AI and the Economy: A Losing Bet for Working People
Tech billionaires and the Trump administration, with the apparent support of most of the capitalist class, are betting big on artificial intelligence (AI). In fact, AI investments have become the primary driver of US economic growth.
But this is a losing bet for us. The AI boom is not sustainable. And because it is delivering little of value, unbalancing our economy, intensifying our ecological crisis, and threatening the quality and responsiveness of our social institutions, the longer it goes on, the greater the harm done, and the more difficult will be the task of economic and social renewal.
The AI Boom
Many people believe that artificial intelligence is an ethereal technology, “living” in the clouds. In reality, AI systems are firmly rooted. They need electricity for training and operation, water for cooling, and racks of servers with chips built using hard-to-acquire minerals, all of which must be housed and accessed in gigantic data centers.
It is the massive spending on these data centers and their associated equipment and software that is most responsible for the US economy’s current growth. The economist Jason Furman estimated that US GDP growth in the first half of 2025 was almost entirely due to these AI-related investments. Excluding them, GDP growth, on an annualized basis, would have been minimal, only 0.1 percent. OECD researchers held a more pessimistic view, believing that without that spending the US would have been in outright recession.
Yearly AI-related capital spending by the biggest tech companies – Google, Amazon, Meta, and Microsoft – rose from $150-billion in 2022 to $360-billion in 2025. And they collectively plan to spend a substantially greater amount, $650-billion, in 2026. Bloomberg reports that the “companies’ estimates for [2026] are expected either near or to surpass their budgets for the past three years combined.” To put that spending in perspective: “The largest US-based automakers, construction equipment manufacturers, railroads, defense contractors, wireless carriers, parcel-delivery outfits, along with Exxon Mobil Corp., Intel Corp., Walmart Inc., and the spun-off progeny of General Electric – 21 companies – are projected to spend a combined $180-billion in 2026.”
These four tech companies are not the only ones investing in data centers. xAI, which was merged with SpaceX in 2026, completed a massive data center in 2025, with another still under construction. Oracle has recently become a major supplier of cloud services and, according to Larry Ellison, its CEO, the company aims to build “more cloud infrastructure data centers than all its infrastructure competitors combined.” In 2025, it signed a $300-billion contract with OpenAI to provide five years of computing services.
AI’s growth-supporting effects are also felt through another channel–the stock market. The November 2022 launch of the AI chatbot ChatGPT sparked an explosive growth in the value of a group of tech stocks known as the “Magnificent 7” – NVIDIA, Microsoft, Alphabet, Apple, Meta, Tesla, and Amazon. These stocks currently account for close to 40 percent of the value of the S&P 500 and are responsible for approximately 80 percent of the market’s overall rise in 2025. They generated an average return of 27.5 percent in 2025, compared with 7 percent for the rest of the S&P 500, producing a sizeable market capitalization-weighted gain of 17.5 percent.
In line with the class nature of the US economy, the wealthiest 10 percent of Americans, owners of close to 90 percent of the stock market, were the chief beneficiaries of this market showing. The wealth effect, where a rise in asset values encourages an increase in consumption, then played its part. The consumption share of the top 20 percent of earners soared over 2025, reaching 60 percent of total US consumption by the end of the year.
The celebration of AI as economic savior distracts from the fact that it is a very specific type of AI, known as generative artificial intelligence, that is largely responsible for the boom. Artificial intelligence technologies are typically divided into two main groups: machine learning and generative AI. Machine learning models use algorithms to identify patterns, make decisions, and improve their performance through experience. They do not generate new content. Generative AI models, which are trained on large data sets, can produce human-like text and respond to and manipulate audio and image inputs. The best known are ChatGPT (OpenAI), Gemini (Google), Claude (Anthropic), Grok (xAI), Copilot (Microsoft), and Llama (Meta).
The owners of these generative AI systems are locked in intense competition, with each hoping to secure market dominance and the resulting monopoly profits. But that is only the short-term goal. They also appear to believe that further development of their respective AI systems will produce a higher level Artificial General Intelligence (AGI), a superintelligence that will lead to, in the words of Mark Zuckerberg (Meta’s CEO), the “creation and discovery of new things that aren’t imaginable today.” Sam Altman (OpenAI’s CEO) believes that a soon-to-be created AGI will provide a solution to global warming, enable us to colonize space, and live forever with our minds uploaded to computers. Elon Musk (xAI’s CEO) believes that AI-powered robots will soon make work optional and money irrelevant.
The race to achieve market dominance, and eventually AGI, pushes these companies to continually offer new models that are said to be faster, more reliable, and more powerful. And it is this competitive upgrading that is propelling data-center construction and the economy’s growth. The reason is that existing data centers cannot easily be retrofitted to accommodate the needs of the new models, which require a greater number of larger server racks, each with more numerous energy hungry powerful chips, and more complex energy and cooling systems.
Trouble Ahead
Despite all the excitement and confident assertions that generative AI is a revolutionary technology capable of transforming the US economy for the better, the AI boom is likely near exhaustion. That is because these advanced AI systems suffer from serious and unescapable flaws and limitations that make them incapable of serving as a bridge to anything resembling AGI, and too unreliable and expensive (if priced to cover cost) to win widespread adoption by sufficient numbers of individuals or businesses.
Despite the use of the term ‘intelligence’, these systems do not think or reason. They operate by probabilistically selecting words or images based on pattern recognition developed from training on massive data sets built largely from material scrapped from the web. As a consequence, they periodically make nonsensical connections, leading them to produce factually inaccurate responses. This proclivity to “hallucinate” makes them untrustworthy, as the many lawyers, doctors, journalists, coders, students, and business owners who have relied on them have discovered. And because these systems are trained on largely unfiltered material from the web, they can also produce output that replicates the hateful and discriminatory material found there, making their use unacceptable in a variety of social, educational, and employment settings.
The companies developing these systems generally downplay the seriousness of these and other related problems, claiming they will be overcome with better and larger data sets, more sophisticated algorithms, and greater computational power. However, new human-created material in sufficient quantity for additional training has proven difficult to obtain because AI generated material now dominates the web. While some developers claim that this “synthetic data” is just as useful as human-generated material, studies have found that its use leads not just to a loss of accuracy but to a structural degradation of how reality is represented or, in the words of tech researchers, “model collapse.” As for the problem of hallucinations, even OpenAI employed researchers have concluded that “large language models will always produce hallucinations due to fundamental mathematical constraints that cannot be solved through better engineering.”
Not surprisingly then, businesses employing AI have found productivity gains difficult to realize. An MIT Media Lab study, reported on by Forbes, concluded that “AI pilot failure is officially the norm – 95 percent of corporate AI initiatives show zero return.” A survey of more than 1,000 enterprises across North America and Europe found that 42 percent had abandoned most of their AI initiatives in 2025, up from 17 percent in 2024.
The upshot? None of the major generative AI systems are profitable or on the road to profitability. OpenAI’s ChatGPT is the most widely used system. Yet, as tech commentator Ed Zitron points out, the company lost $5-billion in 2024 and will likely lose upwards of $8-billion in 2025.
An article in The Conversation, a nonprofit news organization, offers some insights into why:
“Free [generative AI] services, and cheap subscription services like ChatGPT and Gemini, cost a lot of money to run. OpenAI CEO Sam Altman has been candid about how much money his firm spends, once quipping that every time users say ‘please’ or ‘thank you’ to ChatGPT, it costs the firm millions. Exactly how much OpenAI loses per chat is anyone’s guess, but Altman has also said even paid pro accounts lose money because of the high computing costs that come with each query.”
Some analysts estimate OpenAI might run out of cash by mid-2027 without new funding. OpenAI itself forecasts a loss of $14-billion in 2026 and expects to continue to make huge losses totaling $44-billion until 2029.
Things are not much better for Meta, Amazon, Microsoft, Google, and Tesla, which have their own AI systems and also build and operate their own data centers. Collectively, these firms spent more than $560-billion on AI-related capital expenditures over the years 2023-2025, all to generate combined earnings, not profits, of only $35-billion.
Despite model shortcomings and profit challenges, the major AI players remain determined to press ahead. But with planned spending far outstripping revenue, they can do so only if they are able to obtain the required funds from debt and venture capital markets. And the amounts needed are sizeable. For example, OpenAI has signed deals committing it to spend some $1.4-trillion dollars over the next five years, including $500-billion to purchase chips from NVIDIA, $300-billion for computing services from Oracle, $22-billion for computing services from CoreWeave, and an unknown amount to Broadcom to help it develop and deploy racks of its own designed chips. Oracle, for its part, is planning to raise some $50-billion in 2026 through a combination of debt and equity sales to finance its construction activity.
Even the largest data center builders are finding it necessary to tap debt markets. As Bloomberg explains:
“More than $3-trillion. That’s the staggering price tag to build the data centers needed to prepare for the artificial intelligence boom. Not even the world’s biggest technology companies – not Amazon.com, not Microsoft or Meta Platforms – are prepared to foot the bill with only their own cash.
So where will the money come from? Debt markets.
Which ones? All of them.
Blue-chip bonds, junk debt, private credit and complex asset-backed pools of loans. ‘The numbers are like nothing any of us who have been in this business for 25 years have seen,’ says Matt McQueen, who oversees global credit, securitized products, and municipal banking and markets at Bank of America Corp. ‘You have to turn over all avenues to make this work’.”
For the moment, it appears that lenders and investors are willing to back the AI bet. But with AI developers unable to produce a dependable, cost-effective, and broadly useful product, company revenue projections are bound to disappoint, and there will come a time when lenders and investors will simply refuse to throw good money after bad. When that moment arrives, the AI boom is finished.
OpenAI may be the most vulnerable to such a financial squeeze. As noted above, it has signed a number of agreements to purchase services from other companies. However, at the rate it is burning through money, it may not be long before its financing needs outgrow what lenders and investors will find acceptable. If they pull back, OpenAI will be forced to retrench, slashing employment and investment, with negative consequences for its stock price, development program, and the companies counting on its business. Oracle is one of those companies. It borrowed heavily to finance its data center construction counting on OpenAI for most of its future revenue. Without that revenue, Oracle’s own financial situation will quickly deteriorate. Both OpenAI and Oracle are major customers of NVIDIA, so their difficulties will affect its bottom line. And on it goes.
A growing number of investment analysts are starting to take this danger seriously. NPR reports that “Morgan Stanley analysts estimate that Big Tech companies will dish out about $3-trillion on AI infrastructure through 2028, with their own cash flows covering only half of that,” leading one analyst to say, ‘If the market for artificial intelligence were even to steady in its growth, pretty quickly we will have over-built capacity, and the debt will be worthless, and the financial institutions will lose money.’”
In early February 2026, these concerns led, as Bloomberg describes:
“to a series of punishing [stock market] selloffs, wiping more than $1-trillion from the market values of big tech companies… [This] marks a major break from the sentiment of the last few years, when speculation that AI would set off a transformative productivity boom kept pushing stock prices higher… But the pile of money the tech giants are throwing at AI is getting so big that there’s increasing skepticism about whether it can continue.”
In fact, there are signs that even tech players are growing worried. OpenAI was encouraged to pursue its spending plans because NVIDIA had agreed to make a $100-billion investment in the company. However, only months later, NVIDIA walked back that commitment, with Jensen Huang, the company’s CEO, claiming that the deal was nonbinding. Bloomberg cites reports that Huang has privately voiced concerns about OpenAI’s business strategy and standing relative to its competitors.
The problem, of course, is not a matter of competition. Rather it is that these generative AI systems cannot deliver what they promise. Surveys may show significant business and public use, but paying customers are few and far between. While OpenAI claims more than 500 million weekly users, only 15.5 million are paying subscribers, which as Zitron notes, “is an absolutely putrid conversion rate.” And this still beats Google, whose latest Gemini model is now getting rave reviews. As Zitron explains, “When you look at the actual business lines, the revenues are pathetic, with Google’s Gemini Enterprise only having eight million paying subscribers… which could mean everything from ‘paying $17 to $30 a month for a Google workspace with Gemini account’ to ‘has used the Gemini Enterprise API’.”
The Way Forward
The AI boom will end. But it would be a mistake for us to just wait for that to happen. It could take years, and every year it continues, we pay a price. The massive investment in generative AI and its data center infrastructure is drawing funds away from areas of greater social importance, leaving our economy ever more unbalanced and incapable of responding to our needs.
The hyperscale data centers are themselves enormously harmful. They disrupt communities, displace needed agricultural land, siphon off tax revenue needed to fund social services, push up electricity prices, stress local energy systems and water resources, and contribute to global warming. Encouragingly, community groups and environmental organizations are finding new and effective ways to resist the construction of new data centers and, in some cases, block the operation of existing ones.
AI developers are also aggressively working to embed generative AI systems into as many aspects of our lives as fast as possible. They appear to recognize the growing popular distrust and disapproval of their systems and no longer count on their “organic adoption” by consumers, social institutions, or government agencies. Rather, as the writer Matt Seybold so well puts it, “They have moved on to a new dream of forced adoption mandated by government and managerial coercion.” We can already see signs of their efforts in our schools, health institutions, newsrooms, film studios, and social media, although resistance, especially from unions, is growing.
The end of the AI boom doesn’t mean that tech companies will abandon their efforts to profit from required use of generative AI systems or that our economy will automatically generate a new center of economic vitality. That means we need to deepen our own organizing efforts, with a focus on building a more coordinated and stronger fight for a technology policy and economy that serves majority interests. •
This article first published on the Reports from the Economic Front website.
