The AI spending boom is beginning to mirror the classic pattern of historic capital busts, Bank of America has warned, where heavy investment and cutthroat competition erode returns even as the underlying technology advances. In a note published October 16, strategist Sebastian Raedler outlined a familiar sequence: high early margins, aggressive investment, excess capacity and weakening pricing power.
The scale of the outlay is staggering. Consensus expectations for hyperscalers’ capital expenditure over the next 12 months have tripled from 330 billion dollars in early 2025 to approximately 1 trillion dollars, equivalent to roughly 3 per cent of U.S. GDP. That surpasses the 2.2 per cent of GDP spent during the 1870s railroad boom and the 1.1 per cent during the 1990s technology boom, according to the note.
Signs the AI spending boom is entering a precarious phase
Raedler’s warning is not a prediction that a bust is inevitable, but a risk assessment that the AI infrastructure buildout may be reaching a dangerous stage. Several warning signs are already flashing. Average AI token prices, the units of text that language models process and bill for, have reportedly fallen about 50 per cent since summer, partly because of competition from Chinese open-weight models.
Usage and revenue are also diverging. Data cited by Raedler from the corporate spending platform Ramp shows U.S. business AI spending easing slightly since August despite sharply rising token volumes, suggesting weaker monetisation. Meanwhile, AI-related stocks are under pressure: the Philadelphia Semiconductor Index has declined for seven consecutive sessions and sits around 13 per cent below its recent high. Financing risk is mounting too, with Reuters reporting on October 7 that major technology companies are increasingly turning to debt and equity markets to fund AI and cloud expansion, moving beyond the sector’s traditional reliance on internally generated cash. The AI spending boom is also changing how the buildout is paid for: leverage adds a new danger, because if returns disappoint, the debt raised to fund the AI spending boom will still have to be serviced.
Why the AI spending boom echoes railways and the internet
“This monetization disappointment in the face of intense competition is precisely the failure mode seen in past capital cycles,” Raedler wrote. “Most major capital cycles of the past have ended in disappointment, even though the underlying technology often lived up to expectations,” he added, pointing to railways, telecommunications and the internet. All created enormous economic value, but investors lost money when capital flooded into projects whose returns failed to justify their costs.
The AI spending boom fits the template uncomfortably well, Raedler argues: strong initial returns encouraged companies to invest aggressively, and the resulting capacity is now putting pressure on prices. Earlier this month, BofA’s chief investment strategist Michael Hartnett made a related comparison, writing that AI had become “the biggest bubble since railroads” and noting that the AI spending boom has already taken hyperscale cloud capex toward 3.5 to 4 per cent of U.S. GDP next year, still below the roughly 5 per cent reached during the railroad construction boom, he wrote.
What investors should watch in the AI spending boom
The critical question is whether the industry’s long-term promise can generate sufficient returns to justify the spending. For now, the tension is visible in plain sight: billions continue to flow into data centres, advanced chips and computing capacity, making AI one of the biggest bets in U.S. industrial history, even as pricing power deteriorates.
Raedler cautions that a technology can transform industries and deliver substantial economic benefits without generating attractive returns for every company investing in it. Whether the AI spending boom converts infrastructure into earnings will be judged on token prices, capacity utilisation and cloud margins in the quarters ahead. Investors will be watching token prices, capacity utilisation and the profitability of the largest cloud providers for evidence that the AI spending boom can convert infrastructure into earnings, or whether it is condemned to repeat the disappointments of the capital cycles that came before it. For a market that has priced in years of uninterrupted growth, the answer matters enormously.