When Al emerged some years ago, commentators said that Al and robots were going to take over the world, that we were all going to be unemployed, and so on. But at the time, this was more conjecture than anything. Without investment, AI as we know it would never have become a reality.
[Originally published on Marxist.com]
Since 2023, investments by tech companies have made this possible. Hundreds of billions have been ploughed into new data centers. The massive, highly speculative investment boom then led to fears that they were investing in a bubble. We wrote an article in the autumn comparing it to the railway bubble of the 1840s, and we weren’t the only ones.
Is there a bubble?
The essential problem was that it was highly doubtful that these companies were going to start making money from this. The jury is still out on whether the providers of the Large Language Models (LLM) like OpenAI and Anthropic will be able to pay for all this investment. Anthropic, at least, appears to be getting there, and currently expects to turn a profit by 2028.
But that doesn’t mean there is no bubble. The current valuations of AI stocks not only expect AI to pay for itself, but to allow for an exponential growth in profits. NVIDIA’s stock market valuation implies a quadrupling of its profits from an already very high level. Samsung is the same, and Taiwan Semiconductor Manufacturing Company (TSMC) is not far behind. Broadcom is making $30 billion per year, but its valuation assumes a future profit of $200 billion or so.
The fear of missing out on constantly rising stock prices is guiding investors, which is typical of a bubble. But AI is also a new market that will quickly become monopolized, and everyone wants to get in there before it does. That’s why investors are throwing the kitchen sink at it.
A bubble means companies are betting big, but not all of these bets will pay off. Some will go bust and a few monopolies will emerge as winners.
For now, the investments have been largely funded by the hoards of cash accumulated by the big tech companies known as “hyperscalers” (providers of cloud storage infrastructure). This limits the risk of a bust spreading to the credit and banking system. However, as the cash piles of Google, Microsoft, and others start to run out, they are starting to borrow more and more. Oracle, which has the least cash of the “Big 5” hyperscalers, had a wobble a couple of months ago when investors started to worry about their level of debt.
Deployment of AI
The more lasting impact will not be from the bubble, but from the potential of the technology itself. The railway bubble in the 1840s didn’t mean the end of the railway, and the busting of the dot-com bubble didn’t mean the end of the internet.
Leaving aside some of the exaggerated claims the companies make—such as those made about Anthropic’s Mythos model—there is a clear practical use for AI, particularly when it runs as an agent on your computer.
This will be problematic in and of itself. It is already beginning to have an impact on employment. 7% of large UK businesses have already used AI to cut staff. The most immediate effect is a fall in new job postings, which are down 38% in high-exposure jobs like coders and translators. This obviously affects young people just entering the labor market more than any other group.
A recent BBC interview with former British Prime Minister Rishi Sunak, who is now a lobbyist for both Microsoft and Anthropic, discusses this question:
[Sunak] said company bosses were privately acknowledging to him that recruitment of young people is flattening because of the technology… [He said] it is becoming tougher for young people to get jobs in service sectors such as law, accountancy and the creative industries.
Speaking about company bosses, Sunak added:
They’re talking about this concept that they think they can continue to grow their businesses without having to significantly increase employment because they’re starting to see how they can deploy AI.
According to the UK government, about 70% of UK workers are in occupations with tasks that could be performed or enhanced by AI, a higher share than in the US and many other advanced economies, reflecting the UK’s service‑heavy economy.
In the US, companies cited AI as the reason for 54,000 layoffs in 2025. This is five percent of the total. But it is accelerating. In the first three months of this year, the figure was 13%. And last month, Al became the leading cause for layoffs, with 25% of the total.
Some economists say in defense of AI that it will create jobs as well as destroy them. Naturally, for the bourgeois economists, the market will solve all problems. They cite past leaps in productivity when capitalism was relatively healthy. But in a period like the present, technological innovation will only intensify the crisis. In the past few months, Amazon sacked 30,000 workers, particularly in middle management. Now they want to rehire 10,000 AI specialists, a third of what they just got rid of. Tellingly, they are going to use AI to vet and interview the applicants.
Programmers have so far been most affected, as their employers are more adept at new technology of this sort. But there is also the future impact on other professions, like law. Anthropic is working on a “legal” version of its model, and it likely will do to the legal profession what its “code” version has done to programming.
If Anthropic or one of its competitors were to succeed, then lawyers would be checking the work of AI, not writing contracts themselves. They’ll probably prepare their cases based on research carried out by AI, rather than by junior staff. Law schools are discussing reducing class sizes, worrying that the need for lawyers in the future is going to be reduced.
Harvard Medical School has discovered that AI is also very good at medical diagnosis. Some studies even found that it could spot cancer before a specialist could. This kind of thing is what it would excel at: recognizing patterns.
Nevertheless, the adoption of AI is taking its time. It is a new tool, and it will take time for workers and companies to work out how to use it. It will require reorganizing workplaces and workers learning how to become supervisors of these AI agents, in the same way as many industrial workers have become supervisors of machines and robots.
This means there’s a lag. Companies aren’t going to simply change how they work overnight. But it is going to have massive implications.
Anger against the system
There is a prospect of large-scale unemployment, starting among university graduates. Those who keep their jobs will face deskilling, effectively being reduced to the minders of AI agents.
AI threatens to inflame the anger that workers already feel against the system and the bosses. It wouldn’t be the first time in history that automation and deskilling have led to a layer of the petty bourgeoisie or working class becoming radicalized.
The so-called hyperscalers are facing increasing resistance to the building of data centers: people don’t like their energy bills getting higher or scarce water supplies being diverted to cool data centers.
Sam Altman, the CEO of OpenAI, has had his home attacked several times. One of the suspects had even penned an “anti-AI manifesto.”
In the hands of the working class, AI and its complement, robotics—which is highly developed in China—have the potential to liberate humanity from the drudgery of work. It is the basis on which a communist society can be built. But in the hands of the capitalists, this technology threatens unemployment, and for those in work, higher intensity, and longer working days. Already, there are reports of workers having to work longer hours as a result of AI.
This is nothing new. It has happened time and time again. Marx explained this process in Capital. He wrote that new machinery, far from leading to an improvement in working conditions, leads to an extension of the working day for workers, who have to work harder to ensure the capitalists make back the cost of their investment. And so Meta is laying off 8,000 workers and leaving another 6,000 roles unfilled, in order to placate shareholders worried about the cost of AI investment.
Still, for all their investments, it isn’t enough. Unless they come up with a way to radically improve the efficiency of these models, they are really struggling to generate enough computing power to run them. Anthropic, for example, has been complaining that OpenAI is hogging all the processing power, leaving little for them to run their own models on.
For Al to be widely adopted, it will require far more investment. Compared to previous investment booms, this boom has been fairly mild. For example, the railway construction booms in the US and the UK consumed something like 8–10% of the economy. The AI investment boom, by comparison, is only consuming 1–2% of the US economy, and much less elsewhere. This fits a pattern. Western economies have plenty of money for speculation—in stocks, Bitcoin, etc.—but relatively little to actually develop the economy.
Meanwhile, with a crisis of overproduction dominating the world economy, the capitalists, not for the first time in history, are scratching their heads trying to work out: “if I sack all my workers, then who will buy my products?”
The truth is, if AI fails to achieve what it promises, it will cause problems; if it succeeds, it will cause even more problems. And this will add to the multitude of problems already facing the world economy: the Iran War, debt levels, protectionism, etc.