South Korea has announced its participation in the UAE’s Stargate Project, a vast new international AI data-centre initiative positioned as a cornerstone of the next phase of global technological competition. President Lee Jae-myung has framed this move as part of a national drive to turn South Korea into a leading global AI power and a regional hub for next-generation computing. At the same time, even the most prominent AI companies in the world remain deeply unprofitable, burning through enormous amounts of capital while promising imminent technological transformation.

Taken together, these developments reveal more than speculative excess. They expose a deeper tension at the heart of capitalism itself: the drive to eliminate human labour through automation while still depending on that labour as the ultimate source of profit. As investment in Artificial General Intelligence (AGI) accelerates, this contradiction shifts from abstract theory into an immediate economic and political threat.

The question haunting workers worldwide is stark: Will AGI and advanced robots finally achieve capital’s dream of production without workers? Can machines that build machines eliminate the need for human labor entirely? The answer requires us to move beyond technical speculation to examine the social relations that govern how technology is deployed under capitalism.

The Technology

But first let us acquaint ourselves with the some technical details.

Artificial General Intelligence, or “AGI,” is often defined as an AI system that can perform any intellectual task a human can do (or at least the vast majority of them), and can flexibly learn new tasks without being rebuilt from scratch.

Our current best systems, like ChatGPT-style large language models or advanced image models, are not AGI. They can be extremely impressive in specific contexts, but they lack several key features that human intelligence takes for granted.

Long-term memory: Humans can build up a rich, structured memory over years. We remember people, places, projects, and can draw on this history flexibly. Current front-end AI systems only see a limited “context window” at a time. They can be hooked up to external tools and databases, but this is still a crude approximation to a lifetime of experience.

Continual learning and self-improvement: Humans learn continuously from experience, mistakes, and changing environments. We do not need to be “shut down and retrained” from scratch. By contrast, today’s large-scale models are mostly trained in big offline batches just once. Updating them is just too expensive and time consuming.

Generalisation across domains: Human beings routinely jump between very different tasks and settings: cooking, arguing about politics, comforting a friend, fixing a motorbike, etc. Current AI is powerful but it can fail badly outside its training distribution or when the environment shifts.

Robust world models, causality, and embodiment: Humans are embodied creatures with a sense of physics, social cues, and cause-and-effect that comes from living in, and acting on, the world. Most current models operate on text, images, or other data streams with no direct physical presence. They can imitate causal reasoning, but they do not yet have a reliable, grounded understanding of how actions change the world.

Because of these gaps, expert estimates of “when AGI will arrive” span an extremely wide range. Some surveys find a substantial minority of AI researchers giving something like a 25% chance by the early 2030s, and around 50% by the 2040s. Other analyses are considerably more conservative, putting the chance of genuinely transformative AGI in the next 20 years at very low levels. Prominent industry figures send mixed signals. Sam Altman of OpenAI has suggested AGI may arrive “sooner than people think,’ sometimes hinting at timelines measured in single digits of years. Demis Hassabis of DeepMind has spoken of a 5–10 year horizon. Others, such as Yann LeCun at Meta, argue that AGI is “not around the corner” and will require many years, even decades, of fundamental research. The spread of these forecasts tells us something important: there is no scientific consensus on the AGI timeline. A realistic path to AGI, if it happens, will almost certainly not look like a single switch being flipped. It is more likely to be a series of steps: better memory here, more robust reasoning there, increasing integration with robots, new training methods. Each step changes what systems can do in real workplaces and social settings, and each happens within existing power structures and economic incentives. We also hear about Artificial Superintelligence (ASI); systems that greatly surpass human intelligence in all cognitive dimensions and can potentially improve themselves recursively. Many AI safety researchers think that if AGI is achieved, the step to ASI could happen quickly. Others believe this is speculative and that intelligence scaling will plateau long before such runaway effects.

The Hype

While key players talk about an upcoming AGI breakthrough leading to technological utopia, many commentators have started describing the current moment as an AI bubble. Institutions like the IMF and Bank of England openly warn that AI-related stock valuations could unwind sharply, drawing explicit parallels with the late-1990s tech boom.

What is actually going on underneath the rhetoric? Part of the answer lies in financial engineering around hardware and infrastructure. Nvidia, the main supplier of high-end AI GPUs, has become the central node. Analysts describe a kind of “vendor financing”: Nvidia takes equity stakes or guarantees loans so customers can raise debt to build data centres filled with Nvidia chips. Alongside this, a new market of GPU-backed loans and securitisations has appeared, where cutting-edge chips are used as collateral for billions of dollars of credit, reminiscent of the late-1990s telecom bubble, where equipment was used to back risky lending on the assumption that demand would stay sky-high.

States are also heavily involved. Governments from the US to South Korea are pouring public money into “AI infrastructure”, subsidising data centres, power projects and bulk GPU purchases in the name of national competitiveness. South Korea recently announced joining the UAE’s Stargate project, billed as the world’s largest AI data campus outside the US. Seoul presents this as a strategic move to become a “regional AI hub,” with Korean ministries talking about making the country a “top-three global AI leader’ and the “AI Capital of Asia.” The project involves over KRW 30 trillion in initial investment.

On the software side, companies like OpenAI show extraordinary revenue growth but not yet profitability. OpenAI’s annualised revenue climbed from about $2 billion in 2023 to over $13 billion by mid-2025. Yet disclosures indicate roughly $5 billion in losses on $3.7 billion revenue in 2024, and a projected $14 billion in losses for 2025, with hopes of turning cash-flow positive only around 2029. Even the most “successful” AI lab has not yet demonstrated a stable, profitable business model at the scale implied by its valuation.

Put these pieces together and you get a picture where capital is chasing a story: AGI is coming soon, and whoever builds it will command enormous profits and geopolitical power. Therefore, it is rational to spend trillions now on chips, data centres, and models, even if current uses are modest and margins thin. This is where the AGI narrative plays a crucial role—as a justification device for extreme valuations and huge capital expenditures. The more distant and transformative the promised future, the easier it is to wave away short-term contradictions.

An AI Utopia?

Elon Musk recently summed up a common claim at a US–Saudi investment forum: in 10 or 20 years, he said, work will be “optional,” money will become “irrelevant,’ and AI and robotics will “eliminate poverty” and make everyone wealthy. It sounds almost like a futuristic socialist utopia: abundance, no work, no poverty. But when a billionaire owner of car factories and data centres tells us his technology will abolish work and poverty, we should be more than a little suspicious.

The problem is straightforward. Under capitalism, robots are not “society’s robots”; they are private property belonging to corporations and wealthy investors. The gains in productivity do not automatically appear as leisure and security for the majority; they appear first as higher profits for owners, and as job losses for workers. Musk’s utopia of “optional work’ means, in practice, that work is “optional” for the capitalist—he can choose not to hire you at all.

Musk claims AI will eliminate poverty. But we could eliminate poverty with the technology we already have. The world produces more than enough food, housing materials, and medical knowledge to ensure a decent life for everyone. Persistent poverty is a structural feature of capitalist accumulation—a permanent pressure on workers that disciplines them to accept low wages and bad conditions. A reserve army of the unemployed is not an accident of capitalism; it is one of its operating mechanisms.

The promise that “AI will make everyone rich” functions to justify massive investment while deflecting attention from who will own these systems and who will benefit. It suggests we can keep capitalism’s basic structure intact and somehow still get socialist utopia results. History gives us little reason to believe this. Under capitalism, new technologies serve profit first, not human need.

AI and Human Labor

AI and robots are not some extreme deviation from what has come before. Since the industrial revolution, capital has constantly introduced new machines that change how work is done. For workers, this has always had a double character: machines can reduce brutal toil, but are also used to squeeze more output, increase labor intensity, and keep wages under pressure. The tension is not between “humans” and “machines’ in the abstract, but between workers and the way machines are deployed under capitalism.

Workers have not been passive. The history of the labor movement is full of struggles over how new technology is introduced. In South Korea, Hyundai workers have repeatedly fought for job security in the face of restructuring and automation, from the mass strikes of 1987 through to the present. In 2017, the Hyundai Motor union explicitly demanded guarantees that robotics and AI would not be used as a pretext for layoffs.

At Samsung’s semiconductor plants, the National Samsung Electronics Union—30,000 members strong—shattered the company’s 40-year ‘no-union’ tradition with their first-ever strike in 2024. They fought for pay and conditions under the exemption to allow 64-hour work weeks justified by ‘global competition.’ Their struggle reveals a crucial truth: even in the most automated facilities, capital still needs human workers to squeeze.

From a dialectical perspective, the development of capitalism’s “forces of production” is marked by long periods of quantitative change punctuated by qualitative leaps. We can see this pattern in the current AI boom itself. Large language models grew incrementally - more data, more parameters, better optimisation - but at some point those quantitative increases crossed a threshold and produced a qualitative shift in capability. What had looked like “just scaling” suddenly manifested as something different in kind, not just degree. This is a real leap, but it is not a miracle; it is prepared by the prior accumulation of compute, data, algorithms and the labour of thousands of engineers and researchers.

Marx wrote that “at a certain stage of development, the material productive forces of society come into conflict with the existing relations of production.” A sharp jump in AI capability is exactly that kind of development in the forces of production. But even dramatic leaps in technological capabilities do not teleport us into a new society overnight. They have to be materialised in data centres and factories, translated into software products, wired into logistics systems, and imposed in actual workplaces. That is where the existing relations of production reassert themselves, and where class struggle becomes decisive.

The impact of AI and robotics therefore appears in very concrete, but uneven ways. A car factory reorganises a production line around more robots and fewer workers. An office introduces AI tools that allow managers to demand more output from the same staff. None of these changes on their own bring an immediate “end of work”, but each alters the balance of power, the pace of work and the bargaining position of workers. The quantitative spread of automation, punctuated by qualitative jumps in capability, is refracted through the relations of production into wage levels, job security, working hours and control on the shop floor.

We should resist two opposite myths: the capitalist utopia where AGI abolishes work automatically with no struggle, and fatalistic doom where workers are swept away with no chance to respond. The historical pattern is more complex: capital pushes new technology to increase exploitation, but workers push back, sometimes successfully. AGI and advanced robotics will fit into that same pattern, unless we consciously fight to change the underlying social relations.

The key question is not “Will AGI kill all jobs overnight?” but “Who will control the pace and direction of automation, and how will the gains be distributed?’ If workers organise, they can fight to turn each technological advance - whether incremental or abrupt - into an opportunity for shorter hours instead of layoffs, safer work instead of intensification, and democratic control over how new machines are used. The future of human labor under AGI is not written in the code; it will be written in the struggles around how that code is used. And while technical breakthroughs can be abrupt, their social and economic consequences unfold through concrete institutions, material constraints and class struggle.

The Automation Paradox

To understand the deeper economic contradictions in the AGI-utopia story, we need to examine how capitalism actually generates profit. If we imagine a world where robots mine raw materials, run factories, repair other robots, and manage logistics—doing almost everything humans currently do—we face a question with concrete implications: if capitalism depends on extracting profit from human labor, what happens when human labor is no longer necessary?

We can make a basic distinction about commodities that helps address this question. Every product under capitalism has two sides:

Use value is what something actually does for people—food feeds you, a house shelters you, a phone lets you communicate. This is about the physical or practical usefulness of things.

Exchange value is what something trades for on the market, expressed in price. This ultimately reflects the socially necessary labor time needed to produce it—not the labor of one particular worker in one particular factory, but the average labor time under normal conditions.

These two dimensions don’t always align. Clean air has enormous use value but no exchange value because it isn’t produced as a commodity. A luxury handbag has only modest use value but high exchange value. The key point is that exchange value is not a natural property of objects. It’s a social relation between people—specifically, between people who produce commodities for sale under capitalism.

Robots and AI systems can clearly produce use values. In our hypothetical future, they could generate vast quantities of food, housing, medical treatments, and consumer goods with minimal human effort. The material wealth of society—the sheer abundance of useful things—could be extraordinary. But here’s the problem: this abundance of use values doesn’t automatically translate into exchange value or profit under capitalist rules.

Living Labor and Dead Labor

In Marx’s framework, machines represent “dead labor”—past human labor crystallized into tools and infrastructure. Workers are “living labor’—the only element that can create new value beyond the value of inputs.

Machines gradually transfer their value into products as they wear out (depreciation). A robot costing $100,000 that lasts ten years adds about $10,000 per year to whatever it produces. But it doesn’t create surplus value; it merely passes on its own cost. Living labor creates surplus by working beyond the time needed to reproduce its own wage.

Consider a simplified example. A factory owner invests $70,000 in machinery and materials and $30,000 in wages. Workers produce goods that sell for $120,000. The machinery and materials transfer their $70,000 value into the product. Wages are recovered. That leaves $20,000 as profit—surplus value created because workers produced more value than they received.

Now imagine near-full automation: $100,000 in robots and materials, essentially $0 in wages. The robots produce goods. But what can these goods sell for in a competitive market?

Under competition, prices tend to gravitate toward costs plus an average rate of profit across the economy. But if our hypothetical factory has eliminated human labor, and if this is happening across many sectors, where does the surplus come from?

In the short term, an individual capitalist who automates ahead of competitors can make extra profits by producing goods more cheaply while still charging near the old price. But as automation spreads through the sector, competition forces prices down toward the new, lower costs. The temporary advantage disappears. And if this process happens economy-wide—if robots and AGI replace human labor across most sectors—you face a systemic problem.

In Marx’s analysis, if you progressively eliminate living labor from production, you are simultaneously eliminating the source of surplus value. You might still have massive productive capacity and mountains of useful goods, but the mechanism that generates profit has been hollowed out. Capital’s dream of production without labor costs becomes self-undermining: you can have abundant use values, or you can have a functioning profit system, but not both indefinitely on the same basis.

This isn’t just a theoretical puzzle. We can already see versions of this playing out. Industries with very high automation and low labor content—think of highly automated warehouses—often show intense competition that squeezes margins despite enormous productivity. Profit has to be extracted somewhere, which is why such firms often rely on monopoly positions, excess profits through intellectual property (which, unlike true rent, tend to erode over time as technological advantages fade), or financial engineering rather than producing and selling commodities in competitive markets.

Why This Matters

If capital’s drive toward automation ultimately undermines its own profit base, we should expect capitalism to approach full automation in a crisis-ridden, unstable way. We should expect repeated cycles where firms automate to gain competitive advantage, only to discover they’ve collectively undercut the market for their products. We should expect enormous pressure to create artificial scarcity, rent-seeking, and financialization as alternatives to profit from production. We should expect enormous pressure to create artificial scarcity, strengthen monopoly power, and expand financialization as alternatives to profit from production.

The contradiction between capitalism’s drive to eliminate labor and its dependence on labor as the source of profit doesn’t disappear with better technology. It points toward a deeper structural problem: the capitalist organization of production increasingly conflicts with the productive potential of modern technology. Marx’s analysis of value and automation doesn’t mean robots can’t produce abundant useful goods. It means capitalism can’t coherently organize that abundance. The future isn’t a technical question about what robots can do. It’s a political question about who owns them and in whose interest they operate.

Beyond Technological Salvation

The AGI dream functions today much as the industrial revolution did for Victorian capitalists: a promise that the next wave of machinery will finally resolve the tensions inherent in the system. But the contradictions we have traced—between use value and exchange value, between capital’s drive to eliminate labor and its dependence on labor for profit, between technological abundance and artificial scarcity—are not technical problems awaiting technical solutions. They are social contradictions rooted in private ownership of the means of production.

AGI will not abolish capitalism’s contradictions. It will sharpen them. The greater the productive potential of the technology, the more absurd it becomes that access to the necessities of life depends on selling your labor to those who own the machines. The utopia that Musk and others promise—abundance without struggle, wealth without redistribution, freedom without collective ownership—is not a destination we can reach by following the road of private accumulation. It is a destination that requires changing the road entirely.

The real choice is not between humans and machines. It is between a society organized around profit and one organized around human need. AGI does not answer that question. It only makes it more urgent.