Engineers wearing cleanroom suits are inspecting the memory production process in a semiconductor cleanroom.
Driven by exploding AI demand, the Korean semiconductor industry is executing record-breaking facility investments. [Photo = Korea Business Review DB]
Semiconductors Have Stopped Being Mere 'Components'
There has long been an established formula in the semiconductor industry. When demand clusters, prices rise; when prices rise, supply increases; and when supply overflows, prices plummet—a cyclical pattern known as the 'Silicon Cycle.' Semiconductor companies have spent decades oscillating between booms and busts on top of this wave. However, this formula is now breaking down.
The year 2026 is projected to be recorded as the inaugural year when a new paradigm called 'AI memory' clearly separates the winners and losers in the market, going beyond a mere industry recovery.
While the past memory semiconductor market was a battle of low-mix high-volume production and cost reduction, technological prowess in High Bandwidth Memory (HBM)—a custom, high-performance memory tailored for customers—has now become a core indicator determining corporate survival and prosperity.
What has changed? Summarized in a single word, it is 'AI.' And HBM (High Bandwidth Memory), the specialized semiconductor required by that AI, has now become the bottleneck of the entire global technology industry, placing South Korea's Samsung Electronics and SK Hynix back on the center stage of the global economy.
What is a 'Supercycle'?
In the semiconductor industry, a supercycle refers to a rare phase where an industry upswing persists for several years, breaking away from the typical rise-and-fall pattern.
Bank of America (BofA) defined 2026 as a "supercycle similar to the 1990s boom," forecasting that global DRAM sales will surge by 51% year-on-year, NAND by 45%, and Average Selling Prices (ASP) will rise by 33% for DRAM and 26% for NAND.
While the figures are striking, what warrants closer attention is the nature of the rationale behind them. Past supercycles were triggered by the emergence of new devices such as the proliferation of PCs and the launch of smartphones.
This time is different. A strong structural momentum is forming on the demand side. Global AI data center investments alone are projected to reach $820 billion in 2026, up 32% year-on-year. With new investment expansions previewed beyond North American Big Tech—including the EU, the UAE, and the 'Stargate' project in the U.S.—large-scale demand for next-generation memory such as HBM4 is highly likely to arise.
This means that the data center infrastructure itself is transforming into a massive memory-consuming machine. Furthermore, as the center of gravity in the AI market shifts from the existing 'Training' stage to the 'Inference' stage, the semiconductor demand structure is also reorganizing from a focus on computing performance to one centered on memory capacity and cost efficiency.
What is HBM and Why Does Everyone Want It?
HBM can be explained in a single sentence — the only memory capable of supplying data at the speed AI thinks.
The "memory wall" warned by Professor Wulf in 1995 is, 30 years later, meeting the largest computing demand in history known as AI, fundamentally altering the industrial landscape.
In the past, "how fast a processor can be built" was the core question of the semiconductor industry, but the question for 2026 is this: "How fast, and how much data can be supplied to the processor?"
NVIDIA's AI accelerators (GPUs) perform tens of trillions of operations per second. The problem is that general memory cannot keep up and supply the data required for these operations. HBM is a product that innovatively increases data transmission speed by vertically stacking multiple DRAM chips, and it is currently deployed in all the latest AI accelerators, including NVIDIA's H100, H200, and Blackwell.
Samsung Electronics and SK Hynix have reportedly raised the prices of HBM3E slated for delivery in 2026 by about 20%. This phenomenon occurs as explosive demand for AI accelerators exceeds supply, proving that memory semiconductors are transforming from general commodities into premium strategic goods.
Samsung vs. SK Hynix — Same Supercycle, Different Positions
At the forefront of this supercycle, the two Korean companies hold sharply contrasting starting points.
The market estimated the 2025 HBM market share at SK Hynix 61%, Micron 22%, and Samsung Electronics 18%. SK Hynix is projected to maintain its No. 1 position in 2026 with a 50% market share by revenue, and SK Hynix's HBM sales forecast for 2026 stands at 41.2 trillion won.
While Samsung Electronics struggled with yield issues in supplying NVIDIA during the HBM3E generation, SK Hynix rapidly converted its unrivaled technological superiority into market share. Macquarie diagnosed that "the current memory shortage goes beyond a simple supply-demand imbalance and is causing bottlenecks across the entire IT supply chain," asserting that there are no silver bullets to resolve the supply shortage through 2027 or even 2028.
Samsung's counterattack is attempted at HBM4. Industry watchers estimate that Samsung Electronics' HBM sales in 2026 will reach 24 trillion won, a 189% increase from the previous year. Although falling short of SK Hynix in absolute scale, Samsung Electronics' pursuit speed is much steeper in terms of growth rate alone.
The market views that the competition between these two companies will ultimately reach a watershed at HBM4 and the subsequent generation, HBM4E.
The 250 Trillion Won 'Jackpot' — The Scale of the Investment War
The scale of profits this supercycle will generate is unprecedented. Observations have been raised in the market that the combined operating profits of both SK Hynix and Samsung Electronics in 2026 could exceed 200 trillion won and reach up to 250 trillion won.
Record-breaking earnings are leading to record-breaking reinvestments. Following 29 trillion won in 2025, SK Hynix plans to expand facility investments to the mid-30 trillion won range in 2026, pouring over 20 trillion won into the Cheongju M15X fab to foster it as a production base for HBM3E and HBM4, while executing an ultra-large-scale investment in the Yongin Semiconductor Cluster, expanding it fivefold from the initial 120 trillion won to 600 trillion won.
U.S. localization is also accelerating.
SK Hynix is constructing a $3.87 billion AI memory packaging facility in Indiana, U.S., aiming for operations by 2028. This is a strategic move that preemptively responds to the manufacturing reshoring pressure of the second Trump administration while simultaneously meeting the 'supply chain domestic diversification' demands of its major clients.
Ripple Effects on the Entire Korean Economy
This supercycle is not just a story for semiconductor companies. March semiconductor exports recorded an astonishing growth rate of 151.4% year-on-year, breaking an unprecedented record by surpassing $30 billion in monthly exports. This was thanks to soaring prices driven by explosive demand and supply shortages for artificial intelligence (AI) semiconductors, with total exports also renewing historic highs at $86.13 billion.
Analyses suggest that semiconductor exports in 2026 will reach around $188 billion, an 11% increase from the previous year. According to the Export-Import Bank of Korea, the memory semiconductor market is projected to grow by more than 30% year-on-year, and the foundry market is expected to reach a scale of $210 billion, up 20% year-on-year.
This semiconductor export boom is precisely one of the reasons why South Korea's trade balance can maintain equilibrium amidst adverse factors such as the Middle East war, high exchange rates, and high inflation.
As facility investments by Samsung Electronics and SK Hynix are concentrated, a structural trickle-down effect is occurring, opening large-scale order pipelines for domestic mid-sized companies in the semiconductor equipment, materials, and components sectors.
Conclusion: The Shadow of the Supercycle and Beyond
However, the flip side of the supercycle must also be faced. The smartphone and PC markets are facing component shortages and price hike pressures, pushed aside by the priority given to server memory supply. Smartphone shipments in 2026 are expected to experience negative growth of 2%, and PC shipments are expected to decline by 2.4%. It is the paradox where a boom for AI translates into supply pressures for the general consumer electronics ecosystem.
And as always, the supercycle is not eternal. Some research institutions are warning of the possibility of price adjustments post-2026 due to intensifying competition and expanded production capacities, requiring investors to closely monitor the peak timing of the cycle.
Nevertheless, the fundamental reason why this cycle differs from past ones lies in the nature of demand. PCs have a replacement cycle, and the smartphone market is saturated.
However, the infrastructure investments humanity is pouring into AI are still at the preliminary stage. Data centers are being built, AI models are growing larger, and inference demand is increasing exponentially at this very moment.
In a world where semiconductors have stopped being 'components' and become the 'strategic assets' of the AI era, South Korea stands at the center of fierce technological competition for the position of the most vital supplier of those assets.