Wall Street’s focus is shifting from how quickly artificial intelligence will expand to whether the companies building its infrastructure can generate enough profit to pay for it. After years of strong stock market gains driven by AI enthusiasm, a sell-off in technology shares is exposing the financial pressures behind the boom, with rising borrowing costs and enormous capital spending making investors more selective about which businesses they are prepared to fund.
Few companies illustrate the change more clearly than CoreWeave, the US cloud computing firm specialising in AI and backed by Nvidia. Its shares have fallen 36% over the past month, while the yield demanded by investors holding its bonds has surged to 16%, a level associated with growing concerns about a company’s ability to meet its financial obligations. CoreWeave has liabilities of $50 billion and plans to continue raising money to finance its expansion, intensifying scrutiny of whether future revenues will justify the scale of its investment.
The company’s latest financial figures illustrate the challenge. At the end of the second quarter, CoreWeave reported a $104 billion order backlog and revenue of $2.575 billion, but its net loss widened to $626 million from $290 million in the same period a year earlier. The company spent $640 million on interest payments alone. The figures have drawn attention to the vulnerability of so-called neocloud providers, which buy processors, install them in data centres and sell computing capacity to customers, but operate with more fragile financial structures than established cloud giants such as Amazon, Microsoft and Alphabet.
The pressure is spreading across the sector. Semiconductor companies have fallen 19% over the past 12 trading sessions, putting the index on course for its worst run since 2008. The index is down 27% from its June peak. Meanwhile, Australia’s Firmus and Britain’s Nscale have postponed plans to list on the stock market as investor sentiment deteriorates. South Korea’s Elice, which specialises in data-centre infrastructure, plunged 32% on its stock market debut this week before recovering its offer price two days later. The volatility has highlighted investors’ growing reluctance to finance AI-related ventures without clearer evidence of their capacity to generate sustainable returns.
The latest correction comes amid a significant change in financial conditions. Earlier market setbacks, including the arrival of DeepSeek in January 2025 and the announcement of tariffs two months later, drove semiconductor shares down 34.8%. Investors subsequently regained confidence in the technology. This time, however, central banks are maintaining restrictive monetary policies to contain inflation, bond yields are at their highest levels in two decades, and expectations of further debt issuance are increasing the cost of financing expansion.
Goldman Sachs has identified a shift in investors’ appetite for risk. “Our risk appetite indicator has fallen sharply over the past three weeks and moved into negative territory for the first time since March 2026,” the bank said in a report. Unlike in previous corrections, when investors often used falling prices to buy technology shares more cheaply, market participants are becoming more reluctant to increase their exposure. Money is moving towards companies with stronger balance sheets and more predictable businesses, with Goldman noting that high-quality stocks in Europe and the United States are benefiting from resilient margins and financial strength.
Yet evidence of continuing demand complicates the argument that the AI boom is losing its economic foundations. Taiwan Semiconductor Manufacturing Company (TSMC) exceeded market expectations and raised its outlook, while ASML, the Dutch manufacturer of chipmaking equipment, recorded record sales of €11.662 billion. ASML said anticipated demand over the next two years exceeded available capacity. The concern is therefore less about whether customers want AI computing power than whether the companies supplying it can finance the infrastructure required to meet that demand.
The scale of the investment is substantial. According to Bloomberg data, the market has absorbed $450 billion in AI-related debt this year alone. Meta increased its projected spending for 2026 to between $135 billion and $145 billion this week, prompting a sell-off in its shares. Paddy Flood, co-head of global equity research at Schroders, said the industry had increasingly turned to external financing as investment needs expanded. Goldman Sachs estimates that AI-related companies account for a quarter of all dollar-denominated corporate debt issuance this year by investment-grade borrowers.
The prospect of a bubble has returned to the centre of the debate. At the Future of Global Markets conference in New York this week, Ken Griffin, founder of Citadel, warned that markets were “almost certainly” experiencing another bubble. Larry Fink, chairman of BlackRock, offered a different assessment, arguing that demand remained strong and that the principal challenge was deploying infrastructure quickly enough to serve it.
Some analysts believe the recent sell-off has already priced in significant risks. Bank of America argues that leading semiconductor companies, including Nvidia, Broadcom and Micron, are trading at valuations more characteristic of a market that has already suffered a collapse than one still experiencing expansion. Goldman Sachs identifies a different danger: not necessarily excessive share prices, but expectations of future profits that may prove unsustainable. “The market could be developing an earnings bubble,” its analysts warned, drawing a parallel with the banking sector before the 2008 financial crisis, when rapid profit growth ultimately proved impossible to maintain.
Barclays, however, cautions against treating CoreWeave’s financing difficulties as evidence that the entire AI investment model is failing. The bank says the $1.5 billion to $2 billion in CoreWeave subordinated bonds is being treated as a barometer of an AI investment cycle worth more than $1 trillion. Those bonds may accurately reflect the risks attached to that particular debt, its analysts argue, without demonstrating that the wider industry’s business model is broken or that financing for the sector is running out.
Enrico Chinello, head of next-generation technology research at Julius Baer, similarly attributes recent financing pressures primarily to a limited group of highly leveraged neocloud companies. “We see some potential financing risks in certain areas of the ecosystem, but these are specific problems affecting particular cases and do not reveal weaknesses in demand or in the industry’s fundamentals, which remain solid,” he said.
