The race for AI dominance is fueled by a mountain of debt from the data center rush, leading experts to warn of turbulence ahead, even if it’s not a repeat of the dot-com crash.
- An ‘Air Pocket’ Looms: Bank of America predicts the AI market is heading for an “air pocket”—a period of turbulence where aggressive capital spending outpaces monetization—rather than a full-blown 2000-style tech bubble.
- Debt-Fueled Expansion: Hyperscalers like Google, Amazon, and Meta are issuing debt at unprecedented rates to fund the AI infrastructure boom, with the top five companies raising $121 billion this year alone, four times their recent annual average.
- Skepticism as a Safeguard: Despite the frenzy, analysts believe there is enough “healthy skepticism” from investors and markets, evidenced by stock selloffs after high spending announcements, to prevent the kind of speculative bubble seen in the dotcom era.
The artificial intelligence revolution is in full swing, but while the narrative is one of limitless potential, a more cautious story is unfolding in the financial backrooms of Big Tech. According to Bank of America Global Research, we are not on the verge of another tech bubble bursting like it did in 2000. However, that doesn’t mean investors should expect a smooth ride. The massive, debt-fueled capital expenditures on data centers are creating significant risk, leading experts to forecast an impending “air pocket” for the AI sector.
Savita Subramanian, BofA’s head of U.S. equity and quantitative strategy, has been a leading voice in this analysis. “Is this 2000? Are we in a bubble? No,” Subramanian stated, but she quickly followed with a crucial caveat: “Will AI continue unfettered in leadership? Also, no.” Her argument is that while today’s AI boom is supported by real earnings growth and less extreme speculation than the dotcom era, the financial foundation is becoming precarious. “On AI, in our view, investors should get ready for an air pocket,” she wrote in a recent note. “Monetization is to be determined, and power is the bottleneck and will take a while to build out. So for now, investors are buying the dream.” This cautious stance has led BofA to forecast a modest 4% upside for the S&P 500 in 2026, a stark contrast to more bullish predictions from the likes of Deutsche Bank.
Interestingly, this sense of caution is seen as a positive sign. Jean Boivin, head of the BlackRock Investment Institute, echoed Subramanian’s sentiment, suggesting that the widespread discussion about a potential bubble is precisely what might prevent one. “We don’t think the bubble framing is that useful at this stage for investors,” Boivin said. “There is so much talk about the potential of the bubble … People are conscious of the risk. It’s when there’s no discussion of that that we should be more worried.” This “healthy skepticism” acts as a natural check on market hype. A prime example occurred after Meta’s October earnings report, when CEO Mark Zuckerberg‘s admission of a $2 billion increase in capital expenditure guidance triggered a 9% drop in the company’s shares.
The core of the “air pocket” wariness lies in the staggering sums being spent on AI infrastructure. According to the Dell’Oro Group, spending on data centers by hyperscalers surged 53% year-over-year to an astonishing $134 billion in just the first quarter of this year. Tech giants are in an arms race, with Google recently pledging $40 billion to expand its AI compute infrastructure in Texas. The problem, as Subramanian points out, is that “capex funded by operating cash flow is running out.”
This is where the danger truly lies. To keep up, the five main hyperscalers—Amazon, Google, Meta, Microsoft, and Oracle—are increasingly funding their operations through debt. BofA analyst Yuri Seliger noted that these companies issued $121 billion in debt this year, a figure four times their average annual debt issuance over the past five years. He anticipates they will raise another $100 million in debt in 2026. The supply of AI infrastructure is exploding, having increased by over 1,000% from 2024 to 2025, but it’s being built on borrowed money.
This raises a critical question: can these massive investments ever turn a profit? IBM CEO Arvind Krishna is skeptical. By his “back-of-the-napkin math,” the path to profitability is incredibly steep. “It’s my view that there’s no way you’re going to get a return on that, because $8 trillion of capex means you need roughly $800 billion of profit just to pay for the interest,” Krishna explained. Compounding the problem is the rapid pace of technological advancement, which could render today’s cutting-edge infrastructure obsolete in a short time. “You’ve got to use it all in five years because at that point, you’ve got to throw it away and refill it.” This makes the hyperscalers’ big bets a high-stakes gamble, suggesting that while the AI dream is alive and well, the financial reality may soon bring it down to earth with a jolt.