South Korea occupies an unusual position in the artificial intelligence (AI) hardware economy. Two of its leading memory firms — SK Hynix and Samsung Electronics — supply the high-bandwidth memory (HBM) on which every frontier AI system depends, and in June 2026, semiconductors accounted for 43.8 per cent of the country's global exports.
South Korea's strength sits at the bottom of the AI stack — the layered chain running from chips and memory, through data centres and compute, to models and the applications on which they are built — and the value of that layer is set by memory prices Seoul does not control. The government is now working on converting that commercial position into an ambitious national initiative spanning chips, computing infrastructure, domestic AI models, and industrial robotics — the full stack.
Whether South Korea succeeds in this initiative will depend on access to inputs it does not command: electricity, water, sufficient engineering talent, upstream materials sourced from Japan, and export-control clearance from Washington. Ottawa holds, or can host, several of these inputs, yet the AI-chip nexus remains marginal to a Canada–South Korea bilateral agenda. Moving it to the centre is the opportunity.
South Korea's distinctive position in the AI-chip nexus
Each economy in Northeast Asia has addressed the same AI-chip issue differently. Japan has rebuilt its global industrial influence around upstream chokepoints — silicon wafers, light-sensitive coatings (i.e. photoresists), specialty chemicals, and fabrication equipment — and embedded itself in allied production networks.
Taiwan, meanwhile, has expanded its pure-play foundry platform — that is, it builds chips designed by others and designs none of its own — into a wider AI-hardware ecosystem covering advanced packaging, AI servers, and system integration.
Whereas Japan's strength rests on being an indispensable provider of inputs, Taiwan's rests on manufacturing others' designs at a scale no competitor can replicate.
South Korea's position resembles neither of these. Rather, it rests on holding a duopoly in a single critical component, which Seoul is now attempting to extend across the wider stack. Memory has become a highly scarce input in AI infrastructure, and SK Hynix holds roughly two-thirds of the global HBM market by revenue, with Samsung Electronics supplying much of the remaining third.
Unlike foundry capacity, HBM is not something a rival can commission simply by deploying capital. HBM is made by stacking memory chips in layers, and the hard part is getting enough of them to work. That is learned by making them, generation after generation, not by buying equipment.
The scale of South Korea’s domestic commitment reflects a bet that demand for AI infrastructure is durable rather than cyclical.
That being said, the current buildout justifies capital being tied up for a decade. On June 29, South Korean President Lee Jae Myung announced a C$790-billion-plus corporate investment commitment to what his government refers to as the triple axis of semiconductors, physical AI, and data centres.
With this support, Samsung Electronics and SK Hynix will build new memory fabs in the country’s southwest; a separate cluster in Chungcheong that will concentrate on advanced packaging; and SK Group, GS Group, and Naver will lead a data-centre buildout projected to add more than 18 GW of capacity by 2035.
Alongside the fabs, the computing and software layers are also being built. Under an agreement reached at the 2025 APEC summit in Gyeongju, South Korea will deploy roughly 260,000 Nvidia graphics processing units (GPUs) by 2030 across government, Samsung, SK Hynix, Hyundai Motor Group, and Naver Cloud. This will accelerate the actual work of training and running AI models. Additionally, the Ministry of Science and Information and Communications Technology is funding a Sovereign AI Foundation Model Project to produce competitive Korean-language models.
What South Korea is attempting to build in terms of vertical integration is more ambitious than what either Tokyo or Taipei has attempted. It is also, however, a bet that rests on uncertain foundations. Samsung's foundry business holds less than 10 per cent of the global market, compared with Taiwan’s TSMC, which holds roughly 70 per cent. In addition, South Korea’s chip design sector remains thinly developed, with promising accelerator firms such as Rebellions and FuriosaAI forced to compete against Nvidia’s more entrenched software ecosystem. In brief, while South Korea dominates one layer of the stack, it has to buy its way into the other layers.
The problems stacking up for Seoul
The first — and most immediate — physical constraint South Korea will face in turning this ambition into reality is access to electricity. The Yongin semiconductor cluster, which currently has a local generation capacity of less than 2 GW, will require an estimated 15 to 16 GW once it reaches full operation — or nearly a quarter of the demand of the Seoul metropolitan region, with its estimated 25 million people.
In August 2026, the government, Korea Electric Power Corporation, Samsung, and SK Hynix signed agreements setting out a phased schedule for closing that gap, with the new southwestern complex receiving initial power for fab operations beginning in 2029 and Yongin not reaching full supply until 2041. The constraint, therefore, is that whereas fabs can be financed within a decade, transmission networks and electricity generation cannot.
The second constraint is talent. The Korea Semiconductor Industry Association projects the chip industry will need 304,000 workers by 2031 and is projected to fall roughly 54,000 short of that number.
In the 2024 regular university admission round, the five corporate-contract semiconductor programmes — two tied to Samsung, three to SK Hynix — offered 77 places and saw 138 admitted students decline them. Most left to pursue medicine, pharmacy, or Seoul National University's science faculties, forgoing a guaranteed job at the country's two largest chipmakers.
The third constraint sits inside the HBM stack itself: South Korea's dependence on Japan. Stacking memory chips into HBM requires materials and tools dominated by Japanese suppliers — a large share of silicon wafers comes from Shin-Etsu, and DISCO holds over 90 per cent of the global market for the equipment that grinds and cuts the wafers.
Seoul's target of raising its self-sufficiency in materials, parts, and equipment to 50 per cent by 2030 acknowledges the exposure without revealing how far South Korea is from attaining this self-sufficiency. The economy that supplies the world's AI memory is itself downstream of Japanese inputs.
The political constraints Korea faces are also significant — and they originate largely in Washington. In September, the U.S. Commerce Department revoked validated end-user status for Samsung's and SK Hynix's China facilities, replacing a standing authorization with year-by-year licensing.
Because roughly a third of both firms’ memory output is manufactured in China, a material share of Korean production is now at the mercy of renewing annual political permission by another country rather than settled commercial terms. On tariffs, under the November 2025 bilateral framework, Washington committed to giving Korean semiconductors terms "no less favorable" than those offered in a future agreement covering at least as much semiconductor trade. South Korean officials had understood this arrangement as parity with Taiwan.
The clause was tested within two months. In January 2026, Washington imposed a 25 per cent tariff on certain AI chips, then announced a Taiwan deal built on a different principle. Taiwanese firms expanding production in the U.S. may import chips tariff-free up to two and a half times the capacity they add, in exchange for US$250 billion in investment. Because Taiwan's benefit is tied to U.S. investment rather than expressed as a tariff rate, what "no less favorable" would require in South Korea's case is unclear. Taiwan also received an assurance of treatment no worse than that given to others. Interpretation of both provisions rests with the U.S.
Building a Canada–South Korea AI-chip agenda
The hurdles standing in the way of Seoul building the full AI stack create an opening for closer Canada–South Korea co-operation.
The good news is that the institutional architecture for this co-operation already exists. It includes the Canada–Korea Free Trade Agreement, the 2022 Comprehensive Strategic Partnership, a 2023 critical minerals and clean energy memorandum of understanding (MOU), and a January 2026 MOU establishing a Canada–Korea Industrial Co-operation Committee.
However, these frameworks have not yet focused on AI-chip content. The April 2026 Team Canada Trade Mission to Seoul produced nine signed agreements covering lithium, cybersecurity, tourism, and industrial equipment, but none on semiconductors or AI computing. Research funding shows the same gap — maybe more starkly. Canada's joint research call with Taiwan supports research on semiconductors and artificial intelligence, while the equivalent call with South Korea supports research on quantum technologies.
Canada's own “AI for All” strategy, launched in June 2026, includes pillars on sovereign AI compute and on trusted partnerships. Three steps would give the Canada-Korea agenda greater substance.
First, energy should be treated as a matter of industrial co-operation rather than a commodity trade. In June 2026, South Korea agreed to triple Canadian crude imports and is aiming to import 3.4 million tonnes of Canadian liquefied natural gas (LNG) annually, contingent on its participation in the LNG Canada Phase II. But South Korea's AI buildout is constrained by electricity, not by fuel supply. Cargoes of crude and LNG can be bought and even redirected. Generating capacity and transmission lines, however, cannot be imported and take a decade to build. The strategic energy security dialogue both governments co-chair should add a standing workstream on powering AI infrastructure: grid planning, small modular reactors, and joint feasibility work on situating Korean-operated compute capacity in Canadian jurisdictions with surplus firm power. South Korea can finance fabs and data centres.
Second, co-operation on critical minerals should be extended to semiconductor inputs. South Korea imports more than 95 per cent of its critical minerals, and, after China's 2023 export controls, Canada emerged as the dominant supplier of the country’s germanium, with gallium following a similar path — critical components for high-compound semiconductors and data centre power electronics. In July 2026, Ottawa committed up to C$400 million through the Canada Growth Fund to roughly double germanium and antimony capacity and potentially add gallium production. The International Energy Agency ranks both metals among those most exposed to supply vulnerability as semiconductors and AI drive demand. The joint Canada–Korea stockpiling plan, due by the end of 2026, should therefore cover semiconductor minor metals explicitly, with Korean offtake commitments attached to Canadian midstream projects — not just battery inputs.
Third, channels for AI-semiconductor research and talent should be opened. Canada's comparative advantage lies in compound semiconductors, photonics, and advanced packaging, and it has the facilities to prove it. Examples include the Canadian Photonics Fabrication Centre, IBM's Bromont packaging facility, and the C$223-million FABrIC initiative. These would help relieve Korea’s bottlenecks in HBM stacking, optical interconnects, and energy-efficient hardware.
To benefit from an exchange of talent, the next year’s batch of priorities by the Natural Sciences and Engineering Research Council of Canada with Korea's National Research Foundation should target AI semiconductors and quantum, with built-in provisions for industry participation and mobility for qualified graduate student researchers. This would help fill Korea's engineering shortfall while giving Canadian researchers better access to the world's leading memory firms. Discussions between experts and government officials in the two countries in 2025 identified one specific reciprocal opportunity: deploying Korean AI hardware in Canadian data centres.
Korea's AI-chip strategy will be settled by whether it can secure the physical and political inputs to build at the scale it has announced. Canada holds several of those inputs, and it has the frameworks through which to offer them. The opportunity is to make the AI-chip nexus a working part of the partnership.
• Edited by Erin Williams, Director of Programs, and Ted Fraser, Senior Editor, APF Canada.
