Concerns about an AI bubble are escalating
CFOTO/Sipa USA/Alamy
The staggering sum of hundreds of billions dedicated to AI has seemingly inflated a precarious global financial bubble, risking the stability of companies and investors who may find themselves burdened with unmanageable debt, supported by the paltry earnings currently generated by AI services. What consequences could this hold for the technology fueling the current financial frenzy?
In recent times, alarms about a possible AI bubble have been sounded by the International Monetary Fund, the Bank of England, the CEO of the largest US bank, and even OpenAI’s Sam Altman. “What we’re witnessing is not merely a stock market phenomenon; it’s an investment and public policy bubble,” states David Edgerton from King’s College London.
The circular nature of certain transactions among major AI corporations is also sparking concern. For instance, Nvidia, the supplier of GPU chips driving the AI surge, has pledged up to $100 billion for OpenAI to establish a new data center packed with Nvidia’s own products. In a reciprocal agreement, OpenAI may acquire a 10 percent stake in its rival chip manufacturer, AMD.
The threat of a bursting AI bubble is further exacerbated by the sheer scale of investment; data centers alone account for at least $400 billion in annual expenditures, as reported by Morgan Stanley Wealth Management. While the US GDP grew by 3.8 percent in the last quarter, economist Jason Furman from Harvard University estimates that excluding data centers, the GDP would have increased by only 0.1 percent during the first half of the year.
Carl-Benedikt Frey from the University of Oxford explains that such aggressive investment activity is not atypical in technological advancements, arguing it would actually be unusual if the economy could efficiently invest in new infrastructure at precisely the appropriate rate to match demand. “Overbuilding is a common occurrence: we saw it during the railroad expansion and again in the dot-com bubble,” he remarks.
We must consider whether the repercussions of an AI bubble bust would solely impact the companies directly involved or whether they might trigger broader economic fallout. Frey notes that many costly data centers are constructed “off balance sheet,” meaning they are established as new companies funded by external investors or banks that hold the associated risks and potential returns.
This raises uncertainties surrounding who might bear the brunt of financial risks. A data center might have its financing sourced from numerous tech billionaires or conventional banks, and significant losses could precipitate a banking crisis capable of rippling through the overall economy. “While this doesn’t immediately signal an impending financial crisis, the lack of transparency indicates potential risks,” Frey adds.
Benjamin Arold at the University of Cambridge suggests that the disparity between profits and company valuations reveals how out of sync public perception is with the actual financial performance of businesses. He flags these metrics for today’s tech firms as troubling.
“The last time we saw such low ratios was 25 years ago, coinciding with the dot-com bubble,” highlights Arold. “It’s conceivable that the outcome could be positive, but I wouldn’t place any bets on it.”
James Poskett from the University of Warwick forecasts a downturn in the AI sector that could lead to the demise of many businesses, although he stresses that this doesn’t mean the technology itself will vanish. “It’s crucial not to conflate this with the assumption that the technology is flawed or will be abandoned,” Poskett warns. “We may experience an AI decline, but that does not portend the end of AI itself.”
Similar to how the consolidation of railway companies post-bust resulted in a robust rail system, and the collapse of tech firms during the dot-com bust established a foundation of extensive fiber optic networks, we can expect to retain valuable technological advancements, according to Poskett.
For consumers, the deflation of the AI bubble could mean fewer options, potentially increased costs for access, and a slower rate of innovations. It may compel us to confront the reality that employing an enormously expensive tool like GPT-5 to draft an email is akin to using a sledgehammer for minor tasks, with the true costs previously obscured by the zealous AI race. “At present, we’re enjoying a lot of perks, but eventually, these companies must generate profits,” Poskett concludes.
Topics:
