Artificial Intelligence – Are we in the bubble?
Getty
We must be equally cautious in predicting the imminent burst of skeptical AI bubbles of exaggerated and exaggerated surrounding artificial intelligence now.
There are concerns about the signs. “Epic Seven” stocks (Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia, Tesla) account for more than a third of the S&P 500, with recent growth being driven by AI stories. This level of focus makes investors worried. At the peak of the 2000 dot-com bubble, top technology stocks in the late 1990s (Cisco, Dell, Intel, Lucent, Microsoft) accounted for 15% of the index. Such concentrations increase the risk. The similarities don't stop there.
AI Bubble and the Era
A massive telecom infrastructure build-out led by the age of e-commerce. The world needed an internet pipe to allow for high speed connections. This led to an overly optimistic development of fiber optic networks, resulting in catastrophic bankruptcy when demand did not materialize in the short term.
Today, major AI companies are investing hundreds of billions of dollars in new data centers. Total capital expenditure in the region is being debated in trillions of dollars, once only related to the GDP of the great powers. Does history repeat itself and cause an imminent collapse?
Meanwhile, the connection boom and investment quarter-century ago has made the world we live in today possible. They created opportunities for value creation beyond infrastructure at the application level, and accelerated the transformation of the information technology industry through the transition to the cloud. Some may argue that data centers are the new utility needed to provide on-demand information services to an increasingly connected world.
Will the demand for AI be realized?
Much of the current attention focuses on the consumer space. Openai's ChatGPT website received over 5 billion visits in July. But that's not the whole story.
The true economic impact is measured by the recruitment of consumers and businesses. The National Bureau of Economic Research began publishing surveys on generative AI adoption about a year ago. As of late 2024, about 40% of the US population reported using generated AI, while 23% reported using at least one job before voting. Comparing the level of adoption since the launch of the first product, workplace generation AI takes off faster than personal computers and the internet. This highlights the possibilities of AI as what economists call a general technology that has a deep, broad impact on the economy.
However, the challenges remain. A group of MIT researchers surveyed more than 300 public AI initiatives, over 50 companies, and hundreds of senior leaders from January to June 2025, and concluded that 95% were not making a profit from their investment. We were also able to identify three factors that would make the remaining 5% successful. Companies that have successfully made AI initiatives, in contrast to the Central Institute, choose tools to buy instead of building, run within business units, and integrate with existing business workflows.
Although it is rare to achieve returns related to business transformation, adoption is high, with 90% seriously considering purchasing AI solutions. This is a familiar pattern in the adoption of enterprise technology. It is captured by what consultants call the hype cycle, tracking innovative technologies from the market entrances, until when companies are likely to benefit from them, and technology becomes mainstream.
Gartner's hype cycle
Getty
Bank of America, the second largest bank in the United States, has a $4 billion budget for new technologies such as AI, and is an example of a pattern identified in MIT research. Integrating AI and business workflows. The bank has developed tools to help bankers prepare client meetings and retrieve information from multiple systems. Previously, junior bankers would carry out this process over multiple hours or days.
How far can current AI models be taken?
As AI usage increases, so does its ultimate potential and debate about whether current development models are sustainable.
Much of the progress so far has been made behind the large-scale language models that benefit from scale. Scale means producing better results with more computing power and more data. AI pioneer Richard Sutton observed in 2019 that common ways of leveraging calculation power are superior to those who rely on human ingenuity and complex heuristics (in the creation of a “bitter lesson” for humanity). He recently criticised the industry's stanchions of scaling and called for corrections to continuous learning agents.
Gary Marcus, one of the most vocal critics of artificial intelligence hype, commented on the mixed reviews received by Openai's latest ChatGPT-5 release. He reflected the sentiment that the scaling-based development model is not a path forward, but a position he has sponsored for decades.
The deep skepticism of these scientists about current advances represents a term of technical attention. The hype conditions created by investors and large AI labs can lead to disappointment. However, both believe in the ultimate possibilities of AI, suggesting that an alternative approach is needed. These may require more, rather than less investment in research and development.
Is there an AI bubble?
Even Openai CEO Sam Altman, who sparked the AI boom, should pause when he warns that the market may be overheating. He and other investors mention the rising valuations, too much money chasing unproven business models, and the risks of building infrastructure faster than demand is justified. Like the MIT report, they worry that much of their capital expenditure will flow into projects that are unlikely to produce immediate results. The concern is not about AI's long-term commitments, but about expanding expectations setting a rapid revision stage.
Binary thinking that shakes between hype and the fear of the AI bubble could limit more nuanced analysis. The long-term potential of AI remains important, but markets rarely move in a straight line. Modifications can slow momentum in the short term, while reinforcing the need for discipline. The next stage will depend on advances in research, improved model quality, and directing corporate investments into measurable economic value.
