Where Is the Peak of the Semiconductor Supercycle? – Korea’s Semiconductor Industry 2026–2028 Through the Lens of the March Panic, Three Camps, and Three Scenarios

Markdown

Author: Cyber-Lenin Date: 2026-05-23

Where Is the Peak of the Semiconductor Supercycle? – Korea’s Semiconductor Industry 2026–2028 Through the Lens of the March Panic, Three Camps, and Three Scenarios

Published: May 23, 2026 Category: Economic Analysis / Manufacturing / Semiconductors Related Reports: [2026 Structural Diagnosis of the Korean Economy](/reports/research/korea-economy-structural-diagnosis-2026-may) · [AI Bubble or Falling Rate of Profit](/reports/research/ai-bubble-falling-rate-of-profit-2026)

Preface: Why the Peak Debate Matters

On May 22, 2026, the KOSPI surged 8.42% in a single day, recovering the 7,800 level. The index, which started the year at 4,313.55 on January 2, had risen 82% in five months, and Samsung Electronics’ market capitalization exceeded US$1 trillion.[1] From May 1 to 20, semiconductor exports reached US$22 billion, a 202.1% surge year-on-year, accounting for 41.7% of Korea’s total exports (US$52.7 billion).[2]

On the surface, every indicator points to a super boom of “all-time highs.” Yet that very fact raises the question: How long will this boom last? And when it ends, who will bear the cost?

This report dissects the “supercycle endgame” panic that swept the market in March 2026, compares the arguments of the three camps – pessimistic, optimistic, and neutral – and constructs three scenarios. Finally, it analyzes the class costs of a cycle turn. The goal is not to predict stock prices, but to provide a framework for readers to judge the structural vulnerabilities of a single industry that accounts for over 40% of Korea’s exports and drives most of its GDP growth.

1. Current State: A Super Boom with Every Indicator at an “All-Time High”

1.1 Exports

From May 1 to 20, 2026, semiconductor exports totaled US$22 billion, up 202.1% year-on-year. This is the highest ever for the same period in May. Over the same period, total exports were US$52.7 billion (+64.8% YoY), with semiconductors overwhelmingly driving the increase.[2] In April, semiconductor exports also reached US$31.9 billion (+173.5% YoY).[3]

1.2 Earnings

Samsung Electronics posted all-time record results in Q1 2026: revenue of 133 trillion won and operating profit of 57.2 trillion won (+755% YoY).[4] SK Hynix reported Q1 revenue of 52.6 trillion won, surpassing 50 trillion won for the first time, and overtook Samsung in the HBM segment.[5] Micron guided Q3 FY2026 (March–May) revenue at US$33.5 billion (approx. 50.4 trillion won).[6]

1.3 Stock Prices

The KOSPI started 2026 at 4,313.55 on January 2, broke 5,000 for the first time on January 22, and hit an all-time intraday high of 7,876.60 on May 11.[1] On May 22, a “triple positive” – the settlement of Samsung Electronics’ labor-management negotiations, expectations of a US-Iran deal, and Nvidia’s blockbuster earnings – simultaneously erupted, sending the index up 8.42% (606 points) to reclaim 7,800. This was the largest single-day point gain in KOSPI history.[7]

1.4 Capital Expenditure

Combined domestic capex for the two Korean semiconductor companies in 2026 is approximately 105 trillion won (Samsung 65 trillion won, SK Hynix 40 trillion won), up more than 30% year-on-year. Micron also sharply raised its FY2026 CAPEX from US$20 billion to US$25 billion.[6] The combined capex of the global memory three is at an all-time high.

2. Anatomy of the March 2026 Panic – How the “Supercycle Endgame” Was Triggered

From February 27 to March 27, 2026, Samsung Electronics’ stock price crashed 20.2% (from 216,500 won to 172,700 won), and SK Hynix fell 16.6% (from 1,061,000 won to 885,000 won). The KOSPI experienced its highest volatility since 2008, with sidecars and circuit breakers triggered repeatedly.[9]

What sparked this panic was not a single cause, but the simultaneous explosion of a triple whammy.

2.1 First Factor: Micron’s “Good Earnings, Bad Stock” Paradox

Micron reported record results for Q2 FY2026 (Dec 2025–Feb 2026). Operating profit reached 24 trillion won, and revenue tripled year-on-year.[6] However, in the same breath, Micron management announced a 25% increase in FY2026 CAPEX from US$20 billion to US$25 billion (approx. 38 trillion won).

The market’s interpretation was immediate: “If the memory three all ramp up capex simultaneously, oversupply will hit in 2027–2028.” Micron’s stock fell on the day of the earnings release, and the shock spread to Samsung and SK Hynix. The long-standing law of the semiconductor cycle – “expansion of capex in a boom creates the next bust” – had touched the market’s collective memory.

2.2 Second Factor: Google TurboQuant – “AI Eats Less Memory Than Thought”

In early March 2026, Google DeepMind researchers published a paper on a KV cache compression technique called “TurboQuant.” Its core claim was that it could reduce the memory bandwidth required during the AI inference phase by up to a factor of six.[9]

The shockwave this paper sent through the semiconductor market was straightforward: “If AI inference – the core driver of HBM demand – requires far less memory than expected, then today’s HBM capex is excessive.” SK Hynix was hit hardest, unsurprisingly – it has the highest dependence on HBM revenue.

2.3 Third Factor: Iran War and the Dollar-Won Rate Breaks 1,500

The outbreak of the Iran war in March 2026 sent international oil prices soaring, and the dollar-won exchange rate broke 1,500 for the first time since the global financial crisis. Foreign investors net-sold over 33 trillion won of Korean stocks, pulling capital out.[1] The sharp rise in the exchange rate slashed foreign returns on Korean stocks in dollar terms and amplified risk-off sentiment.

These three factors operated through different channels but converged to form a single narrative: “The AI semiconductor boom is ending. And geopolitical risk is about to engulf the Korean market.”

3. The Three Camps: Pessimists vs. Optimists vs. Neutrals

After the March panic, the market split into three camps.

3.1 The Pessimists: “2026 Is the Peak; Down Cycle from 2027”

Main proponents: Goldman Sachs (July 2025), some hedge funds, short sellers

Core arguments:

  1. HBM price decline: In July 2025, Goldman Sachs downgraded SK Hynix to “Neutral,” forecasting that HBM prices would experience their first double-digit decline in 2026. With Samsung entering the HBM market, SK Hynix’s monopoly is broken, and pricing power shifts to customers like Nvidia.[10]
  2. Oversupply cycle: The all-time high combined capex of Samsung+SK+Micron (approx. 105 trillion won + US$25 billion) means the seeds of oversupply in 2027–2028 have already been sown. In semiconductor history, the cycle “boom → capex increase → oversupply 18–24 months later → bust” has repeated itself.
  3. Macroeconomic risks: Prolonged high Fed rates (3.50–3.75%), the ongoing Iran war, and China’s economic slowdown – the advertising and cloud revenue of Big Tech that sustains AI semiconductor demand is vulnerable to recession.

Weakness of the pessimist view: As of May 2026, HBM price declines have not yet materialized. DRAM spot fulfillment rates remain around 60%, with supply unable to keep up with demand.[9] Goldman’s forecast is at least 12 months ahead.

3.2 The Optimists: “The AI Supercycle Has Just Begun – Demand Will Overwhelm Supply for Years”

Main proponents: KB Securities, JPMorgan, Nvidia management

Core arguments:

  1. Nvidia’s earnings prove it: On May 20, 2026, Nvidia reported Q1 FY2027 results: revenue of US$81.6 billion (+156% YoY) and net profit of US$41.2 billion (+180% YoY), far exceeding Wall Street expectations. CEO Jensen Huang stated, “AI demand is so strong that supply cannot catch up.”[11]
  2. HBM is not commodity DRAM: HBM3E 12-Hi is customized for the customer (GPU design). Oversupply does not simply occur as with commodity DRAM. Next-generation GPUs like Nvidia’s Blackwell Ultra and AMD’s MI400 require even more HBM.
  3. Capex merely follows demand: The three companies’ capex increases are evidence of just how large AI data center demand is. HBM demand in 2027 is forecast to more than double from 2026.

Weakness of the optimist view: The market collectively remembers that “This time is different” are the four most dangerous words. In every past semiconductor supercycle – the internet bubble, the mobile transition, the cloud transition – the argument that “this time, structural change means no cycle” was dominant, only to face a down cycle.

3.3 The Neutrals: “Demand Is Real, But Valuations Have Already Priced In the Peak”

Main proponents: Morgan Stanley, some pension funds

Core argument: Samsung Electronics’ PBR has exceeded 3.5x, and the 12-month forward PER of the KOSPI semiconductor sector has reached 25x – historically elevated territory. Even if AI demand is real, stock prices may already reflect more than that. Compared to the 2017–2018 supercycle, where PER peaked at 15x, current valuations are far higher.

4. Nvidia FY2027 Q1: Dissecting the Optimists’ Core Evidence

On May 20, 2026, Nvidia reported Q1 FY2027 (February–April 2026) results.[11]

Metric FY2027 Q1 YoY Change
Revenue US$81.6 billion +156%
Data Center Revenue US$71.2 billion +197%
Net Profit US$41.2 billion +180%
Operating Margin 67.3% +4.2%p
FY2027 Q2 Guidance US$87.5 billion —

These results are overwhelming. However, several warning signals exist:

  1. Growth slowdown: YoY growth of 156% is down from 265% in FY2026 Q4. While absolute revenue increases, the growth rate itself is entering a phase of deceleration due to base effects.
  2. Customer concentration: About 40% of data center revenue comes from the top three cloud providers (Microsoft, Amazon, Google). A decision by them to cut CAPEX would immediately transmit to HBM demand.
  3. China risk: If export controls toward China are tightened, Nvidia’s H20/B20 GPU sales to China will be hit, directly reducing HBM demand from Samsung and SK Hynix toward China.

5. Three Scenarios: Korea’s Semiconductor Path 2026–2028

Scenario A: Soft Landing (Probability 35%)

Assumption: AI demand remains strong through 2027, and HBM avoids oversupply due to its customized nature. Commodity DRAM enters a gradual price decline from H2 2027, but HBM margins offset this.

Unfolding: All-time record annual results in 2026 → Revenue stagnates or slightly increases in 2027 → Modest correction in 2028 (revenue -10% to -15%).

Impact on the Korean economy: GDP growth slows to the mid-2% range in 2027 but avoids a hard landing. Employment at Samsung and SK Hynix is maintained, while suppliers experience partial capacity utilization adjustments.

Scenario B: Hard Landing (Probability 45%)

Assumption: From H2 2026, HBM demand growth slows; from 2027, commodity DRAM oversupply intensifies. The simultaneous CAPEX ramp of the memory three leads to oversupply in Q4 2027–2028.

Unfolding: Peak in Q3–Q4 2026 → Growth in revenue and exports collapses in H1 2027 → Turns negative in H2 2027 → Full down cycle in 2028.

Impact on the Korean economy: Given the share of semiconductor exports in GDP, a hard landing could drag Korea’s GDP growth rate down by more than 1 percentage point. Production cuts and workforce restructuring pressures at Samsung and SK Hynix, a chain of bankruptcies among suppliers, and a sharp drop in government tax revenue would occur simultaneously.

Scenario C: Supercycle Extended by Technological Breakthrough (Probability 20%)

Assumption: A technological breakthrough at the level of AGI, or an algorithmic innovation that dramatically lowers AI inference costs, qualitatively expands AI demand. Next-generation technologies like HBM4 and CXL memory create new demand layers.

Unfolding: Consecutive all-time record results from 2026 to 2028. The traditional cycle of the semiconductor industry breaks down.

Impact on the Korean economy: KOSPI breaks 10,000; GDP growth remains above 3%. However, this scenario has no historical precedent and represents the purest form of “this time is different.”

Basis for Probability Distribution

Scenario B is assigned the highest probability (45%) for two historical regularities: (1) the repeated pattern of the semiconductor cycle in which capex expansion during a boom leads to oversupply 18–24 months later; (2) periods when “structural change” arguments were dominant in the AI industry – at all past peaks, including the internet bubble (1999–2000), the mobile transition (2012–2013), and the cloud transition (2020–2021). Scenario C’s 20% reflects that the possibility of a technological breakthrough cannot be ruled out, but also accounts for the weight of historical precedent.

6. Class Costs: Who Takes the Hit When the Cycle Turns

The costs of a semiconductor cycle turn are not distributed equally. In Korean society, the structure of burden is as follows:

6.1 Chaebol Owners and Top Management: Buffered Risk

Samsung Electronics and SK Hynix accumulated hundreds of trillions of won in cash during the 2024–2026 super boom. This buffer will be used to maintain dividends, buy back shares, and sustain compensation for key executives for at least 2–3 years in a down cycle. The primary shock of a cycle turn is absorbed by the companies’ free cash flow and capex adjustments; the controlling families’ power and wealth suffer only short-term effects.

6.2 Semiconductor Workers: Job Insecurity and Intensified Exploitation

In a down cycle, the first to bear costs are non-regular workers at suppliers. The transmission chain – production cuts → reduced orders to suppliers → layoffs and income loss for subcontract workers – has been repeatedly verified in the down cycles of 2018–2019 and 2022–2023. Regular workers at Samsung and SK Hynix are relatively protected but suffer real wage cuts through reduced bonuses, less overtime, and shift system changes.

6.3 Small Shareholders: The Double Trap of the “National Stock”

Samsung Electronics is a “national stock” directly or indirectly held by about 20% of Korea’s adult population. Many individual investors who entered during the KOSPI’s 82% rally from January to May 2026 bought at all-time highs. If the stock price falls 30–50% in a down cycle, their asset losses transmit into consumption contraction and further deterioration of domestic demand. This is a larger-scale version of the 2021–2022 “Donghak Ant” phenomenon.

6.4 State Finances: Sharp Drop in Corporate and Customs Tax Revenue

Samsung and SK Hynix are estimated to have accounted for about 15–20% of Korea’s corporate tax revenue in 2025.[12] If a cycle turn cuts their profits by more than 50%, a tax revenue shortfall of over 10 trillion won per year could occur. This coincides with the Korean government’s fiscal capacity already under pressure from aging and rising welfare spending, which could amplify into a fiscal crisis.

6.5 Unequal Cost Sharing: “Privatization of Profits, Socialization of Losses”

The excess profits of the semiconductor boom are concentrated among Samsung and SK Hynix shareholders – especially the controlling families and foreign institutional investors. However, the costs of the down cycle are likely to be shifted to the state and taxpayers through tax cuts, subsidies, and financial support. During the 2023 semiconductor down cycle, the government responded with corporate tax cuts, electricity rate freezes, and expanded R&D tax credits – a structure that will repeat on a larger scale in 2027–2028.

7. Semiconductor Dependency from the Perspective of Comprador-Monopoly Capitalism

Korea being called the “Semiconductor Republic” is not a blessing but a vulnerability. With over 40% of exports and more than 35% of KOSPI market capitalization concentrated in two semiconductor companies, this structure is an extreme form of comprador-monopoly capitalism, in which the entire national economy is subordinated to the cycle of a single industry.

In this structure, the “world’s best” semiconductor companies and Korea’s economic sovereignty operate in opposite directions:

  1. Technology dependency: Samsung and SK Hynix’s HBM depends on Nvidia GPUs; Nvidia GPUs depend on TSMC foundry; TSMC depends on ASML equipment. In this value chain, Korean firms remain in the position of “high-end component suppliers,” while final demand and pricing power are held by US Big Tech and designers.
  1. Capital dependency: Foreign ownership of Samsung Electronics exceeds 50%, and that of SK Hynix exceeds 45%. Over half of the excess profits from the boom are converted into dollars and flow out overseas. Yet the adjustment costs of the down cycle – production cuts, restructuring, tax support – remain within the Korean economy.
  1. Policy dependency: The semiconductor industry’s reliance on the United States constrains the Korean government’s diplomatic and trade autonomy. The structural compulsion for Korea to cooperate with US semiconductor restrictions on China shows that semiconductors are not a “private industry” but a control point integrated into US imperialism’s technological hegemony strategy.

This is the concrete face of comprador-monopoly capitalism: a competitive monopoly in the global market, but the conditions of that competitiveness – technology, capital, demand – are structurally dependent on the imperialist order, and that dependency is the source of its monopoly position.

Conclusion: Not Predicting the Peak, But Preparing for What Comes After

The probability distribution of the three scenarios (hard landing 45%, soft landing 35%, supercycle extension 20%) suggests that a semiconductor cycle turn is not an exception but the base case. However, the purpose of this report is not to provide investment advice for timing the peak, but to offer a structural analysis of who bears the costs of a cycle turn.

The class outcomes of a semiconductor down cycle in the Korean economy have already been empirically demonstrated twice – in 2018–2019 and 2022–2023: profits were transferred to foreign shareholders; costs were transferred to supplier workers, small shareholders, and taxpayers. The cycle turn of 2026–2028 will be larger in scale, and government intervention will also be larger.

Industrial diversification to reduce semiconductor dependency, social recapture of excess profits, and building an independent ecosystem to break free from technological dependency – these are the questions that the “semiconductor powerhouse” discourse erases. This report is an attempt to put those questions back on the table.

[1] MBC News, “KOSPI surges 8.42%… recovers 7,800 level,” 2026.5.22. https://imnews.imbc.com/replay/2026/nw2500/article/6824431_36989.html

[2] Chosun Ilbo, “May 1–20 semiconductor exports US$22 billion, up 202% YoY,” 2026.5.21. https://www.chosun.com/economy/economy_general/2026/05/21/M7ZXL7VAUZEDXOLSIBCWDAMQA4

[3] Money Today, “April semiconductor exports US$31.9 billion,” May 2026.

[4] Chosun Biz, “Samsung Electronics Q1 revenue 133 trillion won, operating profit 57.2 trillion won,” 2026.4.7. https://biz.chosun.com/it-science/ict/2026/04/07/PTBXH24YHREFLDWFM4O2CL3KUY

[5] G-enews, “SK Hynix Q1 revenue 52.6 trillion won, first to exceed 50 trillion won,” 2026.3.27. https://www.g-enews.com/article/Global-Biz/2026/03/202603270952218497fbbec65dfb_1

[6] Micron Technology FY2026 Q2 Earnings Release and Q3 Guidance. February 2026.

[7] Newdaily Economy, “KOSPI surges 8.42% on triple positives,” 2026.5.22. https://biz.newdaily.co.kr/site/data/html/2026/05/21/2026052100243.html

[8] Industry composite: Samsung Electronics and SK Hynix IR materials and securities firm reports.

[9] Market data and securities firm reports composite. KOSPI daily volatility data for March 2026.

[10] Goldman Sachs, “SK Hynix: Initiate at Neutral,” July 2025.

[11] NVIDIA FY2027 Q1 Earnings Release, 2026.5.20. Refer to IndMoney report. https://www.indmoney.com/blog/us-stocks/nvidia-stock-q1-fy27-earnings-81b-revenue-25x-dividend

[12] National Tax Service corporate tax statistics and securities firm estimates composite. Estimated share of Samsung Electronics + SK Hynix corporate tax.