dynamic random access memory (DRAM) per month.

Deepening ties with Nvidia and OpenAI mark a turning point in South Korea's AI strategy. South Korea is moving more decisively to ensure the competitiveness of its AI infrastructure amid intensifying global competition by anchoring its AI ambitions in partnerships with the world's leading companies.

AI basic law

In January, South Korea enacted a framework law on AI development and building trust infrastructure, which is scheduled to go into effect on January 22 next year, making it the first country in the world to enact such a policy.

The law aims to promote and regulate the domestic AI industry and defines high-risk AI systems. Introducing safety, transparency and accountability obligations. and establish a governance framework that covers areas such as data use, human oversight, and accountability.

The law, widely referred to as the AI ​​Basic Act, is expected to provide regulatory clarity for companies developing or deploying AI systems in South Korea. This is expected to build trust in AI technology, reduce legal uncertainty, and accelerate AI adoption across industries by setting clear rules around liability and risk management.

However, concerns have been raised about the unclear definition of high-risk AI, the difficulties small and medium-sized businesses and startups may face in complying with the rules, and the vagueness of labeling requirements for AI-generated content. Industry observers have warned that such uncertainties could stifle innovation and delay commercialization, especially in the early stages of implementation.

In response, the government is currently making efforts to clarify the standards through enforcement orders and collaboration between industry, the government, and civil society groups. Authorities are also considering a grace period for penalties, including administrative fines, to give businesses time to adapt after the law takes effect.

Sovereign AI

With the inauguration of the Lee Jae-myung administration in June, South Korea has made AI a top policy priority and set an ambitious goal of becoming one of the world's top three AI powers based on AI sovereignty. Lee appointed AI expert Bae Kyung-hoon as minister of science, technology, information and communications, elevating the post to the level of deputy prime minister and giving him broader authority over AI, semiconductors and digital policy.

The President also launched the President's Council on a National AI Strategy to oversee the government's plans to build a national AI computing center, power generation facilities, and swathes of related infrastructure, as well as foster business and industrial growth with a focus on neural processing units and non-GPU AI accelerators.

As part of the push for AI sovereignty, the government has allocated a national budget of 10.1 trillion won ($6.94 billion) to AI, roughly three times this year's level, and launched a project to develop unique foundational models through a competitive program involving the nation's leading AI model developers.

Through these measures, the government seeks to position AI as a strategic national asset while reducing dependence on foreign high-tech companies.

Real-world applications and agent AI

The domestic AI industry has seen a rapid shift in focus this year, moving from developing language models on a large or small scale to applying AI to real-world use cases.

Large conglomerates are already deploying AI chatbots for internal knowledge management and customer engagement, automating office document processing, data analysis, and demand forecasting across their operations.

AI is also increasingly being used in advertising and content creation, and advances in translation and interpretation technology are enabling companies to expand their services to customers around the world. Although still in their infancy, agent AI systems are also emerging that perform various tasks on behalf of users, such as making restaurant and hotel reservations and completing payments.

In the manufacturing industry, physical AI is one of the most popular applications where intelligence is embedded directly into production systems, with AI being deployed for real-time process anomaly detection and automated quality inspection on production lines. Coupled with advances in robotics, companies are accelerating efforts toward broader manufacturing automation.

super cycle of memory

The rapid expansion of AI services has sharply increased the demand for data centers equipped with AI accelerators and memory chips, causing a supercycle.

The upward trend has been particularly concentrated in HBM, with memory chip makers such as Samsung Electronics and SK Hynix struggling to keep up with demand even with production running at full capacity.

This supply bottleneck is pushing up HBM prices, with companies reportedly raising prices for HBM3E chips by about 20%. Considering that the two major memory companies are planning to start mass production of next-generation HBM4 next year, the fact that the prices of current-generation products remain high is considered abnormal.

The price of general-purpose DRAM products is also rising as memory manufacturers concentrate their manufacturing capacity on HBM. DRAM contract prices have risen about 420% this year, jumping from $3.75 in January to $19.50 in November, according to market research firm TrendForce.

This impact is expected to spread to the prices of other computing devices such as smartphones.

Analysts are almost certain that both Samsung Electronics and SK Hynix will enjoy solid earnings in the coming months. Meanwhile, market attention is now shifting to how AI trends can benefit other chips such as low-power memory, application-specific integrated circuits, system semiconductors, and NAND flash memory.