ON MAY 1, 2026, the United States Department of Defence finalised classified artificial intelligence contracts with eight corporations: OpenAI, Google, Microsoft, Amazon, Nvidia, Oracle, SpaceX, and the startup Reflection AI. These agreements will deploy large language models on classified military networks. The one major AI laboratory that was excluded, Anthropic, was dropped because the Trump administration blacklisted it for insisting on safety guardrails. The message was unambiguous: the technology will serve the war machine, and those who demand even minimal safeguards will be cut out. This single episode captures the logic of generative AI under imperial capitalism more precisely than any technical debate about hallucinations or reliability ever could.

The growing backlash against generative AI, now visible across classrooms, labour movements, artist collectives, and communities fighting data centre construction, is often framed as a response to a technology that has been poorly managed. This framing is inadequate. Generative AI has not been mismanaged. It has been managed with ruthless consistency in the interests of the tech monopolies that own it, deploy it, and profit from it. The harms are not aberrations. They are the system functioning exactly as the relations of production demand.

The concentration of capital tells the story most starkly. In 2026, four corporations alone plan to spend over $700 billion on capital expenditure, the bulk of it driven by AI: Amazon, approximately $200 billion; Alphabet, $190 billion; Microsoft, $190 billion; and Meta, between $125 billion and $145 billion. Global AI capital expenditure will reach $1.3 trillion by 2030, according to UBS. A study by the MIT Media Lab found that 95 per cent of corporate generative AI pilots fail to deliver any measurable return. The speculative character of this investment is not a flaw in an otherwise rational process. It is accumulation for the sake of accumulation, the classic capitalist compulsion to expand regardless of productive use in hopes of securing a monopoly position, with the inevitable crisis deferred onto workers and the public.

The assault on labour is equally systematic. In the first six months of 2025, 77,999 technology sector jobs were eliminated and attributed directly to AI. Wall Street banks have announced plans to cut 200,000 positions over the next three to five years. The World Economic Forum projects 92 million jobs displaced globally by 2030 as a result of AI, automation, and related labour market shifts. In India, the devastation is concentrated precisely where it can least be absorbed. Tata Consultancy Services cut 12,200 workers in July 2025. In the 2024-25 hiring cycle, India's major software exporters recruited only 70,000 to 80,000 fresh engineers, the lowest in two decades, from a pool of 1.5 million graduates produced every year. The ratio tells the story: twenty applicants for every position, in a sector that was supposed to be India's engine of middle-class employment.

Most chillingly, generative AI has been integrated directly into military kill chains. In Gaza, the Israeli military deployed two AI systems, Lavender and Gospel, to automate the identification and targeting of Palestinians. Gospel generated 100 bombing targets per day; human analysts working traditional methods might produce 50 in an entire year. Lavender compiled a database of tens of thousands of Palestinian men flagged as targets, with an acknowledged error rate of approximately 10 per cent. Military sources revealed that for every junior operative marked by Lavender, the army authorised the killing of 15 to 20 civilians. Human review of each target amounted to, in the words of one intelligence officer, 20 seconds of approval, a rubber stamp.

Operation Epic Fury demonstrated that Gaza was not an aberration. AI-enabled kill chains are now embedded in imperial military doctrine. When the United States and Israel launched their attack on Iran in February 2026, Palantir's Maven Smart System, the Pentagon's flagship AI targeting platform, enabled the striking of over 1,000 targets within the first 24 hours. By mid-March, the number had crossed 6,000. The system compressed the kill chain from hours to minutes, generating prioritised target lists faster than human operators could review them. Among the consequences was the destruction of the Shajareh Tayyebeh elementary school in Minab on the very first day of the war. A US Tomahawk cruise missile is assessed to have struck the school, killing at least 170 schoolgirls. In February 2025, Google had quietly removed from its public website the explicit pledge not to develop AI for weapons or surveillance. The progression from Silicon Valley boardroom to automated kill list is not a perversion of the technology. It is the technology fulfilling its function under imperialism.

The environmental costs are staggering. Global data centre electricity consumption is projected to reach 1,050 terawatt-hours by 2026, more than double the 460 TWh consumed in 2022. The water footprint of AI systems alone could reach 765 billion litres in 2025. The carbon footprint is estimated at 32 to 80 million tonnes of CO2 in 2025. These costs are socialised across communities, grid infrastructure, and the global climate. The profits flow to the same handful of corporations. In the second quarter of 2025 alone, activists in the United States stalled $98 billion in data centre projects, from Virginia to Indiana to Arizona, as residents confronted the reality of what "AI infrastructure" means for their water, their electricity bills, and their land.

Meanwhile, the technology is being weaponised against the vulnerable. Deepfake content has exploded from 500,000 files in 2023 to a projected 8 million in 2025. Between 96 and 98 per cent of all deepfake videos are non-consensual intimate imagery; 99 to 100 per cent of victims are women. AI-related academic misconduct now accounts for 60 to 64 per cent of all cheating cases in higher education globally. Deepfake-enabled fraud exceeded $200 million in losses in a single quarter. These are not unintended consequences. They are the inevitable products of a technology released without safeguards because safeguards slow the rate of capital accumulation.

None of this is happening because regulation failed. It is happening because the regulatory apparatus has been captured. More than 450 organisations now lobby on AI in the United States, up from 6 in 2016. Meta launched a Super PAC in 2025 with tens of millions of dollars to oppose state-level AI regulation. The US has no federal AI law. In Europe, the Commission's "Digital Omnibus" proposals threaten to gut the AI Act and GDPR before full implementation; 69 per cent of Commission meetings in 2025 were with business groups and only 16 per cent with civil society groups and NGOs. The regulatory apparatus has been captured before any meaningful regulation was enacted. Under neoliberal capitalism, the state does not stand above the market to regulate it. The state serves capital.

The backlash, however, is no longer confined to commentary. In February 2026, hundreds marched past the London headquarters of OpenAI, Google DeepMind, and Meta in one of the largest protests against AI yet seen. In the United States, an unlikely coalition of labour activists, democratic socialists, church leaders, and even sections of the political right signed a Pro-Human AI Declaration. A 2025 Pew Research poll found five times as many Americans concerned as excited about the increased use of AI in daily life. College students in the class of 2026 cited as their primary reason for refusing AI tools: "I think it is important I do this myself." This is not technophobia. It is the instinctive resistance of working people, students, and communities to a technology that is being used against their interests by a class that has no intention of sharing the gains.

The parallel with the dot-com bubble is instructive. When that bubble burst, it destroyed speculative capital but left behind the internet infrastructure on which the current tech monopolies and pervasive surveillance were built. The AI bubble, whenever it corrects, will leave behind a permanent infrastructure: data centres draining water and power, a labour market stripped of entry-level work, a military apparatus with automated kill chains, a surveillance architecture that will not be dismantled because share prices fell. Under capitalism, the debris of each speculative cycle becomes the foundation for the next round of extraction.

None of this is a verdict on artificial intelligence as such. AI has enormous potential for human progress: in healthcare diagnostics, in climate modelling, in the discovery of new materials, in planning to meet human needs rather than maximise private profit. The technology is not the problem. The problem is what happens when a technology of this power is placed in the hands of a monopolistic oligarchy with deep ties to the imperial war machine, no democratic accountability, and every incentive to socialise the costs while privatising the gains. The resistance now building, from factory floors to university campuses to the communities surrounding the data centres, will only become effective when it recognises this: the fight is not against the machine. It is against the class that owns it.