Semiconductors & AI Accelerators

Sector Overview: The Foundation of Korea's Technology Economy

Semiconductors are not merely one sector among many in South Korea's economy; they are the foundational industry upon which the country's technological identity, export competitiveness, and geopolitical relevance rest. In 2025, semiconductor exports accounted for approximately USD 140 billion, representing the single largest category of Korean exports and a critical contributor to the nation's current account surplus. Samsung Electronics and SK Hynix together command dominant global positions in memory semiconductors, while Samsung Foundry operates the world's second-largest contract chip manufacturing business.

The K-Moonshot initiative positions semiconductors at the centre of its technology strategy through Mission 11 (Ultra-High-Performance, Low-Power AI Accelerators), which targets the development of domestically-designed AI chips that can reduce Korea's dependency on NVIDIA for AI computing hardware. This mission operates within a broader context where Korea's semiconductor sector faces both its greatest opportunity (the explosive growth in AI chip demand) and its most pressing challenges (Samsung Foundry's competitive gap with TSMC, the geopolitical weaponization of chip supply chains, and the need to extend dominance from memory into logic and AI accelerator design).

Understanding Korea's semiconductor sector requires disaggregating three distinct competitive domains: memory semiconductors (where Korea dominates globally), foundry manufacturing (where Samsung trails TSMC), and AI accelerator design (where Korea is building capability through startups). Each domain has different competitive dynamics, different strategic implications, and different relationships to the K-Moonshot programme.

Memory Semiconductors: Unchallenged Global Leadership

Korea's dominance in memory semiconductors is one of the most complete market positions in any global technology sector. Samsung Electronics holds the number one position in both DRAM (dynamic random-access memory) and NAND flash memory, the two primary categories of semiconductor memory. SK Hynix holds the number two position in DRAM and number three in NAND. Together, the two Korean companies control approximately 70 percent of the global DRAM market and over 50 percent of the global NAND market.

This duopoly, shared with American company Micron Technology as the third major player, has been sustained for decades through continuous investment in manufacturing technology, aggressive capital expenditure cycles, and a willingness to endure the memory industry's notorious boom-bust pricing cycles. Samsung and SK Hynix invest a combined total exceeding USD 40 billion annually in semiconductor capital expenditure, primarily in memory fabrication facility construction and equipment.

High Bandwidth Memory: The AI Windfall

The explosive growth in AI computing demand has produced a specific windfall for Korean memory manufacturers through High Bandwidth Memory (HBM), a specialised memory product that has become essential for AI accelerator chips. HBM stacks multiple DRAM dies vertically using through-silicon vias (TSVs), connecting them through a wide interface that delivers memory bandwidth an order of magnitude higher than conventional DRAM. Every major AI accelerator chip, including NVIDIA's H100, H200, and Blackwell GPUs, requires HBM to feed data to its processing cores at sufficient speed.

SK Hynix has established a commanding position in HBM, with an estimated 50 percent global market share. The company was the first to qualify HBM3E (the latest HBM generation) with NVIDIA, securing a substantial share of orders for the AI chip maker's data centre GPUs. Samsung, despite being the overall memory market leader, has faced widely-reported challenges in qualifying its HBM products with NVIDIA, a commercially painful gap that has allowed SK Hynix to capture disproportionate HBM market share.

The HBM market has grown from a niche product category to a market estimated at USD 25-30 billion in 2025, with growth rates exceeding 100 percent year-over-year. Industry analysts project continued rapid growth as AI infrastructure buildout expands globally, with the HBM market potentially reaching USD 60-80 billion by 2028. This growth trajectory makes HBM one of the most commercially significant product categories in the semiconductor industry and a primary driver of Korea's semiconductor export expansion.

Samsung's HBM qualification challenges with NVIDIA illustrate a broader competitive dynamic. While Samsung possesses the manufacturing technology to produce HBM, the exacting reliability and performance standards required by NVIDIA's qualification process have proven more challenging to meet than anticipated. Samsung has committed substantial engineering resources to resolving these issues, and progress in HBM3E qualification has been reported through late 2025 and early 2026, but the delay has cost Samsung significant market share in the highest-margin segment of the memory market.

Samsung Foundry: The TSMC Challenge

Samsung Foundry, the contract chip manufacturing division of Samsung Electronics, operates the world's second-largest foundry business but faces a competitive gap with TSMC (Taiwan Semiconductor Manufacturing Company) that has widened in recent years. TSMC commands approximately 60 percent of the global foundry market by revenue, while Samsung Foundry holds approximately 12-15 percent. More critically, TSMC dominates the most advanced process nodes (3nm, 5nm) where the highest-value AI chips, mobile processors, and high-performance computing chips are manufactured.

Samsung Foundry was the first to introduce Gate-All-Around (GAA) transistor architecture in mass production with its 3nm process node, a genuine technological achievement that preceded TSMC's GAA introduction. However, yield rates (the percentage of functional chips produced) at Samsung's advanced nodes have lagged behind TSMC's, resulting in higher effective manufacturing costs and reducing Samsung Foundry's attractiveness to the most demanding customers. NVIDIA, Apple, Qualcomm, AMD, and other major chip designers overwhelmingly manufacture their leading-edge products at TSMC.

The foundry competitive gap has significant implications for K-Moonshot's AI chip sovereignty objectives. Korean AI chip startups, including Rebellions, FuriosaAI, and DeepX, require access to advanced foundry manufacturing to produce competitive AI accelerators. If Samsung Foundry cannot match TSMC's yield and performance at advanced nodes, Korean AI chip designers face a choice between manufacturing domestically at potentially lower yields or outsourcing to TSMC, which would undermine the sovereignty objectives of Mission 11.

Samsung's response includes massive capital investment (a USD 17 billion fab under construction in Taylor, Texas, plus ongoing investment in Pyeongtaek, South Korea), technology roadmap acceleration toward 2nm and below, and focused efforts to improve yield through enhanced process control and manufacturing discipline. The company has also expanded its advanced packaging capabilities, recognising that chiplet-based architectures and heterogeneous integration are becoming as competitively important as transistor scaling.

The AI Accelerator Chip Gap

Korea's semiconductor sector presents a striking asymmetry: global dominance in memory and manufacturing, but near-complete absence in AI accelerator chip design. NVIDIA, an American company, commands an estimated 80-90 percent of the AI accelerator market with its GPU lineup. AMD provides the primary alternative for high-end AI training. Google, Amazon, and Microsoft have each developed proprietary AI chips for their cloud platforms. Korea, despite being the world's leading semiconductor manufacturer, designs none of the AI accelerator chips that drive the current AI revolution.

This gap is the specific target of K-Moonshot Mission 11. The mission's strategic rationale is straightforward: if AI computing is the defining technology of the current era, and if access to AI computing hardware can be weaponized through export controls (as the US has demonstrated with China), then Korea's national AI ambitions cannot rest entirely on foreign-designed accelerators. Domestic AI chip design capability is treated as a strategic necessity comparable to domestic energy production or defence manufacturing.

The Korean government's approach to closing the AI chip gap operates through multiple channels. The AI semiconductor startup trinity, Rebellions, FuriosaAI, and DeepX, receives venture capital support, government R&D grants, and preferential procurement consideration. Samsung Electronics' internal AI chip development efforts, including NPU cores for Exynos mobile processors and research-stage data centre accelerators, provide a large-company development path. And the broader ecosystem of model optimisation companies like Nota AI and software tool developers supports the software ecosystem required for AI chip adoption.

The AI Semiconductor Trinity

Korea's three primary AI chip startups each target distinct segments of the AI accelerator market, creating a portfolio approach to domestic chip sovereignty.

Rebellions, following its December 2024 merger with Sapeon Korea, is the most capitalised Korean AI chip company at approximately 2 trillion KRW valuation. The combined entity develops the Rebel and ATOM chips for large-scale data centre inference, with a 2027 IPO targeted. Rebellions' positioning as the "K-NVIDIA" reflects both the scale of its ambition and the weight of expectation placed on the company by government and investors.

FuriosaAI develops the RNGD (Renegade) AI accelerator, targeting data centre inference with an architecture emphasising ultra-low power consumption. The company rejected a reported acquisition approach from Meta in 2024, choosing to pursue an independent path with a targeted IPO in 2027. FuriosaAI's decision to remain Korean-owned reinforced the national AI chip sovereignty narrative and maintained competitive pressure within the domestic ecosystem.

DeepX focuses on edge AI inference processors, targeting the billions of AI-enabled devices (smartphones, robots, automotive, IoT) that require on-device AI processing within severe power constraints. DeepX's DX-M1 NPU, fabricated on Samsung Foundry's 5nm process, addresses the highest-volume segment of the AI chip market.

The trinity approach, supporting three companies rather than selecting a single national champion, preserves competitive dynamics and ensures that Korea develops capabilities across the full spectrum of AI chip applications. It also provides multiple shots on goal in a market where the probability of any single startup displacing NVIDIA is low, but the probability that at least one Korean AI chip company achieves meaningful market traction is substantially higher.

Advanced Packaging: The Next Competitive Frontier

The semiconductor industry's competitive frontier is shifting from transistor scaling alone to advanced packaging technologies that integrate multiple chip dies into a single package. TSMC's CoWoS (Chip-on-Wafer-on-Substrate) and InFO (Integrated Fan-Out) packaging technologies have become critical enablers of AI chip performance, as they enable the integration of AI processor dies with HBM stacks in configurations that maximise memory bandwidth and minimise power consumption.

Korea holds a unique position in advanced packaging because the technology requires both leading-edge manufacturing capability (Samsung's domain) and HBM stacking expertise (SK Hynix's domain). The convergence of these capabilities could give Korean companies a structural advantage in developing next-generation packaging solutions that combine logic chip manufacturing with HBM integration in a single vertically-integrated process. Samsung's investments in advanced packaging capacity at its Pyeongtaek campus and SK Hynix's expansion of HBM packaging facilities in Icheon reflect the strategic importance both companies attach to this technology.

TSMC's current dominance in CoWoS packaging for NVIDIA GPUs creates a capacity bottleneck that has constrained AI chip production. Korean companies' ability to offer alternative advanced packaging solutions could attract AI chip designers seeking to diversify their manufacturing supply chain, potentially improving Samsung Foundry's competitive position in the process.

US Export Controls and Geopolitical Dynamics

The US export control regime, particularly the October 2022 and subsequent restrictions on advanced semiconductor technology sales to China, has reshaped the competitive landscape of the global semiconductor industry in ways that profoundly affect Korea's position.

Korean semiconductor companies face a complex compliance environment. Samsung and SK Hynix both operate significant manufacturing facilities in China (Samsung in Xi'an for NAND, SK Hynix in Dalian and Wuxi for DRAM and NAND) that have received temporary exemptions from the most restrictive export control provisions. These exemptions, subject to periodic renewal, create ongoing uncertainty about the long-term viability of Korean semiconductor operations in China, which represent substantial sunk capital investments.

The export controls have also created indirect benefits for Korea's AI chip ecosystem. By restricting China's access to advanced AI accelerators, the controls have accelerated China's domestic AI chip development (Huawei Ascend), demonstrating that export restrictions can catalyse domestic chip development in target countries. This dynamic supports the Korean government's argument for domestic AI chip development: if access to foreign AI chips can be restricted for geopolitical reasons, developing domestic alternatives is a prudent strategic hedge even for US allies.

The Korea-US chip alliance provides a framework for managing these dynamics. Samsung's USD 17 billion fab investment in Taylor, Texas, positions the company within the US CHIPS Act ecosystem and strengthens bilateral technology relations. However, the alliance also creates potential tensions: US industrial policy favouring domestic chip manufacturing (CHIPS Act subsidies for TSMC's Arizona fabs) could compete with Samsung Foundry for customers who might otherwise manufacture in Korea.

Workforce and Talent Pipeline

Korea's semiconductor workforce is one of the most experienced and specialised in the world, with Samsung and SK Hynix together employing over 100,000 semiconductor engineers and technicians. The workforce pipeline is anchored by university programmes at KAIST, Seoul National University, POSTECH, and Sungkyunkwan University (which operates a Samsung-affiliated semiconductor engineering programme).

However, the sector faces workforce challenges. The expansion of semiconductor manufacturing globally, driven by CHIPS Act investments in the US, European Chips Act funding, and Japanese semiconductor revival programmes, is creating intense competition for semiconductor engineers. Korean engineers, among the most experienced globally, are recruitment targets for TSMC's Arizona operations, Intel's expansion programmes, and numerous international semiconductor ventures. Retaining domestic semiconductor talent against international salary premiums is a strategic human resources challenge.

The AI chip startup ecosystem compounds this talent pressure. Rebellions, FuriosaAI, DeepX, and numerous smaller startups recruit from the same engineering talent pool as Samsung and SK Hynix, creating intra-Korean competition that drives up compensation costs across the sector. The government's semiconductor workforce development programmes, including expanded university admissions and industry-sponsored graduate programmes, aim to increase the talent supply, but the multi-year lag between programme expansion and workforce output creates near-term constraints.

Investment Landscape

Korea's semiconductor sector absorbs enormous capital investment. Samsung Electronics and SK Hynix together invest over USD 40 billion annually in semiconductor capital expenditure, ranking among the largest corporate capital investors globally. This investment funds new fabrication facility construction, manufacturing equipment procurement, and R&D for next-generation process technologies.

The K-Moonshot budget framework provides additional government support through R&D grants for AI chip development, subsidised access to fabrication capacity for AI chip startups, and tax incentives for semiconductor facility investment. The Korean government's semiconductor support package includes tax credits of up to 25 percent for facility investment and expanded R&D deductions for semiconductor technology development.

The venture capital ecosystem channels substantial capital to AI semiconductor startups. The 45.5 percent of Korean venture investment flowing to AI companies in 2025 includes significant allocations to the AI chip startups, with Rebellions, FuriosaAI, and DeepX collectively raising over USD 1 billion in venture and growth capital. The National Growth Fund's 250 billion KRW investment in Rebellions represents the government's direct equity commitment to AI chip sovereignty.

Risk Assessment

Samsung Foundry competitiveness is the sector's most consequential risk. If Samsung Foundry cannot close the yield and performance gap with TSMC at advanced nodes, Korea's AI chip sovereignty objectives face a structural constraint: Korean-designed chips would need to be manufactured abroad to be competitive, undermining the sovereignty rationale. Samsung's massive capital investment and engineering resources provide cause for cautious optimism, but TSMC's lead has proven resistant to competitors' efforts for over a decade.

Geopolitical exposure through China-based manufacturing facilities creates ongoing risk. The approximately USD 30 billion in combined Samsung and SK Hynix China investments are vulnerable to changes in US export control exemptions, Chinese government policies, and broader US-China technology decoupling dynamics. A forced exit from China operations would involve substantial write-downs and capacity loss.

HBM concentration risk emerges from the sector's growing reliance on a single product category. While HBM's explosive growth has been enormously profitable for SK Hynix and increasingly for Samsung, the product's value is entirely derived from AI accelerator demand. Any significant slowdown in AI infrastructure investment, whether from technology disappointment, economic recession, or demand saturation, would disproportionately affect Korean memory companies' revenue and profitability.

AI chip startup execution risk is inherent in the semiconductor startup model. The history of AI chip startups challenging NVIDIA is littered with failures, from Graphcore's declining relevance to Habana Labs' limited post-Intel-acquisition impact. Korean AI chip startups must overcome not just technical challenges but the NVIDIA CUDA software ecosystem's deep entrenchment, a barrier that has proved more durable than hardware specifications alone.

Strategic Outlook

Korea's semiconductor sector occupies a position of extraordinary strength in memory and manufacturing, combined with an acknowledged vulnerability in AI accelerator design that K-Moonshot Mission 11 is explicitly designed to address. The sector's trajectory will be determined by three parallel competitive contests: Samsung Foundry versus TSMC in advanced manufacturing, SK Hynix versus Samsung (and Micron) in HBM, and the Korean AI chip trinity versus NVIDIA in accelerator design.

The AI revolution has transformed the semiconductor sector from a cyclical commodity business into a strategic asset class where national competitiveness in AI computing is partially determined by access to, and control over, chip design and manufacturing capability. Korea's position at the centre of the global semiconductor supply chain, as the dominant memory producer and the second-largest foundry operator, provides leverage that few nations possess. Converting this manufacturing leverage into AI chip design leadership is the central strategic challenge facing K-Moonshot's semiconductor mission, and the outcome will substantially determine whether Korea remains a semiconductor superpower in the AI era or becomes primarily a manufacturing service provider for foreign chip designers.