Since the rise in long-term bond yields (interest rates) which began in the middle of last month, when the yield on the benchmark 10-year US Treasury passed 5 percent, the issue has been when this will start to hit the stock market and particularly the AI boom.
That question is starting to be answered as interest-rate sensitive AI stocks undergo a significant fall amid rising concerns about how many heavily indebted companies will even be able to survive in the new higher interest rate environment.
The most vulnerable area, at least to this point, is the so-called “neocloud” sector, comprising companies that have borrowed heavily to invest in chips, usually from Nvidia, which are then rented to firms seeking to train AI models.
Last week, the initial public offering (IPO) of the Australian neocloud firm Firmus collapsed when investors concluded its business model did not stack up.
It has been followed this week by what could prove to be an existential crisis for the US neocloud firm CoreWeave.
As a report in the Wall Street Journal put it: “The Nvidia-backed compute provider has become a weather vane for one of the riskiest—and more highly leveraged corners of the AI trade. This week, it endured a tornado.”
At one point, the company’s shares were down 28 percent for the week, accompanied by a steep selloff in its bonds. CoreWeave bonds fell to below 80 cents on the dollar, regarded as being in distress, and the yield rose close to 15 percent.
The spread on CoreWeave’s five-year credit default swaps, a derivative by which investors seek to insure themselves against a default, rose above 11 percentage points. This meant investors had to pay $11 per $100 of CoreWeave debt to try to protect themselves.
The main problem is debt, as it is with many other AI companies, and CoreWeave has a lot of it. According to FactSet, it has around $35 billion of debt on its balance sheet and is expected to spend $80 billion over the next three and a half years. But its growing problems are only the sharpest expression of a developing crisis.
According to the Journal, investors are viewing higher-for-longer interest rates as an effective tax on marginal borrowers and, in the words of one analyst, “the marginal borrower in the AI trade is anyone who is not a high-quality hyperscaler”—companies such as Amazon, Google, Meta and Microsoft.
The tech consultancy firm Gartner has forecast that 80 percent of neocloud firms will “cease independent operations” by 2030 because of the vast gulf, running into trillions of dollars, between their capital spending and the returns to be generated from the development of AI.
According to Gartner, data center owners would need to generate revenues of $8.5 trillion by the end of 2030 to cover their costs and generate profits on their investments. But it predicted that consumer and corporate spending would generate “no more than $3.6 trillion.”
“A continued AI infrastructure investment trajectory could leave more than $4 trillion in asset value exposed to writedown risk through 2032,” Gartner said.
For some, the crunch may come much sooner than that. According to Ludovic Subran, chief investment officer at the German financial firm Allianz, whose remarks were cited in the Financial Times: “Data center bonds are having their Darwin moment, where the weakest credits struggle to survive.”
The immediate spark for the turmoil of this week has been the recognition that higher interest rates are here to stay and this will cause major problems for the most highly leveraged. But the longer-term issue is whether the demand for the services provided by AI will be enough to even cover the cost of the buildout, let alone generate a sufficient profit.
As reported in the FT, a recent estimate by the Brookings Institution is that “spending on the AI services by all those data centers would need to account for an astounding 9.2 percent of US GDP by 2032 for their operators to make a decent economic return.”
As with all major technological developments, there is also the threat that competitive pressures from the production of cheaper and equally efficient systems will in effect devalue investments already made.
“If you start to have competitive pressure from China, then margin projections for US tech companies will be revisited,” Charles de Boissezon of Société Générale told the FT.
The turbulence has extended into other areas of the AI ecosystem. Global equity markets had their worst day since July on Wednesday on the back of data which showed that inflation in the US was not coming down, making a further rise in interest rates by the Federal Reserve more likely.
There was also concern because of the revelation that OpenAI’s annualized revenue, an important indicator of AI demand, was $20 billion less than the company had initially indicated.
Anthropic has reported that its third quarter revenue has risen by nearly 50 percent to $16.8 billion. But this was slower than the pace it had indicated at the start of the year and below the expectations of some investors.
Uncertainties in the stock market were exacerbated by the blunt statement of the Broadcom chief executive that some tech investors “have no clue about what they are investing in” and that at times the market resembled a “casino.”
This characteristic is nowhere more clearly manifested than in South Korea, one of the main global centers for the manufacture of chips.
On Wednesday, the semiconductor-heavy Kospi index fell by 8 percent, which triggered a trading halt, one of many which have taken place this year, as shares in the major firms SK Hynix and Samsung Electronics fell by almost 11 percent and 9.3 percent respectively.
The South Korean market has been very much at the center of AI speculation. The Kospi index rose from 2,300 points in April 2025 to more than 9,300 points in June this year at the time of the SpaceX IPO, which is coming to be regarded as marking the peak of the AI boom.
The share price of SK Hynix has fallen by almost 58 percent from its peak, while Samsung’s is down by 44 percent.
There is clearly concern among some of the major corporations about the market gyrations.
Responding to the plunge in CoreWeave shares, Nvidia chief executive Jensen Huang told Bloomberg that the company’s strength was “improving every day” and “I’m optimistic.”
His intervention was motivated not least by the fact that his company has become the largest in the world by market value at around $5.3 trillion on the back of the near-universal use of its chips. If companies like CoreWeave go down, or even experience significant problems, there will be major consequences.
In a sign of the extreme volatility which prevails, CoreWeave’s shares rose by 6 percent after a 28 percent plunge following Huang’s comments.
Such jawboning may bring results on a day-to-day basis. But it cannot change economic and financial fundamentals, above all the upward pressure being exerted on long-term interest rates and their impact on the highly leveraged AI boom.
These issues were highlighted at this week’s annual meeting of the International Monetary Fund (IMF) in Bangkok, Thailand this week where the focus was on the need to bring down government debt, now rapidly heading towards 100 percent of global GDP.
There was a significant moment in the press conference on the Fiscal Monitor report when a questioner noted that, for the past number of years, the IMF theme has been that governments needed to bring down debt, and so what was different now.
Rodrigo Valdés, the director of the Fiscal Affairs Department, replied that the situation was “urgent,” and governments had to act now before the markets forced them to do so.
The report, he said, had “dialed up” the urgency of the message because, in contrast to the recent past when interest rates were at historic lows, the long-term interest rate was now moving above the rate of growth of GDP and debt service costs were rising.
He did not elaborate further but the implications are becoming increasingly apparent: the interest rise threatens not only to burst the highly leveraged AI bubble but also to set off a crisis in the entire financial system, which has become increasingly dependent on debt.