Alliances, policy, memory supercycle define Korea’s AI push in 2025

Nvidia CEO Jensen Huang, middle, speaks throughout the firm’s GeForce Gamer Pageant in Seoul, Oct. 30. From left are Hyundai Motor Group Govt Chair Chung Euisun, Huang and Samsung Electronics Govt Chairman Lee Jae-yong. Yonhap

Korea’s synthetic intelligence (AI) push met a decisive second in 2025 as international alliances, nationwide coverage and {hardware} demand aligned to create noteworthy momentum.

Nvidia and OpenAI deepened cooperation with Korea, unveiling plans for large-scale AI infrastructure backed by the federal government and native conglomerates. On the identical time, the federal government elevated sovereign AI to a core nationwide technique by sharply growing budgets, and the Nationwide Meeting handed a complete AI framework legislation set to take impact subsequent yr, making Korea the primary nation to implement such insurance policies.

Corporations have moved past growing language fashions to develop real-world purposes of AI, whereas the {hardware} sector is having fun with a reminiscence supercycle pushed by surging knowledge middle demand. The increase has created provide bottlenecks not solely in AI-specific high-bandwidth reminiscence (HBM) but in addition in typical reminiscence semiconductors.

President Lee Jae Myung poses with attendees of a gathering with OpenAI CEO Sam Altman, fourth from left, on the presidential workplace in Seoul, Oct. 1. From third from left are Samsung Electronics Govt Chairman Lee Jae-yong, Altman, President Lee and SK Group Chairman Chey Tae-won. Courtesy of presidential workplace

Deepening ties with international AI giants

Nvidia is positioned on the middle of Korea’s efforts to safe ample AI accelerators. In October, Nvidia CEO Jensen Huang visited Korea for 2025 Asia-Pacific Financial Cooperation conferences, and introduced a plan to provide 260,000 Nvidia Blackwell graphics processing items (GPUs) for the federal government, Samsung Electronics, SK Group, Hyundai Motor Group and Naver.

With the Nvidia GPUs, Samsung, SK and Hyundai Motor plan to empower their manufacturing capabilities with AI factories and digital twin applied sciences. Digital twin is a know-how mirroring real-world entities in a digital surroundings. Naver will develop the capability of its hyperscale knowledge facilities whereas enhancing bodily AI purposes. The federal government plans to deploy the GPUs to help AI improvement throughout corporations and industries, develop sovereign AI basis fashions and construct a nationwide computing middle.

OpenAI CEO Sam Altman additionally visited Korea a number of occasions this yr, which resulted in Samsung and SK becoming a member of the Stargate undertaking, a $500 billion three way partnership for AI infrastructure within the U.S. As a part of their partnerships, Samsung Electronics and SK hynix will every provide reminiscence chips to satisfy OpenAI’s chip calls for — about 900,000 wafers’ price of high-performance dynamic random-access reminiscence (DRAM) every month.

The deepening ties with Nvidia and OpenAI mark a turning level for Korea’s AI technique. By anchoring its AI ambitions in partnerships with main international gamers, Korea is transferring extra decisively to safe AI infrastructure competitiveness amid intensifying international competitors.

AI Primary Regulation

In January, Korea enacted the Framework Act on the Improvement of AI and the Creation of a Basis for Belief, which is ready to take impact on Jan. 22 subsequent yr, making the nation the primary on the planet to enact such a coverage.

Designed to advertise and regulate the home AI business, the act defines high-risk AI programs; introduces obligations associated to security, transparency and accountability; and establishes a governance framework masking areas comparable to knowledge use, human oversight and legal responsibility.

The act, broadly referred to as the AI Primary Regulation, is predicted to supply regulatory readability for corporations growing or deploying AI programs in Korea. That is anticipated to assist construct belief in AI applied sciences, scale back authorized uncertainty and speed up AI adoption throughout industries by setting clear guidelines on accountability and threat administration.

Nevertheless, issues have been raised over unclear definitions of high-risk AI, the problem smaller companies and startups could face in complying with the foundations, and ambiguity over labeling necessities for AI-generated content material. Business officers warn that such uncertainties may discourage innovation or gradual commercialization, significantly throughout the early levels of implementation.

In response, the federal government is now making efforts to make clear these requirements by enforcement decrees and by coordinating between business, authorities and civic teams. Authorities are additionally contemplating grace intervals for penalties, together with administrative fines, to permit corporations time to adapt as soon as the legislation takes impact.

Members of the Presidential Council on Nationwide Synthetic Intelligence Technique lower a cake in celebration of the primary 100 days of the council’s launch in Seoul, Dec. 15. Yonhap

Sovereign AI

An emblem for corporations collaborating within the authorities’s undertaking to develop Korea’s proprietary synthetic intelligence basis mannequin / Courtesy of Ministry of Science and ICT

With the beginning of the Lee Jae Myung administration in June, Korea has recognized AI as a high coverage precedence, setting an bold purpose of changing into one of many world’s high three AI powers primarily based on AI sovereignty. Lee appointed AI skilled Bae Kyung-hoon as minister of science and ICT and elevated the submit to the extent of deputy prime minister, granting broader authority over AI, semiconductors and digital insurance policies.

The president additionally launched the Presidential Council on Nationwide AI Technique to oversee the federal government’s plans to construct a belt of nationwide AI computing facilities, energy era services and associated infrastructure, whereas fostering progress in companies and industries targeted on neural processing items and non-GPU AI accelerators.

As a part of the push for AI sovereignty, the federal government earmarked 10.1 trillion gained ($6.94 billion) within the nationwide price range for AI, roughly tripling this yr’s stage, and launched a undertaking to develop a proprietary basis mannequin by a aggressive program involving the nation’s main AI mannequin builders.

By these measures, the federal government seeks to place AI as a strategic nationwide asset, whereas reducing the nation’s reliance on international tech corporations.

A robotic arm picks up an object throughout the 2025 Robotic World exhibition at KINTEX in Goyang, Gyeonggi Province, Nov. 5. Yonhap

Actual-world purposes and agentic AI

The home AI business has skilled a fast shift in focus this yr, transferring away from growing massive or small language fashions and towards making use of AI to real-world use circumstances.

Main conglomerates have already launched AI chatbots for inside information administration and buyer response, whereas automating workplace doc processing, knowledge evaluation and demand forecasting throughout their operations.

AI can be more and more being utilized in promoting and content material manufacturing, whereas advances in translation and interpretation applied sciences are enabling corporations to develop companies for international prospects. Though nonetheless at an early stage, agentic AI programs are additionally rising, executing varied duties on behalf of customers comparable to reserving eating places or motels and finishing funds.

In manufacturing, bodily AI — the place intelligence is embedded instantly into manufacturing programs — has turn out to be one of the crucial consultant utility areas, with AI deployed for real-time course of anomaly detection and automatic high quality inspection on manufacturing strains. Mixed with advances in robotics, corporations are accelerating efforts towards broader manufacturing automation.

A high-bandwidth reminiscence 4 chip is displayed at SK hynix’s sales space on the SEDEX 2025 exhibition in Seoul, Oct. 22. Yonhap

Reminiscence supercycle

The fast enlargement in AI companies resulted in exponential progress of demand for knowledge facilities powered by AI accelerators and reminiscence chips, triggering a supercycle.

The uptrend has been significantly centered on HBM, with reminiscence chip makers comparable to Samsung Electronics and SK hynix struggling to maintain tempo with demand even whereas operating manufacturing at full capability.

This provide bottleneck has pushed HBM costs increased, with the businesses reportedly elevating the provision costs of HBM3E chips by about 20 p.c. Provided that the 2 reminiscence giants are set to start mass manufacturing of next-generation HBM4 subsequent yr, the truth that costs for present era merchandise stay excessive is taken into account uncommon.

As reminiscence makers focus their manufacturing capability on HBMs, costs of general-purpose DRAM merchandise have additionally risen. In keeping with market tracker TrendForce, contract costs for DRAM surged about 420 p.c this yr, leaping from $3.75 in January to $19.50 in November.

The affect is predicted to spill over into the costs of different computing units, comparable to smartphones.

Analysts are virtually sure that each Samsung Electronics and SK hynix will get pleasure from strong earnings in coming months. In the meantime, the market’s consideration is now shifting towards how the AI pattern will profit different chips, comparable to low-power reminiscence, application-specific built-in circuits, system semiconductors and NAND flash reminiscence.