Schumpeter’s Death: You’re Not Worried Enough About AI
While some leftists on the internet rage about people generating images with AI, the world is in the midst of the most sweeping development of productive forces in decades — one that, by the logic of competition, is unstoppable.
Artificial intelligence — a term that encompasses far more than generative AI (GenAI) — will, in the coming decades, sharpen the class contradiction within bourgeois states and its outward effects to a degree unmatched by any other crisis of recent decades. This is not a worst-case scenario or the possibility of a distant future, but the sole logical conclusion of a serious engagement with what lies ahead.
This conclusion consists of three parts: the massive contradiction between productive forces and relations of production (illustrated through Schumpeter’s “Creative Destruction”), the qualitative transformation that AI represents within imperialism, and the general dumbing down of the working class.
Productive Forces and Relations of Production
With every major surge in the development of productive forces — that is, the means required to produce commodities — the contradiction between capital and labour sharpens further. This contradiction, manifested here primarily through job losses and their consequences for people and regions, has historically been absorbed in the long run (quantitatively) by a comparable amount of labour demand emerging in a new economic sector.
When the steam engine replaced the looms of the cottage industries, the industrial reserve army of millions of formerly self-sufficient textile workers was, after short- and medium-term immiseration (a halving of real wages within a single working lifetime[1]), eventually absorbed by the gravitational pull of the newly emerging industrial centres (Manchester, the Ruhr Valley), which took in a comparable volume of labour. The drastically reduced production costs enabled by machinery caused global demand for fabrics and clothing to explode to such a degree that the new factories required thousands of workers to operate, maintain, and organise the logistics of the machinery. The workforce migrated out of necessity from the dying cottages into the halls of the textile metropolises, which — while sharpening the fundamental contradiction between capital and labour — nonetheless saw the sheer quantitative level of employment rise in the long run. The relocation of labour power took roughly one generation.
When microchip technology and the first industrial robots entered the factory floors of the 1980s, manual workers lost their livelihoods. In the long run, after short- and medium-term immiseration, this consequence was cushioned by the growth of the service and IT sectors, which absorbed a comparable volume of labour. The costs reduced by automation freed up capital for software development, logistics, and financial services, generating demand for labour in new fields. The workforce migrated from the factory floors into the offices of the information society to administer the economy’s infrastructure. Thus the expansion of the tertiary sector (the service sector) forestalled mass unemployment by deploying the reserve army of workers in the service of computer-aided administration.[2]
A study by bourgeois MIT economist David Autor, examining changes in US job task content between 1977 and 2018, shows that 64.5% of tasks eliminated during that period were “routine tasks” (regular, repeatable workflows). Conversely, 75.6% of newly added tasks were “abstract,” requiring skills such as logical reasoning, creativity, and interpersonal abilities.[3]
With the near-total economic incorporation of artificial intelligence, this mechanism — which Schumpeter understood as “Creative Destruction” — now faces several obstacles. Past sectoral shifts displaced specific forms of physical labour, repetitive movement, and other manual activity. That labour power was substituted within a given sector through automation and complemented in another sector in accordance with increased labour productivity. The (idealised) car painter substituted by a painting robot is (after short- and medium-term immiseration) absorbed as a logistics worker and generates greater labour productivity there — though this naturally does not show up in his wages.
With the foreseeable economy-wide deployment of AI, however, it is not the low- and medium-skilled occupations that are exposed to profit-squeezing rationalisation (as in previous sectoral shifts), but high-skilled occupations in the tertiary sector. In the Policy Brief of the “ai:conomics” project at Maastricht University, the researchers demonstrate empirically that it is precisely the tertiary sector that will be affected by AI-related rationalisation. AI is “more likely to affect workers across a wider skill and wage spectrum than previous technologies”[4]. This means there is no new sector to absorb (as in previous phases of “creative destruction”) the rationalised labour power of the service sector.
In short: the mechanism of “creative destruction” fails in the long run because AI substitutes precisely those cognitive and analytical activities that have until now served as the escape route for earlier waves of automation. AI is a “General Purpose Technology” that will potentially encompass every sector of the economy — ironically abbreviated as GPT.
The AI sub-sector of the tertiary sector (or the quaternary sector, as the digital occupations of the tertiary sector are often called) — which in future will concern itself with prompt engineering and other AI-related activities — does not come close to covering the mass of workers being rationalised and displaced by AI. In previous instances of creative destruction, the developed productive force ultimately required a comparable volume of labour for a correspondingly increased productivity; the deployment of AI does not. While the mechanical loom still required workers to operate machines, perform maintenance, and organise production flows, an AI can directly execute or drastically compress many cognitive tasks (e.g., text production, analysis, programming, or design). This means no new mass employment sector emerges on the same scale as in earlier technological upheavals.
Bourgeois economists draw partly absurd conclusions from the empirical data already available on labour market change. The Anthropic study “Canaries in the Coal Mine?” from last November documents that in the United States, “early-career workers” (aged 22–25) are already registering a relative employment decline of 16% due to AI — though the same does not hold for “experienced workers”[5]. The study distinguishes between “automation” and “augmentation”: in the first category, AI automates work; in the second, it merely supplements it. In the latter category, no effects on hiring are currently observable.[6] I hope the problem with both findings of the study is already apparent on reading: bourgeois-economic studies on AI usage treat AI as a fixed, punctual technical object. Both of the above theses would already be rendered obsolete after just a few more years of AI development.
Particularly striking are the conclusions of bourgeois-economic think tanks like CEPR[7] and Yale’s Budget Lab[8], which repeatedly emphasise that economy-wide AI integration has not yet produced a massive surge in productivity and that concerns about mass rationalisation are therefore unfounded. A CEPR study from February, examining 12,000 European companies, observes “that AI adoption increases labour productivity levels by 4% on average in the EU, with no evidence of reduced employment in the short run.” It is true that there have been no massive waves of AI-driven redundancies so far — but as outlined above: why would a company that can rationalise its workflows through AI not do so? The hurdles currently facing especially small and medium-sized enterprises in AI integration recede day by day, with every new AI model.
Fictitious Capital and the Rate of Profit
The conclusion to be drawn from the so far “modest” productivity gains should be precisely the opposite. Elliot Goodell Ugalde of Queen’s University argues in his paper “Artificial Intelligence and Crisis Economics: Marx, Financial Concentration, and Strategic Risk” that the massive investment in AI on the part of states and corporations is not fundamentally at odds with the production increases that have not yet materialised. He writes:
“Under conditions of stagnating productive returns, AI has become a privileged outlet for valorizing surplus capital […] the present conjuncture reflects a situation in which ‘the real barrier of capitalist production is capital itself’ (Marx 1894, pt. III, ch. 15), such that capital unable to secure adequate returns in production multiplies ‘accumulated claim[s], [and] legal title[s], to future production’ (Marx 1991, ch. 29, n.p.). The AI boom thus operates less as a durable resolution of stagnation than as a spatio-temporal displacement of overaccumulation pressures through enclave concentration, credit expansion, and promissory valuation.”[9]
Thus the massive investment in AI is, of course, an expression of overaccumulation — in this case, speculation on new technology with which the absent profits from production can be offset through fictitious capital. But this is not in contradiction with the very real productive potential that AI will unleash. Added to this is the fact that it is precisely large corporations that have the means to invest in (still) often speculative AI systems:
The CEPR study reveals: among US companies with more than 250 employees, around 45% have already fully implemented AI systems (by comparison: 90% of all German companies are experimenting with or have already introduced GenAI integration), while among smaller firms with 10 to 49 employees only around 24% have done so. Larger companies typically have greater financial resources, the necessary technical expertise, and the economies of scale that make it easier for them to bear the integration costs of new technologies.
While bourgeois economics in the cited studies treats AI merely as a punctual, fixed tool, a massive shift in the organic composition of capital is underway. The ratio of technical expenditure (constant capital) to human labour power (variable capital) is shifting so drastically that human labour — according to Marx the sole source of surplus value — is being systematically displaced from the production process.
AI occupies the tertiary and “quaternary” sectors — the last refuge of wage labour — thereby threatening the emergence of a permanent industrial reserve army for which there is no longer any new sector of absorption. Since AI does not merely supplement cognitive output but ultimately substitutes it, capital is in the long run undermining the very foundation of its own valorisation. While individual frontrunners secure enormous extra profits through automation in the short term, on the macroeconomic scale the value-substance of the system diminishes along with living labour.
The contradiction lies in the fact that the system produces ever more efficiently for a market whose solvent base — wage earners — it simultaneously rationalises away. The “canary in the coal mine” is not only the young employee who can no longer find a way in, but the entire principle of the valorisation of value through human labour, which founders on its own technological development — a development that stands in contradiction to the relations of production.
Imperialism
At the latest with the “One Big Beautiful Bill Act” (OBBBA) of the United States from July 2025, the age of AI as a qualitative development of imperialism began. The legislation tied large portions of military development to simultaneous dominance in AI leadership and prohibited legal regulation of AI companies at the state level — thereby aiming to unleash the full, massively subsidised development of AI as a productive force.[10] AI monopolies — namely Anduril, Palantir, Nvidia, and Meta — are more closely bound to US politics than any other branch of capital. The massive cuts to social spending accompanying the OBBBA signal the severity of the competitive stakes for imperialism.
The European Union preceded this and already in February 2025 launched the “InvestAI” initiative to “mobilise investments in artificial intelligence worth €200 billion,” along with the establishment of new EU funds for the construction of “AI gigafactories” totalling €20 billion.[11]
AI is understood not, like other developments in digital productive forces, merely as a complement to the existing — but as an entirely new, fundamental productive force whose leadership (and what it requires) will in future determine imperial wars and the partition of the world. Marx describes how capital subordinates science and the forces of nature in order to increase relative surplus value. In the era of the industrial revolution, the machine replaced human muscle power; in the era of AI, capital targets the subsumption of the “General Intellect” — of society’s collective knowledge — and its conversion into fixed forms of capital.
Lenin defined imperialism through five characteristics, of which the concentration of production and the rule of monopolies stand first. In the 21st century, this concentration manifests in an oligopolistic structure that controls the entire digital value chain. Companies like Nvidia, Microsoft, Amazon, and Alphabet control not only the software but also the hardware and cloud infrastructure necessary for training AI. The EU is attempting, through “InvestAI” and the construction of four to five gigafactories each equipped with 100,000 high-performance chips[12], to form its own monopolistic bloc to counterbalance US and Chinese dominance. These form the fixed capital without which participation in the world market becomes impossible.
The “InvestAI” initiative is explicitly described as a “CERN for AI”[13], which underscores the character of state-organised large-scale research and production, securing private investments through a “first-loss tranche” (the highest risk at potentially the highest return) from the EU budget. Here one sees the fusion of industrial and banking capital into finance capital — as described by Lenin — today taking the form of venture capital and sovereign wealth funds dictating the direction of technological development.
Finance Capital and Bubble Formation
While the stock valuations of the “Magnificent Seven” (Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, Tesla) climb to astronomical heights, real productivity growth across the broader economy lags behind expectations, as outlined above. Without speculation on AI, US GDP growth in 2025 would have stood at only 0.1 to 0.2%.[14]
This dependence on a technological “silver bullet” is a hallmark of imperialism’s parasitic character. Capital flees from stagnating real production into speculative bets on “Artificial General Intelligence” (AGI) in order to offset the tendential fall in the rate of profit. When these bubbles burst — as the “DeepSeek shock” in January 2025 suggested, when the more efficient Chinese model caused Nvidia’s market value to collapse by nearly 600 billion dollars — the imperialist centre responded not with massive state intervention.[15] Part of this response involves binding AI capital to the “military-industrial complex.”
Companies like Palantir and Anduril are today an integral component of the US war architecture. Alex Karp’s concept of the “Technological Republic” explicitly demands that Silicon Valley shed its scruples about cooperating with the Pentagon in order to secure the “hard power” of the West.[16] The real-world consequences of this cooperation can be observed in the deployment of AI systems in Gaza. Systems such as “Lavender” and “The Gospel” function as instruments of genocide, generating targets with minimal human oversight. These algorithms mark thousands of people as potential targets based on patterns in mass data, with “error rates” of 10%[17] and hundreds of civilian “collateral casualties” per high-value target being knowingly accepted. When Anthropic refused to make its models available for mass surveillance and autonomous weapons, the Pentagon threatened to classify it as a “risk to national security” and to invoke the Defense Production Act[18] to compel the company’s cooperation.
The supremacy of Western AI capital is, however, not absolute. The release of China’s DeepSeek-R1 model at the start of 2025 shook the assumption that only massive amounts of capital and the latest Nvidia chips (subject to export restrictions to China) can produce top-tier performance. DeepSeek employed more efficient algorithms and older hardware (H800) to achieve results comparable to OpenAI’s o1 — at a fraction of the cost. This triggered a crisis of US finance capital, since the entire valuation model of companies like Nvidia rested on the assumption of an enduring technological monopoly position. The imperialist reaction to DeepSeek was telling: rather than competing through better products, it forced the binding of other countries to the “American Tech Stack” through combined hardware-cloud-model packages. This corresponds to the tactic described by Lenin of international cartels dividing up markets and eliminating competitors through political and economic coercion.
Dumbing Down
In principle, humans “dumb down” with every external aid they use in performing cognitive work — but nowhere near to the degree seen with the use of GenAI, or LLMs in particular.
The comprehensive Massachusetts Institute of Technology (MIT) study “Your Brain on ChatGPT” from June 2025 used electroencephalography to compare 54 test subjects who wrote essays over several months without any aids (group a), using search engines (group b), and with the assistance of LLMs (group c)[19]. The study found the following:
Group c registered 55% reduced signal interactions in the brain compared to group a while writing their essays. 83% of writers from group c had difficulty citing their own essay afterwards[20]. Group a showed the highest signal interactions in alpha, delta, and theta brainwaves — the waves responsible for memory, semantics, and creativity. Participants in group c continuously reduced their own work effort over the course of the study and by the end were frequently copying verbatim the responses provided by ChatGPT. Group b showed nowhere near comparable signal interactions or behaviour, and over the entire course of the study registered a “significant increase in brain connectivity across all EEG frequency bands”[21]. The study’s authors conclude that “reliance on AI-driven solutions” could contribute to “cognitive atrophy” — a wasting of brain functions.[22]
The wage worker is in principle alienated from the product of the wage labour they perform — it is therefore only comprehensible that in their cognitive work they reach for every aid available to them. The absolute mass of pupils, students, and wage workers in the tertiary sector now turn to GenAI. Combined with the short-form video economy, what emerges is a subject that has unlearned how to productively absorb and process information.
Of course it is helpful to have Hegel explained by Gemini rather than spending months chewing through the Phenomenology of Spirit. Such utility can also be genuinely productive, making archaic language and complex subject matter more accessible. The overall effect within the existing relations of domination is, however, a different one: a submissive subject, maximally alienated from its own cognitive production and restricted in its capacity for absorption while simultaneously dependent on the filters of a handful of large monopolies — such a subject does not appear to have an interest in questioning the existing order, let alone wanting to change it.
What Is to Be Done?
Among parts of the left there prevails the idealist notion that one should simply stop using generative AI because it is bad for the environment and otherwise immoral. This is, of course, a peculiar appeal, given that not using generative AI carries no added value for resistance against it. Every productive force, and what it means for society, exists in relation to the relations of production and to how the products of that productive force are managed — the sharpening contradiction between productive forces and relations of production is an unalterable feature of capitalism. Opposing generative AI as such is as senseless as if the communist movement of the 1970s had organised resistance against industrial robots.
Individual resistance through the non-use of a thing rests on the false consciousness that the subject can, through an individual consumer choice, halt or morally correct the movement of the material relations of production. This, however, inverts cause and effect. It is not consciousness that determines social being, but social being that determines consciousness. When a new productive force such as generative AI today comes into being, this does not happen out of the moral whim of individual persons, but from the inner dynamic of capitalist production itself. Capitalism is compelled to continuously revolutionise the productive forces in order to increase surplus value and prevail in competition.
Whoever believes that abstaining from the use of such technology constitutes an act of resistance fails to grasp the structure of the problem. The machine — be it the mechanical loom, the industrial robot, or AI — is in itself neither revolutionary nor reactionary. Its social character only emerges through the relations of production within which it exists. The contradiction therefore lies not between human and machine, but between the developed productive forces and the capitalist relations of production, which confine these possibilities within the compulsion for profit, the concentration of property, and exploitation. Under different relations of production — ones not organised around the competition of capitals — AI could genuinely replace large portions of required labour power and relieve the working class. It would enable a life with more leisure in which people can pursue their own productive activities. Even fundamental problems with the implementation of a planned economy effectively cease to exist under conditions of AI-assisted production. We have written about this at length in “The New Planned Economy.”
While part of the left loses itself in moral debates about AI, one of the most sweeping developments of productive forces in decades is already underway. Artificial intelligence will drive the class contradiction in capitalist states to a new level — not because it rationalises only physical or repetitive labour, but because it attacks precisely those cognitive activities of the service sector that have until now served as the catchment basin for earlier waves of automation. Unlike previous technological upheavals, this may well produce no new mass employment sector capable of absorbing the displaced labour power.
Simultaneously, the organic composition of capital is shifting drastically: living labour — the sole source of surplus value — is being increasingly displaced from the production process. In parallel, AI is becoming the central field of imperialist competition between states and monopolies, which are investing enormous resources in controlling this technology. At the same time, its use reinforces a cultural tendency towards alienation from one’s own cognitive labour, as large parts of the population increasingly delegate their intellectual production to algorithmic systems.
The decisive point remains, however: these developments cannot be halted through individual boycotts. AI is an expression of the inner dynamic of capital itself. Whoever wishes to change its social consequences must therefore fight not the technology, but the relations of production within which it is deployed.
[1] https://www.tandfonline.com/doi/full/10.1080/00404969.2024.2437180
[2] https://thevoice.bse.eu/2021/02/23/automation-sectoral-reallocation/
[3] https://hai.stanford.edu/news/assessing-the-real-impact-of-automation-on-jobs
[4] https://aiconomics.eu/en/findings-insights/policybrief-4
[5] https://digitaleconomy.stanford.edu/app/uploads/2025/11/CanariesintheCoalMine_Nov25.pdf, p.1
[6] https://digitaleconomy.stanford.edu/app/uploads/2025/11/CanariesintheCoalMine_Nov25.pdf, p.3
[7] https://cepr.org/voxeu/columns/how-ai-affecting-productivity-and-jobs-europe
[8] https://budgetlab.yale.edu/research/ai-productivity-boom-dont-count-your-productivity-data-chickens
[9] https://www.queensu.ca/cidp/artificial-intelligence-and-crisis-economics-marx-financial-concentration-and-strategic-risk
[10] https://www.nbcnews.com/tech/tech-news/big-beautiful-bill-ai-moratorium-ted-cruz-pass-vote-rcna215111
[11] https://ec.europa.eu/commission/presscorner/detail/de/ip_25_467
[12] https://www.ga-alliance.eu/clien-alert-ai-continent-action-plan/
[13] https://background.tagesspiegel.de/digitalisierung-und-ki/briefing/europa-fehlt-die-grosse-ki-vision
[14] https://finance.yahoo.com/news/without-data-centers-gdp-growth-171546326.html?guccounter=1&guce_referrer=aHR0cHM6Ly93d3cucGVycGxleGl0eS5haS8&guce_referrer_sig=AQAAAGHdf2JUAMqUy7gQHAyBnwjhEHM7k_Y9Xdue-Bazp0fvbnD4AB9FQD2QlMxuh_h4tBs–dd97nX9pEIk76llxzbMdqeyBVOw-XXHyhGJKDkluU3MIDdjwRInyjGwME6h5zGrWS5cC3fQnFDyxZGltUL8yUESAQRgU0U91wafJx0s
[15] https://socialistchina.org/2025/10/24/intelligence-artificial-profits-fictitious/
[16] https://www.commonplace.org/p/alex-karps-technopolitik
[17] https://madhyamamonline.com/world/thousands-targetted-killed-in-gaza-using-israels-killer-ai-algorithms-1277172
[18] https://www.wsws.org/en/articles/2026/02/25/hhsk-f25.html
[19] https://arxiv.org/pdf/2506.08872 p.145 f.
[20] https://arxiv.org/pdf/2506.08872 p.146.
[21] https://time.com/7295195/ai-chatgpt-google-learning-school/
[22] https://arxiv.org/pdf/2506.08872 p.10
