Good evening everyone.

A study by the The Bank of Korea issued just a few month ago stated: If adoption is broad and swift, artificial intelligence could boost Korea’s total productivity by up to 3% and push GDP as much as 13% higher, enough—so the model says—to cancel two-thirds of the growth lost to our rapidly ageing population. It is a striking forecast and this could be one reason presidents, CEOs and capitalists now speak of GPUs almost in the same way 19th-century tycoons spoke of railroads or steel.

Yet figures alone do not tell us who captures the gains, who bears the costs, or why AI has become a fresh terrain of rivalry between corporations and between states. Let’s shift the spotlight from clever algorithms to the social relations that organise them—from the silicon etched in foundries like Samsung’s Hwaseong campus, to the armies of crowd-workers who label data in Manila. Can we read this frenzy through a Marxist Lens and show that AI, as an advanced technology, fits into a common pattern. I will show how AI only intensifies capital’s hunt for profit, and how it reshapes the global division of labour.

AI Introduction

I’d like to start out by giving a brief introduction to machine learning. In conventional software the programmer dictates every step in advance: if A happens, then execute B. Machine learning flips that script. We show the computer million—or billions—of examples of A,B pairs and task the computer to discover the pattern on its own.

A few years ago computer scientists were attempting to see if a computer could learn these patterns in language, so they gave the computer many sentences as ‘input’ and asked to computer figure out the correct output which was just the next word of the sentence. To their surprise after the model was trained the computer could be given the input “What is the capital of France?” And the computer by just trying to keep the string of text going, gave the correct answer “Paris”.

This astounding feat was made possible by several factors:

• a new type of machine learning model known as transformers became available,
• fast GPUs that could do the heavy number crunching required to train the model weights
• a colossal amount of training data (the inputs and outputs) scraped from books, wikis and social media, often without paying the original authors.

The process of training a machine learning model is a guided trail and error approach that minimises the errors inside the computer. At each trial the millions of parameters (or weights) of the model are adjusted until they give minimal error on the predicted output.

Once the model is trained, the weights of the model are frozen. Thus the terabits of input data are condensed into several a billion numbers. But when we want to run the model (known as inference), like when we ask a question of chatGPT, our query is processed through the model - interacting with these billion numbers via a GPU. So its hardware intensive both during training but also during inference.

So why does AI matter now—why the headlines, the policy battles, the first campaign pledge of Korea’s leading presidential candidate being a bulk GPU purchase? To answer, we need more than technical detail. We need to revisit Marx’s distinction between living and dead labour, constant and variable capital, and see how those categories illuminate today’s AI gold rush.

Marxist Economics

Capitalism rests on the production of surplus value—the portion of new value that workers create but do not receive as wages. In factories, living labour combines with raw materials and machinery to generate commodities. The amount spent on wages, Marx calls variable capital (V); the outlay on buildings, equipment and inputs is constant capital (C), which contributes no fresh value of its own but gradually reappears, diminished, in each finished product as it wears out. A machine therefore embodies dead labour: the congealed work of past producers that now ages and depreciates while transferring its fixed value into new commodities. The organic composition of capital—the ratio C/V—indicates how heavily a firm relies on machinery and materials relative to living labour; under competitive pressure this ratio tends to rise, with far-reaching effects on profitability.

Let us now return to the different components of AI in light of these Marxist definitions. The model weights—or encapsulated digital knowledge—can be understood as a form of dead labour. They represent a distillation of countless acts of human work: from the authors, poets, and artists creation, to the labour involved in designing and optimizing the models themselves. Yet unlike physical machinery, this digital dead labour does not degrade through use. It is endlessly reproducible, a kind of zombie labour that continues to function long after the original human input has been forgotten or obscured.

However, the crucial point is that these weights cannot do anything by themselves. To utilize them, one requires a specialized material apparatus: the GPU. The GPU more clearly embodies Marx’s classical notion of dead labour—it is a complex machine created through past human labour, and as it operates, it gradually transfers its value to the commodities it helps produce, through wear, energy use, and depreciation.

New Technology, Old Problems

Each time capital introduces a more productive machine, the pioneer enjoys a brief windfall. Imagine two clothing firms. One still relies on a loom that turns out 10 coats per hour with a single worker; the rival company installs a new loom that delivers 20 coats per hour with the same wage bill (V). Because prices in capitalism gravitate toward the socially necessary labour-time, the early adopter can undersell competitors and still pocket a larger surplus value per coat — at least until everyone else catches up.

But once the new loom becomes general, the market value of a coat is now pegged to 20-per-hour productivity; the extra profit vanishes, while the industry’s average organic composition of capital (C/V) has doubled. Each firm has tied up more money in machines and less in living labour, the sole source of new value. The result is Marx’s tendency of the rate of profit to fall (TRPF):

Rate of profit = Surplus value (S) / (C + V)

If C rises faster than the surplus value S wrung from V, the denominator grows quicker than the numerator and the profit rate declines, even when the mass of profit can still expand. In plain terms, capitalists must run harder—invest ever larger sums in equipment like looms or GPUs—just to stand still in the market.

AI is being adopted by various businesses and for various purposes. So while the exact change to the organic composition of capital will be very inhomogeneous we may still expect that the formulation for the rate of profit to fall remains a strong tendency.

Exploitation

But Marx reminds us that the falling rate of profit is not a mechanical doom-loop; capital constantly looks for “counteracting influences.” The most immediate of these is to squeeze more surplus value out of the same workforce.

If capitalists can drive the rate of exploitation (S/V) upward fast enough, they can keep the overall profit rate afloat even while the organic composition (C/V) climbs. They do this by lengthening the working day, intensifying labour during the same hours, or cutting the value of labour-power itself.

Consider Amazon’s warehouses, where workers now strap on wrist devices that log every motion to the second; any lull counts as “time off task” and can trigger discipline. Here, digital surveillance turns each minute into a battleground over surplus value.

In Korea the state is also willing to stretch the working day. In 2023 the Yoon government proposed a 69-hour work week—which eventually did not pass. Failing at the national level, capital looked sector by sector and in April this year the Ministry of Employment and Labour quietly granted Samsung Semiconductor a permit for 64-hour weeks, renewable every six months. Management justified the regime as unavoidable in the global AI-chip race while the state nodded along.

So we see that advanced tech does not equate to lessened workload on a worker, in fact it can be the exact opposite!

Monopolies and State Capitalism

As Marx observed, profits can be rescued not only by grinding labour harder but by reshaping the market itself. When a few giants corner an industry, they can sell above the value dictated by average social-labour time. And when the state shoulders huge fixed costs or guarantees demand, it further insulates those giants from competitive pressure. Both moves lift the numerator in the profit-rate fraction while preventing the denominator (C+V) from ballooning quite so fast.

Presidential hopefull Lee JaeMyung has already pledged to secure up to 50,000 high-end GPUs for a national AI project and during the recent presidential debates sketched his vision for a Korean AI. He proposed a data centre to be the hub of the next generation of large language models based in Henam and powered by the renewable wind power of the South-East coast.

In the United States under the CHIPS and Science Act the Biden administration has awarded up to $8 billion in direct grants and cheap loans to Intel for its Ohio and Arizona factories, plus a 25% investment tax credit.

Monopoly pricing and state subsidy allow capital to raise surplus value without adding a single worker or overtime hour. By off-loading investment risk onto taxpayers and throttling competition, they keep the average profit rate higher than it would otherwise be. Yet the fix is temporary: protected giants still confront saturated markets abroad, political scrutiny at home, and the sheer scale of new investment required for each technology leap. When those pressures intensify, monopoly alliances spill over national borders.

Imperialism

As Chris Harman argued in Explaining the Crisis, imperialism is not a policy choice but a structural necessity: advanced capitalist states must export capital, secure cheap inputs, and dominate markets to offset falling profit rates at home. AI, with its voracious demand for data, rare minerals, and geopolitical leverage, has become a new frontier for capitalist accumulation, sharpening tensions between imperial powers and deepening oppression in subordinate regions.

The recent $1 trillion loss in U.S. tech stocks following the release of DeepSeek caused a panic in Silicon Valley. Tech executives rushed to Washington to beg for further government protections and actions against China.

Harman emphasizes that “No capitalist wants to face alone a world of bitter, unregulated competition… The state… sees its job as being to carry out lifeboat operations designed to forestall such collapse.” So even as corporations celebrate the free market in theory, they demand active state support in practice.

Nothing illustrates the fusion of state and monopoly quite like Washington’s GPU embargo. Successive rounds of U.S. export controls have blocked Nvidia’s A100, H100 and now even the “China-compliant” H20 chips from reaching the mainland, explicitly to slow China’s model training capacity. A commodity that once circulated freely is now treated as a strategic asset.

South Korea walks a tightrope as a “middle” technological power. Seoul is courted by Washington yet Samsung and SK Hynix operate plants and sell a significant fraction of semiconductors to China. That being said, there was a drastic decline of 31% in sales since February of this year in response to stricter US trade restrictions.

Domestic firms were also “invited” to join the $500 billion U.S. Stargate project, even as Samsung and SK Hynix negotiate carve-outs that would still let them upgrade plants inside China.

Yet Imperialism is not confined to great-power rivalry; it also manifests in the subjugation of weaker states and peoples, often enabled by AI itself. The U.S., for instance, has deepened its support for Israel’s occupation and genocide in Palestine, supplying AI-powered surveillance systems and autonomous drones deployed in Gaza.

China’s imperialism, meanwhile, operates through internal colonization and external expansion. Within its borders, AI fuels a surveillance state, most starkly in Xinjiang, where Uyghur Muslims face internment and algorithmic policing. While, externally, China’s Belt and Road Initiative may looks progressive to some but it is ultimately aimed at subjugating weaker nations to secure raw materials required for advanced tech.

From Chips to Class Struggle

All the above points to a system in profound crisis. But we should remember that crises are also moments of struggle, and the response of the working class will determine the future AI creates.

After the company’s first-ever strike in 2024, the 30,000-member National Samsung Electronics Union forced a wage-deal breakthrough, shattering the firm’s forty-year “no-union” rule. It was also the first strike over pay transparency concerning a controversial incentive metric and the first major strike in the global chip industry, where rapid expansion has been coming at the expense of workplace safety.

Google staff in the US staged sit-ins and walkouts against Project Nimbus, the $1.2 billion AI-cloud contract with Israel; management has fired at least 28 organisers but the campaign widened to DeepMind’s UK researchers, now seeking formal union recognition!

Last month, Microsoft employees disrupted high-level executives speaking at an event celebrating the company’s 50th anniversary, in protest against the company’s role in Israel’s ongoing Genocide. And an Microsoft event in March was preceded by a rally outside that included current and former employees of the tech giant. Protesters projected a sign on to the wall saying: “Microsoft powers genocide”.

In South Korea, workers are already fighting back against AI-driven exploitation. Baemin delivery drivers, whose routes and wages are dictated by AI algorithms, have organized against increasing workloads and declining pay under gig economy platforms. The riders launched a collective action to refuse deliveries over one weekend last month. More than 100 posts, including screenshots titled “Baemin App OFF,” were uploaded to the Rider Union café, a labor union for delivery app workers.

These battles underscore the contradiction at the heart of capitalism: the same technologies states wield to police borders and battle rivals are provoking new solidarities among the workers who design, debug and deliver them. If the first half of the story is a rivalry between blocks of capital, the second is the re-emergence of labour as a conscious antagonist—proof that even in the age of neural networks, class struggle remains the motor of history.

Artificial Intelligence or Anti-Imperialism

In this article I have tried to show that AI as an advanced technology fits into a common pattern. What begins as a technological wonder quickly turns to human exploitation, which is state supported and that ultimately and by logical necessity has to contribute to systemic instability and eventually contribute to Imperialism.

Realising this, means realising that all the well meaning legislation, regulation and safeguards are not enough under the current capitalist system. They are not enough to halt the lurch from one crisis to another.

The crisis of AI under capitalism is undeniable. Instead of being a tool for human liberation, it has the potential to deepen exploitation, increase unemployment, and induce economic instability. But this trajectory is not inevitable. AI is not an autonomous force—it is shaped by the social and economic system in which it develops. The question is simple: who controls AI, and for whose benefit? Under capitalism, AI is a tool for maximizing profit, deepening imperialist rivalries, and expanding surveillance. But under socialism, under workers control, it could be repurposed to reduce working hours, eliminate scarcity, and reorganize production to meet human needs.

This is why the response to AI must go beyond regulation, ethical oversight, or social welfare policies. The capitalist class will not voluntarily create a system where AI benefits workers. A world where AI is used to genuinely free humanity from unnecessary labor will not emerge from government policy alone—it must be won through class struggle.