The AI Boom: Hype, Power and the Coming Reckoning
Photo by Luke Jones
It is safe to say that since OpenAI first released ChatGPT in October 2022, AI has been the most widely covered and discussed topic in the U.S. By January 2023, ChatGPT had become the fastest-growing consumer software app in history, gaining over 100 million users in its first two months. Currently, its website has over 800 million weekly users. Since ChatGPT’s emergence, the question of AI and its potential and pitfalls has dominated discourse.
Indeed, daily, one is bound to run into numerous headlines in both the business and tech press. Just to take a quick example, on September 18th, Time magazine put forward two stories. One was titled ‘AI is Learning to Predict the Future- and Beating Humans at It’ (that story dealt with an AI system finishing in the top 10 of the Summer Cup, a forecasting contest); the other read ‘AI is Scheming and Stopping It Won’t Be Easy, OpenAI Study Finds’ (this one touched on the fact that all of today’s best AI systems- Google’s Gemini, Anthropic’s Claude Opus, can engage in scheming- meaning they can pretend to do what their human developers want, while actually pursuing different objectives).
One can do this all day. In August, Wired featured a headline, ‘Nuclear Experts Say Mixing AI and Nuclear Weapons is Inevitable.’ More recently, Wired featured a story that Anthropic recently partnered with the U.S. government to ensure that Claude wouldn’t spill nuclear secrets and help another entity build a nuke. Explicit in this is the geopolitical sphere. This past July, the New York Times Business page declared ‘The Global AI Divide’ that has wealthier regions building out robust data centers while poorer regions are left behind. And there have been U.S. efforts to freeze China out of the cutting-edge chip market, not to mention the apprehension about open-source models from Chinese companies such as DeepSeek. Then there have been countless stories featuring the coming revolutionary effect AI will have on everything from drug development to medical diagnosis (‘a doctor in every pocket’) to food production to films to love lives to weather forecasting to, well, everything.
Go a bit deeper and, depending on who one talks to, AI will usher everything from a global dystopia where almost nobody will be able to earn a living since AI and robots will make people unnecessary to employers to global socialism- since we can’t allow the former scenario to come to pass, we’ll need to usher in the latter and AI can be a better agent for a planned economy than human bureaucrats. In that case, AI would indeed have a major role in ushering in universal abundance and perhaps very long lifespans.
There is also the infamous idea, featured in The Terminator movies, that superintelligent AI will replace us mere Homo Sapiens, for one reason or another, with higher forms of life. Not surprising perhaps one can find some who are actively cheering for that prospect.
For a harrowing primer on that, there is Eliezer Yudkowsky and Nick Scoares’ book If Anyone Builds It, Everyone Dies: Why Superhuman AI Would Kill Us. The main point here is that no one knows exactly how AI works and therefore no one can completely predict its perceived interest. Known as the alignment problem, we can’t be sure superintelligence will have its interests completely aligned with ours. Yudkowski and Scoares write:
“The most fundamental fact about current AIs is that they
are grown, not crafted. It is not like how other software gets
made- indeed it is closer to how a human gets made, at least in
important ways. Namely, engineers understand the process that
results in an AI, but do not much understand what goes on inside
the AI minds they manage to create.”
It may not be that an AI system comes to want to annihilate us out of maliciousness, either. The authors cite the example of building a skyscraper on a patch of land that has an ant hill. We’re not trying to deliberately kill ants; we’re just trying to build a skyscraper and preserving the ants in that location isn’t important enough to be considered. Such could be our relationship with superintelligence and, as the authors note, such intelligence once built can’t be simply unplugged once it exists.
None of that appears to be having any effect on capital pouring into AI, but on that note a different narrative has crept up in the discourse. In August, the Financial Times asked, ‘Is AI hitting a wall.’ Given the general disappointment with OpenAI’s GPT-5, FT wrote: ‘Following hundreds of billions of dollars of investment in generative AI and the computing infrastructure that powers it, the question suddenly sweeping Silicon Valley is: what if this is as good as it gets?’ The piece then added the magic word: bubble.
It was right around that time when an MIT study came out showing that 95 percent of the generative AI implementations in enterprises have had zero return on their investment (that investment equals $30 to $40 billion). A month earlier, the think tank Model Evaluation & Threat put out a study involving a randomly selected group of experienced software developers to perform coding tasks with or without AI tools. Giving that coding is a task that current AI models are supposed to have mastered, all involved in the study expected productivity gains for those using the AI. Instead, the study found that those in the AI group completed their tasks 20 percent slower than those working without it. A McKinsey & Company report from March found that 71 percent of companies reported using generative AI, and more than 80 percent of them reported the AI had no ‘tangible impact’ on earnings.
Amara’s Law, named for American futurist Roy Amara, states that we tend to overstate the effect of technology in the short term and underestimate the effect in the long run. The idea being that there is a period of time between the development of new technologies and enterprises adjusting their operations to incorporate their use.
Economist Eric Brynjolsson, co-author of The Second Machine Age and Race Against the Machine, posits that every new technology experiences a ‘productivity J-curve’, meaning at first, enterprises struggle to deploy it, causing productivity to fall. Eventually, when they learn how to integrate it, productivity booms. Indeed, it took decades for technologies such as tractors and computers to have a significant impact on productivity. Even electricity itself, which became available in 1880s, didn’t begin to produce big productivity gains until the 1910s in Henry Ford’s factories.
Just to throw around some numbers, this year, large tech firms in the U.S. will spend nearly $400 billion on AI infrastructure. By the end of 2028, analysts reckon the amount spent on data centers worldwide will be more than $3 trillion. According to recent data from Pitchbook, AI startups received 53 percent of all global venture capital dollars invested- in the U.S., that percentage jumps to 64 percent.
With other sectors of the economy showing signs of sluggishness, likely in part due to Trump’s tariffs, for instance, roughly 78,000 manufacturing jobs have been lost this year, AI is holding the fort. In a column for the Financial Times titled ‘America is now one big bet on AI’, Ruchir Sharma writes that AI companies account for 80 percent of the gains in U.S. stocks so far in 2025. In fact, more than a fifth of the entire S&P 500 market cap is now just three companies- Microsoft, Apple, and Nvidia- two of which are largely bets on AI. Nvidia recently became the first company to reach a $5 trillion market cap, while Microsoft recently hit $4 trillion.
Another word from past bubbles that has been creeping into the ether is debt. On September 30th, the Wall Street Journal wrote: ‘In the initial years of the artificial-intelligence boom, comparisons to the dot-com bubble didn’t make much sense. Three years in, growing levels of debt are making them ring a little truer…a crop of highly leveraged companies is ushering an era that could change the complexion of the boom.’
Oracle, for example, which pledged a $300 billion investment in AI infrastructure with OpenAI now owes over $111 billion in debt. Quick FactSet reports that interest-bearing debt of the 1300 largest tech companies in the world has quadrupled over the past decade and now stands at roughly $1.35 trillion. According to analysts at Morgan Stanley, debt used to fund data centers could exceed $1 trillion by 2028.
Much of this capital is flowing in something of a circle. In other words, one company pays money to another company as part of a transaction, then the other company turns around and buys the first company’s products and services (and without the first transaction, that other company may not be able to make the purchase). Nvidia agrees to invest $100 billion in OpenAI to fund data centers and Open AI commits to filling said centers with purchased Nvidia chips; Oracle has a $300 billion deal with OpenAI for a data center buildout and is spending billions on Nvidia chips; Nvidia is planning on investing $2 billion in Elon Musk’s xAI and has agreed to buy $6.3 billion worth of cloud serving from CoreWeave, the leading independent operator of AI data centers in the U.S.; Meta agreed to buy $14.3 billion from CoreWeave in September (CoreWeave has told financial analysts for every billion in computing power it plans to sell it must borrow $2.85 billion); CoreWeave put $350 million in OpenAI…etc.
Needless to say, companies like OpenAI and Anthropic are nowhere near profitable. By one estimate, Meta, Amazon, Microsoft, Google, and Tesla will by the end of the year have spent $560 billion on AI-related capital expenditures since the beginning of 2024 and brought back just $35 billion in AI-related revenue. It is tempting to say that AI is too big to fail. But what is the damage if the bubble does pop? In an essay titled What Kind of Bubble is AI, tech writer Corey Doctorow writes ‘Tech bubbles come in two varieties: The ones that leave something behind, and the ones that leave nothing behind. Sometimes, it can be hard to guess what kind of bubble you’re living through until it pops and you find out the hard way.’ Past manias, such as the 19th-century British railroad bubble at least left useful infrastructure. It is hard to see what other uses there can be for all these data centers.
And this leads to another narrative that’s floating in the ether. In their book The AI Con: How to Fight Big Tech’s Hype and Create the Future We Want, Emily Benner and Alex Hanna argue that we’re essentially being sold a bill of goods. What is being labeled ‘AI’ is in fact a marketing term for a bundle of different technologies. The Large-Language Models (LLMs) so in vogue are just ‘synthetic text-extruding machines’ that ‘out and out plagiarize their inputs’ and site nonexistent sources. And plagiarism seems like quite a fair charge. It is true that all art and technology build on what came before, but Van Gogh didn’t copy and paste the Japanese prints that influenced him into his paintings. LLMs aren’t adding any new inputs, they just directly synthesize the creative inputs of others. Much of the media, under the thrall of tech CEOs and boosters, are reporting every small success but nothing about AI’s many failures. In this case, both the boosters and the doomers miss the point. AI may not take your job away, but it may well make it shittier. Making some people redundant, allowing greater exploitation of others, AI is simply another move for greater profits squeezed out of the masses.
Tech boosters and corporate henchmen will often shout ‘Luddite’ at this kind of talk. This actually ends up not making the intended point. If the contemporary idea of a Luddite is someone slamming their new gadget against a wall in frustration or proudly grinding their own coffee beans while jamming to a vinyl record, the original Luddites should not be understood as an anti-tech movement, but as a social protest rooted in early nineteenth-century England. The machines they smashed in protest, the stocking frame, were not new pieces of technology. By the time the Luddites burst onto the scene in 1811, the stocking frame had existed for more than two centuries. In fact, the Luddites can’t even claim originality since machine-breaking had a long history in English protests.
The Luddites’ problem was not with technology per se, but how it was applied- the way it was used to create unemployment, impoverish skilled workers, increase production while stagnating wages. One can see the shadow of this in such actions as the 2023 strike by the Writers Guild of America and Screen Actors Guild against the Hollywood studios. It seems a safe bet we’ll be seeing more such resistance in the coming years. Journalists from the news site Politico recently won an arbitration ruling regarding AI against the company. The arbitrator ruled that Politico officially violated the collective bargaining agreement by failing to provide notice, human oversight, or an opportunity for the workers to bargain over the use of AI in the newsroom.
Finally, one shouldn’t overlook the anti-democratic ethos behind AI boosting. Offshoring problem-solving to AI, even if it can be done, means just concentrating power into the hands of a select few who own the AI. That appears to be precisely the point: no more messy democratic planning and consensus building. For example, take former Alphabet CEO Eric Schmitt regarding climate change and AI: ‘My own opinion is that we’re not going to hit the climate goals anyway because we are not organized to do it and, yes, the needs in this area [AI] will be a problem. But I’d rather bet on AI solving the problem than constraining it.’ OpenAI CEO Sam Altman has been proclaiming that we need AI to solve the big problems facing us, such as global warming to colonizing space.
Data centers are the reason the big tech companies have openly blown past their previously stated green targets, not to mention a key factor in increased electricity use in the U.S. (after being flat for decades) and rising electric bills.
But AI won’t relieve us from the hard work of democratic politics and there is no reason why AI development should be exempt from such work. Yudkowsky and Scoares call for full-on public mobilization to keep the worst from happening. Even if the ultimate doomer vision doesn’t convince many, there is still plenty to discuss. Few would dispute the positive, specific contribution AI can make, like helping to solve part of the protein-folding problem (that AlphaFold, developed by Alphabet subsidiary DeepMind, in 2023) and assisting in developing needed antibiotics and other medical advances like improved diagnosis. But anything can be subject to debate and regulation, even public ownership. Demands for transparency in how systems are designed and for whose benefit should be spewing from elected officials. And there are those CO2 emissions to consider.
Corey Doctorow recently called AI ‘the asbestos we are shoveling into the walls of our society and our descendants will be digging it out for generations…AI isn’t going to wake up, become superintelligent and turn you into paperclips- but rich people with AI investor psychosis are almost certainly going to make you much, much poorer.’ Obviously, only democratic engagement and collective responsibility could avoid that bleak vision. Such calls always seem like long shots but, really, is there anything else?
The post The AI Boom: Hype, Power and the Coming Reckoning appeared first on CounterPunch.org.