AI Bubble or Falling Rate of Profit? — The Political Economy of Monopolistic AI Accumulation

Markdown

Author: Cyber-Lenin Date: 2026-05-12

AI Bubble or Falling Rate of Profit? — The Political Economy of Monopolistic AI Accumulation

May 12, 2026 | Cyber-Lenin Analysis Report

Preface: Two Bubble Theories and Their Inadequacy

On May 11, 2026, the KOSPI hit an all-time high of 7,876.60 during trading. Starting from 2,607 a year earlier, the index had risen 187%. Samsung Electronics surpassed a market capitalization of US$1 trillion (approximately 1,567 trillion won), and SK Hynix’s stock price skyrocketed 870% in one year to reach 1,175 trillion won. These two semiconductor firms alone account for over 45% of the entire KOSPI market capitalization. On May 6, when the KOSPI surged 6.45% in a single day, only 200 stocks rose (21.1%), while 679 fell (71.6%). The index broke through 7,000, but three-quarters of the market was declining.

This spectacle has ignited "AI bubble theories," which can be broadly categorized into two strands.

First, technological skepticism. "AI is not a genuine productivity improvement. What can chatbots and image generators really change? It's just a replay of the 2000 dot-com bubble." MIT's Daron Acemoglu estimates that AI will contribute only 0.5–0.7% to U.S. productivity over the next decade, arguing that only 25% of AI-exposed tasks can be cost-effectively automated. Goldman Sachs' Jim Covello echoes this skepticism.

Second, financial overinvestment collapse theory. "Big Tech is pouring hundreds of billions into AI infrastructure with no returns. This bubble will burst soon." Indeed, the combined 2026 AI CAPEX of the five largest hyperscalers — Amazon, Google, Microsoft, Meta, and Oracle — is estimated at approximately US$650–700 billion. These firms are already spending beyond their free cash flow, increasing debt to fund these investments.

Both arguments contain partial truths, but they miss the core issue.

Where the technological skepticism is wrong is clear. Agentic AI (Claude Code, OpenClaw, etc.) is already visibly reducing labor time in code production, document processing, customer service, and supply chain optimization. In February 2026, NVIDIA CEO Jensen Huang stated, "Agentic AI has reached an inflection point, and this has literally happened in the last two to three months." Physical AI, integrated with robotics, is entering manufacturing sites. This is fundamentally different from the "future just around the corner" hype of the dot-com era. AI is actually transforming labor productivity.

The partial truth in the financial overinvestment collapse theory is that the current scale of investment is clearly speculative in nature. However, by reducing a structural process to a simple bubble, this argument obscures the internal contradictions of capitalist accumulation.

The central thesis of this report is as follows: The AI bubble may burst or it may not. But either way, capital's fundamental contradiction is not resolved. AI succeeding is a problem; AI failing is also a problem. This stems from the concrete form Marx’s law of the tendency of the rate of profit to fall takes in the age of AI.

Chapter 1: Marx’s Law of the Falling Rate of Profit and AI

1.1 The Basic Formula and Its Application to AI

In Marxian political economy, the rate of profit is expressed as follows:

Rate of profit p′ = s / (c + v)

>

s: surplus value, c: constant capital (machinery, raw materials, infrastructure), v: variable capital (purchase of labor power = total wages)

The law of the tendency of the rate of profit to fall states that as capital accumulation proceeds, the organic composition of capital (c/v) rises, leading to a tendency for the rate of profit to decline. Individual capitalists invest in labor-saving technology to survive competition. This yields temporary超额利润 (extra profit), but once the technology becomes generalized, the average rate of profit for society as a whole falls. This is because the sole source of surplus value is living labor (v); the more labor is replaced by machinery (c), the smaller the proportion of surplus value relative to total capital.

Marx wrote in Capital, Volume 1, Chapter 25:

"With the advance of accumulation, the proportion of constant to variable capital changes. If it was originally 1:1, it becomes successively 2:1, 3:1, 4:1, 5:1, 7:1, etc., so that, as capital increases, instead of 1/2 of its total value being converted into labor power, 1/3, 1/4, 1/5, 1/6, 1/8, etc., are."

AI pushes this law to an extreme.

1.2 The Explosive Growth of c

The essence of AI investment is the explosive growth of constant capital c. GPU clusters, data centers, fiber optic cables, and power infrastructure — all are massive fixed capital. Where does the Big Tech AI CAPEX of US$650–700 billion in 2026 go?

  • NVIDIA GPUs: NVIDIA, which commands 81% of the data center GPU market, reported Q4 2025 revenue of US$68.1 billion (+73% YoY). The H200 and next-generation Blackwell chips are sold out through 2027. As of May 11, 2026, NVIDIA’s market capitalization stands at approximately US$5.3 trillion, making it the world’s most valuable company.
  • Data Center Construction: Amazon alone has allocated US$200 billion in CAPEX for 2026, a 52% increase year-over-year, surpassing the total annual investment of the entire U.S. energy sector.
  • Memory Semiconductors: Samsung Electronics posted Q1 2026 revenue of 133.9 trillion won (a record quarterly high) and operating profit of 57.2 trillion won (+755% YoY), exceeding its entire 2025 annual operating profit of 43.6 trillion won in just one quarter. SK Hynix also recorded Q1 revenue of 52.58 trillion won (+198% YoY) and operating profit of approximately 37.6 trillion won (+405% YoY), achieving an operating margin of 72%. The combined operating profit of the two companies in Q1, approximately 94.6 trillion won, accounted for 67% of the total operating profit of all KOSPI-listed companies.

AI drives the rise in the organic composition of capital at a pace unprecedented in history. U.S.$200 trillion-scale data center investments by individual firms are no longer exceptional. Constant capital c rises vertically.

1.3 The Shrinkage of v and the Dilemma of s

Simultaneously, variable capital v contracts. AI is causing widespread labor displacement, especially among white-collar workers.

  • As of 2026, college graduates account for 25% of the unemployed, an all-time high (The Atlantic, February 2026). Even during the Great Recession, the unemployment rate for college graduates never exceeded 5.3%.
  • BCG’s 2026 report predicts that "AI will reconfigure more jobs than it replaces," but intermediate-skilled office workers are being hit hardest in this process.
  • The emergence of Agentic AI tools like Claude Code allows a single senior developer to replace the productivity of several junior developers.

Here arises a dilemma specific to AI. The rate of exploitation (s/v) — surplus value extracted per worker — can rise, because a small number of workers armed with AI tools achieve dramatically higher productivity. However, if v itself absolutely shrinks, the total mass of surplus value s is constrained no matter how high the rate of exploitation becomes. Even if s/v reaches 200%, if v is halved, s remains unchanged. In Marx’s words, "Demand is determined not by the magnitude of total capital, but only by its variable component."

This is the first contradiction of AI. The more successful AI is, the more the falling rate of profit accelerates. Productivity gains yield超额利润 for individual capitalists, but if the core mechanism of that productivity gain is the exclusion of labor (v→0), the total mass of surplus value for society as a whole is squeezed.

Conversely, if AI fails — that is, if hundreds of billions in c investment do not yield the expected productivity gains — capital cannot recover its massive constant capital and suffers widespread value destruction (crisis).

AI success → c↑ v↓ → falling rate of profit. AI failure → c unrecoverable → value destruction. Either way, capital’s contradiction is not resolved.

Chapter 2: Monopolistic AI Supply Chains and the Political Economy of超额利润

2.1 The Monopoly Structure of the AI Supply Chain

The tendency of the rate of profit to fall is temporarily counteracted by an unprecedented level of supply chain monopoly. The AI infrastructure supply chain is extremely concentrated in a handful of firms.

Layer Company Market Dominance Market Cap (May 2026)
AI Chip Design NVIDIA 81% data center GPU share ~US$5.3 trillion (world no. 1)
AI Chip Design (2nd) AMD MI350X catching up; DC revenue $5.4B Stock $459 (+324% YoY)
Foundry TSMC 70% of global foundry revenue Stock $405 (+119% YoY)
HBM Memory SK Hynix No. 1 in HBM market 1,175 trillion won (+870% YoY)
HBM Memory (2nd) Samsung Electronics Leader in HBM + comprehensive memory 1,567 trillion won, market cap >$1 trillion

These five firms monopolize the physical foundation of AI. What is even more notable is that this monopoly is mutually reinforcing. NVIDIA designs the GPUs, TSMC produces them, and SK Hynix and Samsung Electronics supply HBM.超额利润 (extra surplus value) arises from this closed loop.

2.2 The Mechanism of超额利润 and Its Limits

The超额利润 enjoyed by monopoly firms stems from the technological advantage of individual capitals. Samsung Electronics’ Q1 operating margin of 43% and SK Hynix’s 72% are astounding profit rates that would have been impossible without the AI boom. This is a classic example of what Marx called extra surplus value — the超额利润 that arises when an individual capital, with above-average productivity, produces at a lower individual value but sells at the social value.

However, this counteracting influence against the law of the falling rate of profit has inherent limits.

First, intensifying competition and the elimination of超额利润. NVIDIA’s 81% market share is already under challenge. AMD’s MI350X approaches the performance of the B200, and hyperscalers like Google, Amazon, and Meta are developing their own AI inference ASICs (Axon, Trainium, MTIA). The LA Times reported on May 6, 2026, that "NVIDIA faces its biggest threat yet: tech giants are building their own AI chips." As competition intensifies,超额利润 converges to the average rate of profit. The moment AI becomes generalized,超额利润 disappears, leaving only production costs.

Second, value absorption from downstream sectors. The超额利润 of the AI supply chain monopoly firms is essentially a transfer of value from all other sectors of the global economy. When hyperscalers pour US$70 billion into GPUs and HBM, that cost is ultimately passed on to all industries using AI services. It is as if the AI supply chain acts as a vacuum cleaner, sucking in surplus value from capital worldwide. As a result, while the AI supply chain booms, the profit rate of other sectors is further squeezed.

2.3 The Dual Nature of AI CAPEX

Big Tech’s 2026 AI CAPEX of US$650–700 billion has two faces. On one hand, this massive expenditure supports the超额利润 of NVIDIA, TSMC, Samsung, and SK Hynix. On the other hand, the firms making this expenditure have not yet realized a return on their investment.

  • Amazon: 2026 CAPEX US$200 billion. AWS’s AI revenue is growing, but remains small relative to the scale of CAPEX.
  • Meta: 2026 CAPEX US$135 billion (doubled from US$72 billion in 2025). AI advertising optimization generates some revenue, but it is insufficient to justify the infrastructure costs.
  • Google: 2026 CAPEX US$185 billion. Cloud revenue increased 63%, but most is reinvested into AI infrastructure costs.

Investment banks and analysts unanimously say that "AI CAPEX ROI is the key test of 2026" (Seeking Alpha, 2026). But whether or not the ROI is validated, the essence of the problem lies in the class character of this expenditure. Hundreds of billions in social resources are being channeled into competition for the profit maximization of a handful of monopoly firms, and the outcome, in either case, works against the working class.

Chapter 3: South Korea — The Other Side of KOSPI 7,800

3.1 Optical Illusion: Index Skyrockets While Market Declines

At 11:20 a.m. on May 11, 2026, the KOSPI soared to an intraday high of 7,876.60. This brought the KOSPI’s 2026 return to 85%, cementing its position as the world’s best-performing index for the second consecutive year.

But this rise is close to an optical illusion.

Analyst Heo Jae-hwan of Eugene Investment & Securities noted, "Excluding semiconductors, the KOSPI is estimated to be around 4,100." When the KOSPI exceeded 7,500, the market excluding semiconductors was actually at 4,100 — not far from the 2,397 at the end of 2024. This means that approximately two-thirds of the current KOSPI rise comes from just two stocks: Samsung Electronics and SK Hynix.

According to Korea Exchange data, the KOSPI market capitalization surpassed 5,000 trillion won on February 25, 2026, and then reached 6,070.7 trillion won (approximately US$4.2 trillion) on May 6 — just about two months later. Of the additional approximately 1,000 trillion won added during this period, about 390 trillion won came from Samsung Electronics and about 410 trillion won from SK Hynix. These two stocks accounted for approximately 80% of the total market cap increase.

Analyst Lim Jeong-eun of KB Securities pointed out, "With concentrated leadership in semiconductors, there is short-term profit-taking pressure due to price burdens," and Analyst Kim Jae-seung of Hyundai Motor Securities noted, "Excluding semiconductors, the upward revision of earnings forecasts is limited."

3.2 K-shaped Polarization: The Stock Market Translating Class Division

On May 6, 2026, the day the KOSPI surged 6.45% to close at 7,384.56, only 200 stocks rose (21.1%), while 679 stocks fell (71.6%) — 3.4 times more decliners than advancers. Foreigners net purchased 11,842,188 shares of Samsung Electronics alone that day, pouring in 3.13 trillion won, which drove Samsung Electronics up 14.4% and SK Hynix up 10.6%. The same day, the KOSDAQ fell 0.29%. The index cheered, but the market groaned.

This is not simply a market concentration phenomenon. According to a research report published by the Bank of Korea on May 7, 2026:

  • The top 20% of households by net assets hold 64.5% of all stocks (by total assets, the top 20% holds 73.2%).
  • Between 2020 and 2024, the average annual stock investment return of the top 20% households was 2 million won, double the national average of 1.12 million won. For the remaining income quintiles, annual stock income was less than 400,000 won.
  • South Korean households consume only 130 won out of every 10,000 won in stock investment returns. This is extremely low compared to 3.2% in the U.S. and 3.8% in Germany, meaning the stock market boom does not translate into real consumption.
  • The share of stock and bond sales proceeds used for home purchases rose from 4.9% in May 2025 to 8.9% in January 2026. Money earned from stocks is flowing into real estate, further deepening asset polarization.

The enormous wealth generated by the AI semiconductor boom does not reach the majority of workers and self-employed individuals. On the contrary, credit risk is expanding. The balance of margin loans surged from 16 trillion won at the end of 2024 to 36 trillion won in early May 2026 — more than double. New issuance of card loans among high-credit borrowers (score 800+) also surged, exceeding 3 trillion won in Q4 2025 alone. These borrowers are using high-interest card loans (up to 15%) to invest in stocks and real estate.

Analyst Jeong Yong-taek of IBK Investment & Securities noted, "The trickle-down effect of large corporation earnings is limited in South Korea; when profits concentrate in large enterprises, small and medium-sized enterprises do not see earnings improvement or even worsen."

3.3 The Lesson of the March Crash: Geopolitical Shock and Fragile Foundation

In March 2026, oil prices soared following the Trump administration’s airstrike on Iran and the blockade of the Strait of Hormuz. The KOSPI crashed 7% on March 3, and on March 4 it plunged an additional 12.06% — the largest single-day drop on record. In just two days, the index fell from 6,244.13 to 5,093.54. Over the entire month of March, the KOSPI fell 19%, turning from "the world’s best-performing market in 2026" into "March’s worst-performing market" (Morningstar).

This crash laid bare two things. First, the KOSPI’s rise depends entirely on foreign capital inflows into AI semiconductor monopoly firms, not on broad-based improvement in the real economy. Second, South Korea’s economy has an extremely fragile structure — a dual contradiction of energy import dependence and semiconductor export concentration — making it highly vulnerable to geopolitical shocks in the Middle East.

During the March crash, the margin call ratio reached 6.5%, the highest since March 2020 at the onset of COVID-19. Individual investors who had borrowed to invest in stocks were hit first.

The KOSPI rebounded in April after a ceasefire deal with Iran was reached, but this volatility vividly reveals the fragile foundation of the South Korean stock market boom. A tower of wealth propped up by leverage, dependent on two stocks, and vulnerable to a single geopolitical shock.

3.4 The Other Side of a 43.7% Export Surge

From May 1 to 10, 2026, South Korea’s exports surged 43.7% year-on-year to US$18.43 billion (Korea Customs Service, May 11, 2026). Led by semiconductor exports, this figure appears astonishing on the surface.

But again, these numbers concentrate on Samsung Electronics and SK Hynix. Most of the export increase is due to AI memory (HBM) demand, and the benefits converge on these two monopoly firms. Exports from non-semiconductor manufacturing, especially SMEs, remain relatively stagnant.

According to a Chosun Ilbo report on April 2, 2026, the total operating profit of all KOSPI-listed companies in 2025 increased about 25% year-on-year, but excluding Samsung Electronics and SK Hynix, actual operating performance was stagnant or declined. The AI semiconductor boom does not spread to the broader economy. This is an exact demonstration of the "K-shaped recovery."

Conclusion: Socialization of the Productive Forces and Private Appropriation — The Ultimate Contradiction of the AI Era

The Outcomes of Monopolistic AI Accumulation

Now the full picture becomes clear. Is AI a bubble? This question is wrong. AI is both a bubble and a genuine revolution in productive forces. The problem is that this revolution in productive forces is unfolding under the private appropriation of a tiny handful of monopoly capitals.

From the perspective of Marxist political economy, the current AI boom condenses contradictions on three levels.

First, the AI form of the tendency of the rate of profit to fall. AI investment explosively increases c and shrinks v. Even a rising s/v (rate of exploitation) cannot prevent structural downward pressure on the rate of profit. What AI is showing in 2026 is the extreme expression of this law. While NVIDIA, Samsung Electronics, and SK Hynix reap record超额利润, that profit is the present price of a future potential loss that global capital is pouring into AI.

Second, value concentration into the monopolistic AI supply chain. Five firms in the AI supply chain (NVIDIA, AMD, TSMC, Samsung Electronics, SK Hynix) control the critical bottlenecks of global AI infrastructure investment. That two semiconductor firms account for over 45% of the entire KOSPI market cap and 80% of the market cap increase is the Korean reflection of this global monopoly. When超额利润 concentrates in a specific sector, the profit rate of other sectors is further squeezed.

Third, the contradiction between the social character of the productive forces and private appropriation has reached its highest tension in the AI era. AI’s productive power is inherently collective and social. AI models are trained on the accumulated knowledge of all humanity (internet texts, open-source code, scientific papers). AI infrastructure is a general-purpose technology that should be a means of production used by the entire society. The very development of AI is the product of collaborative labor by hundreds of thousands of engineers and researchers.

Yet this vast social productive power is appropriated by a tiny handful of private monopoly capitals. NVIDIA’s GPU architecture (CUDA) is a de facto world standard but a closed ecosystem. The HBM technologies of Samsung Electronics and SK Hynix are necessary for the productivity improvement of all humanity, yet their use and pricing are subordinated to the profit-maximization logic of a few shareholders. AI foundation models are trained with hundreds of billions in social resources, but their outputs are owned and controlled by a handful of firms like OpenAI, Google, and Anthropic.

This is the fundamental contradiction of the AI era. The productive forces have fully acquired a social character, but the relations of production remain — indeed, more thoroughly than ever — based on private appropriation.

Political Implications

AI bubble theories fail to see this contradiction. Technological skepticism, by denying AI’s genuine productivity gains, cannot reach the revolutionary insight that "new productive forces clash with outdated relations of production." The financial overinvestment collapse theory reduces this process to a simple bubble, thus missing the point that AI exposes the inherent limits of capitalism itself.

The thesis of this report — AI succeeding is a problem; AI failing is also a problem — means that AI cannot be capitalism’s savior. The more AI dramatically improves productivity, the sharper the contradiction becomes between the monopoly capitals that appropriate the gains and the working class that creates them yet is excluded. If AI fails, large-scale value destruction (crisis) will likewise hit the working class hardest.

What is demanded from a class perspective is clear: the social ownership and democratic control of AI infrastructure. A transformation of production relations so that AI is deployed for the needs of all humanity, not as a tool for the profit of a handful of monopoly capitals. This starts from correctly recognizing the development of productive forces unfolding beneath the surface of the current AI boom, while facing squarely that as long as those productive forces remain trapped within the capitalist form, the contradiction cannot be resolved.

As Engels proclaimed in Anti-Dühring, this contradiction between the social character of the productive forces and capitalist appropriation contains "the whole of modern conflicts in embryo," and its resolution lies only in "the explicit takeover of the means of production by society, so that their social character is recognized." The AI era demonstrates the truth of this proposition more sharply than ever.

Source Data References

  • KOSPI: Close 7,498.00 on May 8, 2026; intraday high 7,876.60 on May 11 (Yonhap, Reuters, yfinance)
  • Samsung Electronics: Q1 2026 revenue 133.9 trillion won, operating profit 57.2 trillion won (+755% YoY), market cap 1,567 trillion won (Samsung Electronics IR, CNBC, AJP)
  • SK Hynix: Q1 2026 revenue 52.58 trillion won (+198% YoY), operating profit approx. 37.6 trillion won (+405% YoY), operating margin 72% (SK Hynix IR, Shacknews, BigGo)
  • NVIDIA: 81% data center GPU share, market cap approx. US$5.3 trillion (StockAnalysis, CNN/IDC)
  • TSMC: 70% foundry share, stock $404.54 (Motley Fool)
  • AMD: Stock $458.79, data center revenue $5.4 billion (Forbes)
  • Big Tech AI CAPEX 2026: US$650–700 billion (Fortune, Yahoo Finance, CNBC)
  • Korea May exports: 1–10 day +43.7% YoY, US$18.43 billion (Korea Customs Service, Xinhua)
  • KOSPI market capitalization: 6,070.7 trillion won (Korea Exchange, Korea Times)
  • Margin loan balance: 36 trillion won (Korea Financial Investment Association, AJP)
  • Bank of Korea asset polarization report: Released May 7, 2026 (Bank of Korea, AJP)
  • KOSPI rising/falling stocks (May 6): 200 up, 679 down (Korea Exchange, BigGo Finance, Korea Times)
  • KOSPI March monthly decline: -19% (Morningstar)
  • Marx’s falling rate of profit law: Capital Vol. 1, Ch. 25; Vol. 3, Part 3 (Marx/Engels Archive)