Is nationalization the answer to AI? The wrong question gets in the way. The short answer is no — and understanding why requires a Marxist economic framework, not a nationalist one.

AI poses genuine threats to the working class. On April 26, 2026, Sen. Bernie Sanders hosted a forum bringing together leading U.S., Canadian, and Chinese AI scientists to address several existential threats posed by the drive to develop superintelligent agents capable of performing any human task. The scientists agreed on the threats: superintelligence as an existential risk, loss of human control, and massive job displacement. They differed only on timing — estimates for achieving such agents ranged from one to 20 years — and on the nature of these agents themselves.

Pope Leo XIV’s 2026 encyclical Magnifica Humanitas (“Magnificent Humanity”) warns that artificial intelligence must be “disarmed” to serve the common good rather than corporate domination. While not inherently evil, AI threatens human dignity, jobs, and truth if left unregulated, requiring developers and governments to prioritize moral oversight over unbridled efficiency

The threats are real. But the policy framework most often proposed in response — nationalization — starts from the wrong premise. AI has already transcended national boundaries. Its impacts on the working class of every country cannot be partitioned by nationality.

To understand both the threat and the correct response, we need to return to basics.

The Marxist foundation: Labor Is the source of value

One of the enduring contributions of Marxist political economy is the labor theory of value: human labor-power (variable capital) is the source of value in capitalist production and reproduction. “Reproduction” here means capitalism’s capacity to perpetuate the wage-labor system of class relations — not just produce goods, but reproduce the conditions that keep workers selling their labor to owners.

The surplus value extracted from labor power is not a simple commercial exchange. When a worker sells their labor, the product of that labor is retained and sold by the owner. The worker receives a wage; the capitalist keeps the difference between the wage and the value the worker produced. This mechanism drives capitalism’s insatiable demand for growth — orders of magnitude greater than predecessor systems like mercantilism.

Until recently, it was possible for economists to disagree about whether the exchange value of a commodity reflected abstract human labor (as Marx argued) or Keynes’ “animal spirits.” AI makes that question unavoidable.

The fundamental contradiction: What happens to profit without labor?

AI, in standard Marxist terms, is constant capital — a tool, like a machine. Some argue it should be classified as variable capital because of its generative abilities, but either way the question is the same: if AI displaces wage labor, where does profit come from?

Per the labor theory of value, the answer under capitalist relations is: nowhere. Rents have outlasted capitalism’s predecessors, but sustaining rent-based accumulation — with its attendant forms of coercion — in an era of theoretical abundance makes no sense as a long-term system.

This is not a distant hypothetical. It is the internal contradiction at the heart of the current AI buildout.

The Bubble: $700 billion looking for a business model

The “Magnificent Seven” — Apple, Microsoft, Nvidia, Alphabet, Amazon, Meta, and Tesla — have spent two years setting successive records in AI investment, with combined spending projected at roughly $700 billion this year alone.

What are they getting for it? Wired reported in October 2025 on economists Brent Goldfarb and David Kirsch’s framework for identifying tech bubbles, drawn from 58 historical cases. Their four indicators: abundant uncertainty, “pure play” companies whose survival depends entirely on the bubble technology, novice investors underestimating risk, and compelling but false commercial narratives. AI checks every box. Novice investors remain the biggest vulnerability.

The revenue picture confirms the concern. An MIT study found that 95% of firms adopting generative AI did not profit from the technology. Economist Mike Roberts noted in February 2026 that nearly three years after AI dominated Silicon Valley headlines, most major players — Nvidia excepted — still have no demonstrated long-term business model. OpenAI’s Sam Altman famously explained his monetization strategy as simply asking AI how to make money. The business models improved somewhat, but not by much.

The infrastructure behind the hype: data centers and the energy crisis

Behind the software hype lies a staggering physical buildout. AI has completely transformed data center economics. According to Princeton University’s Markus Center, a frontier data center with 200 megawatts of power capacity costs $8.2 billion. The planned U.S. data center capacity of 200 gigawatts implies $8.2 trillion in capital expenditure over roughly a decade — exceeding every historical infrastructure boom, including railroads, electrification, highways, and telecom fiber. AI-related investment now accounts for essentially all U.S. GDP growth.

Half of that investment goes to computing equipment, another third to facility infrastructure, and the rest to new power generation. The hyperscalers — Meta, Google, and the rest — have transformed the market from multi-tenant real estate into single-tenant AI campuses with bespoke hardware requirements.

It would not surprise Marx that raw electric power is where the rubber hits the road. This infrastructure dependency is the real terrain of a future crisis — not software job displacement alone.

The Citrini scenario: viral, influential, and wrong on Marx

In February 2026, the financial analyst group Citrini Research published a “Global Intelligence Crisis” report that amassed over 22 million views on X after triggering a sell-off in software stocks. Its scenario: AI agents replace human labor across the economy by as early as June 2028, producing unemployment above 10%, a stock market crash of 38%, and a credit and mortgage market meltdown. Unlike previous technological disruptions, Citrini argued, this one would not generate replacement jobs — every new role AI creates renders dozens of existing ones obsolete at a fraction of the pay.

Many mainstream commentators called the Citrini paper “pure Marx” because it predicted a consumer collapse with no recovery — the end of capitalism. They were wrong.

As economist Mike Roberts points out, a consumer-led collapse is not Marx’s theory of crises, even though most mainstream and many left economists believe it is. Marx explicitly and repeatedly rejected the underconsumption theory of crisis.

Marx’s actual theory is based on overinvestment and the falling rate of profit. Capitalists invest in technology and machinery to shed labor and cut costs, temporarily raising profitability. But since only human labor creates value, replacing labor with machines reduces the very source of profit. Over time, profitability falls. Eventually capitalists stop investing — a capital strike. Workers are then laid off, and only at that point does consumption collapse. The consumption crisis is the result, not the cause.

Mainstream critics of Citrini are partly right: if AI raises productivity enough, prices fall and purchasing power holds. But they miss the real doom scenario. Rising productivity means less value is generated, which means profitability falls over time. That is the natural brake Citrini’s model lacks — and it is already built into the system.

It is also worth noting that roughly 60% of U.S. workers today are employed in occupations that did not exist in 1940. Historically, technological change has been the main driver of net employment growth. The timeline of displacement and adaptation matters enormously.

Why nationalization is the wrong frame

The AI agent economy is real and growing. Consumer agents already book travel and make purchases autonomously. The global AI agents market was valued at $5.4 billion in 2024 and is projected to reach $236 billion by 2034. A growing share of business operations will have no human workers — only agents interacting with other agents.

But these agents are digital. They do not make physical goods. Deploying them in the physical world requires extremely expensive robots. The transition is neither as fast nor as clean as the hype suggests.

More fundamentally: the problem is not national. AI research and development is expanding in every country. The existential impacts on the working class cannot be bounded by national borders. Nationalization — placing AI infrastructure under the control of one state — does not address the international character of the threat and risks reinforcing exactly the nationalistic competition Sanders’ forum was convened to warn against.

The correct demands: internationalization and public control

A major collision between private or nationalistic AI platforms and the interests of democracy and socialism now seems inevitable. The AI crisis crystallizes the concrete demands of the anti-monopoly coalition. Drawing on Sanders’ forum with U.S. and Chinese scientists:

AI must be internationalized — development and governance cannot remain in the hands of a handful of monopoly platforms. The energy infrastructure required to power AI must be internationalized. The financial services required to fund both the AI buildout and just transitions for displaced workers must be placed under public control. The scale of the challenge demands not just new policies but a new government — one answerable to the working class, not to the Magnificent Seven.

Nationalization, in isolation, would be the wrong tool for an inherently international problem. The working class has always had to think and organize across borders. AI makes that more urgent, not less.

The opinions of the author do not necessarily reflect the positions of the CPUSA.

Images: AI Sistine Chapel by dazonn.com. CC BY-NC 4.0. Difference between usage of conventional servers and water cooled servers in data centers by Ilya Plekhanov/Burning River Brigade (Facebook). CC BY-SA 4.0.