The popular fear is that AI “wipes out jobs,” and therefore threatens workers first. That is an incomplete diagnosis. The deeper systemic risk is that sufficiently capable AI can weaken the economic mechanisms that make capitalism politically stable and economically self-reinforcing: wages as the primary channel of mass purchasing power, competitive markets that disperse profits over time, and broad social consent that the system is fair enough to endure.
This is not a claim that capitalism must collapse. It is a claim that advanced AI creates a credible pathway where capitalism, left to its default incentives, can undermine its own foundations faster than it “extincts” the working class.
Capitalism’s hidden dependency: the wage–demand loop
In most capitalist economies, the majority of households obtain income primarily through labor. Firms pay wages. Households spend wages. That spending becomes revenue. Revenue funds investment. Investment creates jobs. The loop is not morally pretty, but it is mechanically powerful.
When a technology increases productivity by complementing workers, the loop can strengthen. When a technology increases productivity by replacing the economic importance of labor, the loop can weaken. This is exactly why central bank and academic work tracks “labor’s share” of national income. A sustained decline in labor’s share is not merely a worker problem; it is a macro stability problem. The Philadelphia Fed has argued that generative AI uniquely raises the risk of a long-run decline in labor’s share, potentially threatening the historical stability of that share even if “full employment” is maintained.
In other words, you can have people “employed” while labor becomes less essential to value creation and bargaining power. That is a different kind of disruption than a short-term unemployment spike.
AI is more plausibly labor-replacing at scale than prior automation waves
Classic automation tends to replace specific tasks while creating new tasks where humans retain comparative advantage. That “task creation” is one reason capitalism historically survived large technological shocks without permanently collapsing the labor share.
The problem is that frontier AI targets general cognitive tasks across many occupations, including tasks that used to anchor middle-class wages: drafting, analysis, customer interaction, coding assistance, planning, and coordination. When automation races ahead of new human-advantaged tasks, standard task-based economics predicts downward pressure on labor share and potentially wages, with gains shifting toward capital owners.
That is not a speculative ideological claim. It is a mainstream model result: more automation tends to increase capital’s share and reduce labor’s share unless offset by new tasks that favor labor.
The “production without consumers” risk
If AI allows firms to produce more with fewer labor costs, profits can rise while wage income stagnates. That sounds fine at the firm level, but it introduces a system-level contradiction: who buys the output?
Capitalism can survive pockets of automation because displaced workers can move, wages can adjust, new jobs can emerge, and credit can smooth demand. But if AI diffusion becomes broad and persistent, the economy risks chronic demand weakness: output capacity rises while mass purchasing power lags.
This is why labor share matters so much. A “once-in-a-lifetime” decline in labor’s share is not just distributional; it can mechanically destabilize aggregate demand. The Philadelphia Fed’s warning is precisely about this kind of structural shift.
Capitalism’s second dependency: competition that erodes excess profits
A second stabilizer in capitalism is that, in theory, high profits attract entrants, competition increases, and excess returns get competed away. AI can weaken that stabilizer by amplifying scale effects and barriers to entry.
Frontier AI is characterized by high fixed costs (compute, data infrastructure, talent, compliance, distribution) and very low marginal cost per additional unit of output. That combination can reinforce winner-take-most dynamics: the best models and best distribution channels get more usage, more data, more integration, and more capital to reinvest.
When that happens, the system drifts away from competitive capitalism toward durable rents. People still “work,” but a rising share of economic surplus accrues to a narrow set of owners controlling AI capital and platforms. Evidence-based policy institutions explicitly flag capital-share and return-to-capital channels as central to AI’s inequality risks.
The threat here is not that workers vanish. It is that capitalism becomes less like a broad-based market system and more like a concentrated rent extraction machine.
Why this threatens capitalism more than workers
Workers can be harmed severely without capitalism being threatened. Capitalism has historically tolerated high inequality, weak unions, precarious work, and churn. The “working class” can suffer while the system persists.
What capitalism cannot tolerate indefinitely is a sustained legitimacy crisis combined with structural mechanisms that keep concentrating gains while weakening the wage-based demand foundation. When the median household feels permanently locked out of progress, the political system responds: antitrust, taxation, labor regulation, industrial policy, nationalization pressures, or populist backlash. That is not an external moral critique; it is how political economy reacts when distribution becomes incompatible with social consent.
Recent reporting underscores that the AI buildout is already driving extraordinary capital concentration and investment intensity, raising questions about profitability, market structure, and sustainability.
The most important nuance: this outcome is not inevitable
A strong pro-capitalism rebuttal is that AI can raise productivity so much that living standards rise broadly, even if labor is partially displaced, especially if institutions adapt. The IMF explicitly notes that distributional outcomes depend on complementarity, productivity magnitude, and policy choices.
Empirically, some data so far suggests AI adoption has not yet produced widespread layoffs, and may coincide with firm growth in certain settings.
That nuance does not weaken the “capitalism risk” thesis. It strengthens it: if outcomes depend on institutions, then the real question becomes whether capitalism can reform itself fast enough to keep the loop intact.
What “capitalism surviving AI” actually requires
For capitalism to remain recognizable and stable under advanced AI, it likely needs some combination of:
Broadening ownership of AI capital (so returns do not concentrate purely to a small owner class)
Redistribution mechanisms that preserve mass purchasing power (tax and transfer, social dividend models, wage subsidies, reduced hours with income support)
Aggressive competition policy and interoperability rules to prevent permanent AI platform rents
Labor bargaining modernization so productivity gains translate into household income, not only capital returns
Public investment in human-complementary task creation, not just efficiency automation
This is the crux: AI does not have to “end work” to threaten capitalism. It only has to make labor economically less central while concentrating gains. That combination is historically how market systems slide into instability, backlash, and forced redesign.
Join the conversation! Your thoughts help the community grow.