https://portside.org/2026-07-11/ai-and-working-class

Technology is Contingent

Over two centuries ago, Britain faced high inflation during the Napoleonic Wars. The emerging industrialist class reached a novel solution to cut costs and raise profits: implement new machinery within textile mills, converting skilled labor into unskilled work that could be performed by anyone, even “apprenticed” orphans provided only room and board. 

As a response to this degradation of working conditions, the “Luddites” formed. They engaged in nighttime raids against physical capital, only targeting those machines that most threatened skilled work, and those employers who most drove down living standards. Luddites were labor agitators, working through direct action to halt the erosion of living standards in their industry during an era when collective bargaining was outlawed and the strike was not yet perfected. They were only defeated following ruthless suppression by the British government, including new laws making the destruction of machines punishable by death. 

Looking back from the perch of the present, the Luddites may seem, at best, to be a noble lost cause fighting against the tide of history. But they understood that history is contingent, and the technology the boss implements is but one of those contingencies. 

We face a similar cycle today, as bosses use new technologies to reduce costs and gain more control over workers. Indeed, the parallels are so strong that Gen Z college students, seeing dire working prospects upon graduation, are forming “Luddite” clubs across the nation. These young workers do not face the power loom, but a number of software technologies called “AI” (though they have little to do with artificial intelligence) including but not limited to “generative AI” (popular chatbots like ChatGPT, Claude, and Grok). 

We’ve been awash in AI and its discourse for years now, and virtually all visible changes have been negative. Businesses replace human customer service with low-quality bots. Scammers attempt to pass off AI-generated art, music, video, and books as legitimate. People fall in love with chatbots, descend into AI psychosis, or are egged into suicide by a “helpful” chatbot. Massive new data centers are constructed against the wishes of local residents. And yet billionaires continue to extol AI, and recently, have begun even chiding the general public for “not using it enough.”

To be fair to the media, negative stories are not soft pedaled—bad news drives clicks, after all. But even when the downsides are discussed, AI is almost treated as something inevitable—like a natural disaster. But like the power loom of old, how AI is developed is being shaped not by an inevitable tide of “progress,” but by the whims of individual capitalists.

In this case, the capitalists have made some rather bad bets.

Deflating the Bubble

A disturbing percentage of economic activity is now reliant on the AI industry. Some estimates suggest that between half and nearly all U.S. economic growth is either directly or indirectly from AI, from the boom in data center construction to the growing net worth of wealthy individuals with AI stocks. Indeed, discounting both AI and energy companies (the latter of which are now overvalued due to the ongoing crisis in the Strait of Hormuz), there has been no net increase in stock prices since the beginning of the year. The non-AI sections of the U.S. economy, while not in recession, are treading water. 

But, in spite of huge global spending on AI (some $2.5 trillion projected for 2026), the industry itself loses money. Elon Musk’s Grok lost $6.4 billion last year. OpenAI (owners of Chat GPT) lost a stunning $38.5 billion. And while the smaller Anthropic (owners of Claude, which focuses more on business services) claimed it will make an operating profit in the second quarter, the company used nonstandard accounting, made it clear they would lose money for 2026 as a whole, and lost $5.6 billion last year. 

There are businesses ancillary to the core of AI which make money, including the chip maker NVIDIA and companies constructing new data centers. However, high demand for their products only exists because money-losing AI companies survive. The AI industry is kept afloat by huge external investment by more established tech companies like Microsoft and Amazon, major venture capital firms, and ironically, other companies involved in the AI product chain.

Because private capital subsidizes loss on the promise of (supposed) future returns, AI companies offer their products at a fraction of their real cost. In the case of OpenAI, for example, a quadrupling of 2025 pricing would have been needed to just break even. AI providers attempted to adjust enterprise-level pricing upward this year, which led major customers from Uber to Microsoft to sharply cut back utilization. 

Founding a money-losing company is common in Silicon Valley. Indeed, the whole startup model is based on getting interest from venture capital and hoping you can grow into profitability. Nine out of ten Silicon Valley startups fail, often because the cost to acquire customers exceeds the revenue generated (a problem AI faces right now). However, when a $50 million valuation startup fails, it doesn’t have a wider impact on the U.S economy. If OpenAI and its competitors fail to meet targets, the implications for the broader economy will be staggering. 

The Rise of Data Centers

Part of what makes the current AI speculative bubble so much more damaging is the related proliferation of data centers. Data centers existed for years before the rise of generative AI—Google and Amazon used many of them for cloud storage—but the artificial demand spurred by discounted AI services has set off a nationwide boom. By the end of last year, U.S. data center construction spending was greater than office spending. A controversial Stratos data center scheduled to be built in Utah will cover 40,000 acres and consume more power than the entire state. Why are these mega-projects now so ubiquitous?

The operation and training of AI models requires staggering levels of computation. Generally speaking, an AI model is seen as “better” the more data is used in its training, but that also means that training the most advanced models requires orders of magnitude more energy, and that each question asked may require checking against hundreds of billions of parameters. This “brute force” model may not be the best paradigm for further advancement (China-based DeepSeek has achieved similar results to ChatGPT at a fraction of the cost) but at the moment, American-based AI companies are stuck on this “go big” paradigm.

Data centers are massive energy draws, due both to the direct costs of computation, and the cooling required in order to stop systems from overheating. A “hyperscale” facility can have power usage over 100 times that of a typical office building. Though data centers only utilize four to five percent of U.S. power consumption now, some estimates suggest the overall share will double or triple by 2030. In states with the most data center construction, there has been a massive increase in energy bills. For example, Virginia has seen a 267 percent increase in electricity rates over the last five years.

Some data centers (like the massive Stratos center in Utah) attempt to get around grid concerns by building their own power facilities. However, these are generally powered by fossil fuels like natural gas, driving up energy costs indirectly for consumers, along with greenhouse gas emissions. While the International Energy Agency (IEA) projects a higher percentage of data centers will be driven by renewables by 2035, the rapid growth in total energy usage means more gas and coal will be burned in absolute amounts due to the data center boom. Closer to home, data centers mean aging coal power plants are having their decommissioning pushed back or canceled.

In addition to energy concerns, data centers have massive draws on local water supplies to cool their hardware. A 2024 Department of Energy report found that data centers used 17 billion gallons of water in 2024. Water usage could be up to 73 billion gallons by 2028. One in five data centers are in water-stressed regions, while nearly half of all servers are powered by plants located in water-stressed regions. 

Data centers also reinforce and deepen existing environmental racism and classism. As a use which worsens air quality (most data centers have backup diesel generators, even if they’re on the grid) and creates noise pollution, data centers are often slotted into poor, industrial-zoned areas. A study by the Kapor Foundation of California’s 300+ data centers found that 82 percent locate in communities with already bad air quality, 65 percent in areas with groundwater under threat, and 79 percent in areas with the greatest amount of hazardous waste. 

While sited near working class communities (often communities of color) like the mills of old, data centers have a major difference: they provide few working-class jobs. Hyperscale facilities in rural areas may have up to 200 workers, but smaller-scale ones often only have one or two dozen, most of whom will be engineers and technicians who live as far as possible from the noxious byproducts. 

The strain that data centers put on other sectors of the economy is not limited to their immediate communities. A global shortage in memory chips began in late 2025, with data centers purchasing 70 percent of all chips. This has resulted in a spike in the cost of memory chips by 200 to 400 percent, and forced manufacturers of PCs, gaming consoles, and cellphones to raise prices by double digits this year, contributing to inflation.

Selling the Boss’s Fantasy

As AI’s visibility has grown, and its negative effects have become clearer, Americans have soured on the technology. A Pew survey from June found only 16 percent of Americans had a positive view of AI, with 40 percent holding a negative view. Unlike other recent technologies, attitudes towards AI are harsher among younger generations, in spite of young Americans' more extensive use of AI chatbots. 

All this can lead someone to ask; why are the companies spending tens of billions—why is the global economy spending trillions—to construct a technology that almost no one (other than scammers) appears to be positive about?

The reason: AI isn’t for you, it’s for your boss. 

Taken to its logical conclusion, the AI pitch is “every man a CEO.” Instead of doing work, you have a personal assistant who follows your commands to exacting precision. They never talk back, never tell you your idea is a bad one, and never ask for more money and better working conditions. 

Within businesses, the biggest users and proponents of AI are at the executive level. A June study cited by the Wall Street Journal found that 40 percent of non-managerial workers said AI saved them no work time, with another 27 percent saying it saved them less than two hours of labor a week. In contrast, 19 percent of c-suite executives (CEOs, CFOs, etc.) claimed that AI saved them more than 12 hours a week of labor, with another 57 percent claiming it saved them between 4 and 12 hours of work. 

Viewed through this lens, it also makes sense that established tech companies like Google and Amazon have invested so heavily in AI startups, and that the biggest breakthroughs in generative AI have been in coding. Tech CEOs are chasing a fantasy where physical capital can wholly replace human capital, where they can be a boss with no workers. 

The Future of Work If the AI CEOs Win

There has been a rise in so-called “AI doomers” who foretell an upcoming mass destruction of white-collar jobs, as almost all knowledge work is replaced by more advanced large-language models. While these voices might at first seem anti-AI, they are effectively doing PR work for the AI companies by focusing on how powerful and inevitable the technology is. Mass layoffs might be our nightmare, but they are the boss’s dream. 

The truth is much murkier. Individual studies have claimed that AI reduces task completion times for workers. However, in spite of this, very few firms that have implemented AI have seen cost savings, and there are no indicators that AI investment is increasing U.S. productivity. While a concerning number of layoffs are now due to AI (the single-largest reason cited this spring, with 21,490 in April alone) these losses are as of yet small enough to not have macroeconomic implications.

More troubling is that AI eliminates new entry-level jobs. For decades “learn to code” was a mantra told to high school students, yet according to the Federal Reserve Bank of New York, recent grads in Computer Science are now less able to find work than those with liberal arts degrees like English or History. 

The mid-term future of work and AI isn’t unemployment, it’s correcting substandard, AI-generated work. Even in coding, where the use case is clearest, AI often discovers an unelegent method via brute-force computation that needs refinement. In work involving knowledgeable research, the AI “hallucination” issue is well known (AI will, for example, make up citations in legal briefs for cases that don’t exist). When AI is used in the workplace, the worker’s job shifts from that of a front-line producer of knowledge and creative content to a quasi-supervisor who must check everything generated by machine. In the boss’s dream, this worker will have their load double, then triple, and further as jobs are combined, though this remains part of the boss’s fantasy for now.

It is worth comparing the current risks for white-collar workers to that of manufacturing in decades past. Plant automation led to technological unemployment, with the number of workers in manufacturing in the U.S. falling 35 percent from their all-time high in 1979, despite a doubling in total output. Yet just as notable as the employment decline was the deskilling of work. The more automated production became, the easier it was to split the remaining manual work into individual tasks which could be “rationalized,” leading to growth in lower-wage assemblers and a decline in high-skill craftsmen. A similar trend is likely with AI uptake, as firms attempt to establish processes which streamline prompt generation and double-checking results, making all of us even more like cogs in a machine. 

Creative Destruction

Much emphasis was placed a few years back on how AI training models scanned millions of images, books, and songs without paying anything to rights holders. While this violated the legal rights of millions of creative workers, the biggest legal actions have been taken by major media companies like Disney and Universal. Ultimately, focusing mostly on copyright plays into the hands of these major businesses, and does little to help independent creative voices.

The current threat for AI is not primarily cutting the royalties checks of creative workers for past work, but making future work harder to find. AI has bitten much harder into freelancers than direct employees. Areas varying from graphic design, to production of stock images, to copywriting, have seen work dry up, with much of what’s left “fixing” substandard AI first drafts. 

Independent musicians, self-published authors, and short-form video providers have also had their (meager) returns threatened with the rise of AI generated “slop.” For example, since the public availability of ChatGPT, the number of self-published books on Amazon has risen from 100,000 to 300,000, with the presumption that the majority of all new e-books are now AI. Streaming music company Spotify has promoted AI music on its platform, shuffling it into playlists to reduce royalty payments. And a study by video editing platform Kapwing found six out of ten videos shown to a new user on TikTok are clearly AI generated (the real number might be even higher). 

To be clear, there’s little threat of AI creating breakout stars; consumers have been negative regarding AI-generated creative works when unmasked. However, the cost in time and money for the scammers to create these slop accounts is so trivial that the low return for any one account isn’t a major concern. It does, however, make rising above the ocean of slop much, much more difficult for artists, musicians, and authors. This is an existential threat to creative workers, particularly young creatives from marginalized communities who lack connections, because the rise of social media has also tied most creatives to a handful of platforms, most of which are doing nothing to prevent this tide of slop.

An Economy for Humans, Not Bots

Silicon Valley has become its own strange self-referential creature, with ideologies percolating through it which get little coverage in the wider press. These go well beyond the typical rapacious capitalism of the ownership class, with a strange, quasi-religious millenarianism having overtaken many tech CEOs over the past two decades.

One such term which has gotten wider use in recent decades is “the singularity,” a concept from science fiction first popularized by futurist Ray Kurzweil in 2005. This is a belief that it is inevitable AI will become self-aware and advanced enough to bootstrap itself into superintelligence far beyond humanity. This philosophy comes in both a negative flavor (that the AI god will kill us all) and a positive one (that it will be benevolent and we can all live in a Star Trek-like postscarcity world). Some also include a fantasy that it will allow us to “upload” ourselves into digital immortality. In recent years, a variant of this called TESCREAL has married the concept with eugenics, arguing we should cheer our extinction as a “superior form of consciousness” inherits the world from us.

These seem ridiculous, but when powerful billionaires believe stupid things, we must take them seriously. Elon Musk was briefly the world’s first trillionaire due to the recent SpaceX IPO, which has its value inflated due to the supposed “potential” of Grok. Musk also believes that he’s living inside of an advanced computer simulation. Peter Thiel is worth $20 to $25 billion, effectively created J.D. Vance as a politician, and founded Palantir—an AI firm which facilitates government surveillance and partners with ICE. Last year, when asked by Ross Douthat if he wanted the human race to continue to exist, Thiel was unable to give an unequivocal yes. 

To be clear, there’s no evidence that generative AI comes any closer to self-awareness than any other advances of the last few decades. Due to the sheer scale of the training models, it has much greater capacity, but like other programs, it does nothing if not prompted. We are still far from a Terminator-style scenario where Skynet could rule over us all. And yet, the ahuman philosophies of tech CEOs shape the future nonetheless. They have ceased to see AI as a means towards an end (even just toward their own wealth generation) and increasingly see it as a mystical end unto itself.

The simplest response to this as workers is to reject the entire project in disgust. However, even if the current AI bubble is likely to pop, the advances made will still be there. Some of the worst people in the world will remain at the helm of large companies and continue with their quest to implement them, even if it takes another ten to twenty years. 

As the UE General Executive Board declared in May, we must demand “AI for Public Good, Not Private Profit,” with future development ensuring maximum human flourishing, applied within the workforce in a manner which eases workloads and shortens workweeks, rather than encouraging layoffs, job combinations, and speedup. It should also be powered by clean, renewable energy, built on sustainable infrastructure, work towards energy efficient computation, and only be further expanded as vital human needs are attended to. 

AI likely won’t make the sci-fi fantasy dream of the billionaires come true. But on the chance it can make some dreams come true, let us fight for our dreams instead.

UE Data and Technology Coordinator Christian Cmehil-Warn and Digital Organizers Samantha Cooney and Ebony Thomas contributed important insights to this report.