The AI Revolution Might Be Running Out of Steam
September 19, 2025
The AI Revolution Might Be Running Out of Steam
Despite massive investment and grand promises, AI companies are struggling to deliver returns. The bubble may be deflating, but like the dot-com crash, the aftermath could consolidate power in the hands of tech giants.
In late August, a Silicon Valley news item caught the attention of technology watchers across the country and globe: Meta was, with immediate effect, freezing hiring for its artificial intelligence operation.
The hiring freeze marked a sudden shift in strategy from the Mark Zuckerberg—led company, which had until several weeks ago been turning heads with a talent acquisition strategy that reportedly included offers of $100 million signing bonuses and even larger compensation packages.
Some in the industry took the news of Meta’s hiring freeze as a sign of nothing more than Meta’s struggle to build a competitive AI division — arguing that the fact the company was forced to offer such lavish contracts in the first place was proof that it was struggling to successfully recruit top talent as it struggled with its reputation as a particularly difficult place to work.
But others took the news as a broader sign that, despite the massive incursion of AI into the daily lives of millions in recent years, the AI revolution may be hitting a snag.
“There’s a hype bubble,” Max Read, who covers AI in his newsletter, told Jacobin. “We can all agree that for the last three-and-a-half years, say, there’s just been an absurd amount of overpromising and underdelivering about the capabilities of these large language models, promises about how they’re going to transform the world, promises about how everything’s going to change. I think we can all agree that hype bubble is already on its way to deflation.”
In addition to Meta’s hiring freeze, there are other signs that AI, at least in its present stage of development, is failing to meet those lofty expectations. A study produced by MIT found that 95 percent of generative AI projects are failing to boost revenue, while the rate of AI adoption by larger companies appears to be slowing.
“The amount of money it costs to produce a new model is growing really, really quickly and the gains are getting smaller. ChatGPT-5 is really not better than ChatGPT-4.5. It’s almost no different in terms of [its] capabilities,” Aaron Benanav, a professor at Cornell University, told Jacobin. “Microsoft’s CEO and others have said productivity is supposed to be growing quickly from these technologies, and it would have to: they’re spending so much money that the economic returns would have to be enormous — and it’s just not materializing.”
However, Matthew Ellis, a professor at Portland State University, cautioned that even if there is an AI bubble of sorts, that doesn’t mean AI will necessarily go away when the bubble inevitably pops. Ellis said that if the AI bubble is akin to the dot-com bubble, as OpenAI’s Sam Altman has suggested it might be, that doesn’t mean AI is on the precipice of disappearing or becoming in any way less threatening to workers.
“That bubble burst, and we still have websites, we still have the internet, we still have all of that stuff,” Ellis said. “What happened is what always happens in capital, [which] is that you need to burn off a lot of excess capital and get it out of the market, and then the people who are left over, who are powerful enough, are able to suck up the remainder, integrate it into their control, and they are a much larger holder of the capital in that particular industry — leading, potentially, to big monopolies.”
Absent any sort of public policy intervention, Read predicts that the names of the biggest beneficiaries of the AI bubble bursting will be familiar: Google, Microsoft, and possibly, in the end, Meta too.
“These are the already incumbent huge tech companies who have in-house models. . . In some sense, power is just consolidated in what are already the most powerful companies in the Valley — who already have very close relationships with the national security state, who already have very close relationships with the government, regardless of who is in power,” Read said. “This isn’t really a changing-of-the-guard moment.”
It is instead the less established companies that may struggle to survive the AI bubble — and in any case, the success or failure of individual companies is not necessarily predictive of the future impact of the technology.
“There will probably be cases of this or that AI company that spectacularly fail in embarrassing ways, but the technology is not going to go away because the purpose of the technology is to automate labor away — and that’s what everyone wants to do,” Ellis said.
Read said that while there may be elements of the AI bubble that parallel the dot-com bubble experience, the better comparison for AI’s current position may be the cryptocurrency bubble. That episode ended with the FTX implosion that sent prices tumbling and resulted in the company’s founder, Sam Bankman-Fried, receiving a lengthy prison sentence.
“It felt like this huge, resounding victory, because all of a sudden you weren’t being confronted every time you opened the newspaper or opened Twitter or whatever with a bunch of assholes talking their book about this stuff. But the fact is that Bitcoin, crypto — none of that stuff has gone away,” Read said. “It’s just sort of fallen out of the public eye in the same way. In fact, it’s sort of more powerful, more insidious, worse than ever.”
Read pointed to the amount of money cryptocurrency interests spent during the 2024 election, in which candidates boosted by a pair of crypto-backed super PACs won fifty-three out of fifty-eight races. One of those PACs, Defending American Jobs, spent some $40 million to defeat Senator Sherrod Brown of Ohio — an outspoken critic of the industry who was then the leader of the Senate Banking Committee. The industry is reportedly amassing an even more imposing war chest for the next electoral cycle and flexing its muscle on Capitol Hill, while the price of Bitcoin reached an all-time high in August.
Government Doubles Down on AI Dominance
Another factor that points to the possibility of AI’s staying power is the extent to which the technology is already being embedded in government operations and considered a matter of state concern.
The Trump administration has broadcast very clearly that it wants the United States to be an AI power, calling the country’s “unquestioned and unchallenged global technological dominance” a “national security imperative” in the White House’s recent AI Action Plan. Benanav suggested that this is partially an ideological approach in addition to a practical one, given that the United States has already “lost the race for green technologies” and is thus especially inclined to trumpet the importance of its AI sector.
Individual states have also already begun investing in AI, both by partnering with leading AI companies and lavishing tax breaks on corporations building the physical infrastructure needed to support the technology. Here too, Zuckerberg is leading the way in the scope of his ambition: in June, Meta announced plans to build the largest AI data center on the planet in rural northeastern Louisiana — having been lured to the state with a twenty-year sales tax exemption and the promise of three new on-site power plants financed and built by Entergy, the state’s public utility, at a cost of more than $3 billion.
In total, the tax incentives passed to lure Meta’s investment could cost the state more than $10 million annually through the year 2059. Louisiana is a striking example of the lengths to which states have been willing to go to entice corporations to build data centers, but it is by no means alone: a CNBC investigation found that states have bargained away at least $6 billion in just the last five years to attract data centers.
The environmental impact of these data centers — and of artificial intelligence use generally — appears significant, even amid the many other environmental degradations of the day. Because of the density of power required to run generative AI, a single query on ChatGPT requires ten times the energy of a Google search. The intensity with which AI data centers consume both energy and water is perhaps best grasped at the level of individual centers: a planned AI data center in Wyoming, for instance, will use more electricity than every home in the state combined, while a single Google data center in Council Bluffs, Iowa, used 1.3 billion gallons of water from the local water supply in 2023. There is already evidence to suggest that AI data centers are driving up electricity prices while draining freshwater reserves.
“It’s all part and parcel of the same thing: we’re kind of hurtling down this path, this AI arms race in the midst of ecological collapse, that seems governed almost entirely by these mystical pronouncements from Silicon Valley combined with investor appetite for a new potential source of growth in the midst of global deindustrialization choking off other avenues for investment,” Read said. “All of these things are interlinked, and all of them are bad.”
There is also, Benanav pointed out, a psychological component to the current discourse around AI’s purportedly world-altering ability. Even if the existence of a bubble does not mean the technology is on the precipice of vanishing, workers can still organize to exert control over AI and ensure its development does not rob them of their livelihoods or natural environment.
“It’s always important to know that generally, at least in the last forty to fifty years, computer technologies have not gotten rid of jobs, they’ve just changed jobs,” Benanav said. “And so all the talk about getting rid of jobs is mostly about trying to demoralize people about fighting to improve conditions. Knowing that about the research, that the jobs change more than they go away, it’s really important to fight.”
Search
RECENT PRESS RELEASES
Related Post
