Wall Street strategists are issuing a warning: the decline in AI cloud service providers鈥?share prices will spill over into other corners of the technology trade and could drag the U.S. stock market all the way back to its June lows.
At the center of the storm is CoreWeave, Inc. (ASX: CRWV). After its bonds fell to distressed levels this month, the company鈥檚 share price has dropped 41% since October 6 鈥?the day major U.S. stock indices hit record highs. Over the same period, other 鈥渘eoclouds鈥?鈥?companies that rent out computing power to firms building and running AI models 鈥?have also posted double-digit declines: Iren Ltd. (ASX: IREN) fell 40%, and Nebius Group NV (ASX: NBIS) dropped 36%.
The bonds cracked first, not the stocks
The real trouble started in the bond market. On October 14, the secondary-market yield on a CoreWeave bond maturing in 2032 with a 9.625% coupon briefly touched 15%. Venture capitalist Chamath Palihapitiya posted the yield chart on social media, and Musk replied with just one word: 鈥淵ikes.鈥?During the same period, CoreWeave鈥檚 five-year credit default swap (CDS) spread broke through 1,000 basis points 鈥?for comparison, SoftBank Group sits at about 424 basis points and Oracle at about 288 basis points; the implied probability of default from the five-year CDS rose to roughly 60%.
Behind those numbers is the shape of the business: CoreWeave buys large quantities of GPUs, builds data centers, and then rents out computing power to AI companies by the hour, with expansion funded almost entirely by debt. Its total debt rose from about $15 billion a year earlier to roughly $52 billion at the end of June; second-quarter interest expense was $640 million, up from $267 million a year earlier; quarterly revenue more than doubled to $2.58 billion, but it still posted a net loss of $626 million. One analyst put it bluntly: if the company borrows at a cost of 9% while the best return these investments can generate is only 5%, then every step of expansion is destroying enterprise value.
Wall Street has given it a name 鈥?the 鈥渃anary in the coal mine鈥?of the AI trade. The reason is not its size, but that it is one of the few samples on the AI capital expenditure chain with high-frequency public credit quotes: bonds price cash flow, stocks price narrative, so the alarm always sounds first in the bond market.
It is not just the secondary market valve that is being tightened. In mid-October, South Korean AI computing supplier Elice plunged 32% on its first day of trading in Seoul; Australia鈥檚 Firmus Grid abandoned its public listing plans; and the UK鈥檚 Nscale ran into resistance over its own IPO target valuation. CoreWeave CEO Mike Intrator also acknowledged in an internal memo that 鈥渨eakness and delays surrounding competitors鈥?IPOs and debt financing鈥?are seeping into CoreWeave鈥檚 narrative, while calling the market鈥檚 performance 鈥渋rrational.鈥?
From 鈥渕icro鈥?to 鈥渕acro鈥?/b>
Investors鈥?worries go beyond how much has already been borrowed 鈥?they also include how much more will need to be borrowed after earnings season. Last week, credit risk indicators for some major AI-spending technology companies were pushed to new highs. On that brutal Thursday, October 22, the Nasdaq 100 fell nearly 3%, the S&P 500 posted its worst single-day performance since late July, and the Philadelphia Semiconductor Index plunged 6.72% in one day. Before that, the semiconductor index had already fallen for eight consecutive trading sessions 鈥?its longest losing streak since the 2008 financial crisis.
鈥淚t is now shifting from a micro event to a macro event,鈥?said Brian Reynolds, chief market strategist at Reynolds Strategy LLC, referring to the selloff in neocloud stocks. 鈥淚n other words, it is not just hitting neoclouds 鈥?it is hitting the entire credit market, and then the stock market.鈥?Reynolds expects further selling that could push the S&P 500 back to its early June low 鈥?which would mean a further 4.7% decline from the index鈥檚 Friday close. Based on the S&P 500 closing at 7,623.68 on October 23 and its June 10 low of 7,265.99, that decline is about 4.7%, matching his estimate. He also noted that the S&P 500 had already broken below its 50-day moving average on Thursday and could fall all the way back to the June low, which is exactly where he would be willing to act. 鈥淚 want to stay defensive for the next few weeks,鈥?Reynolds said. 鈥淚 really want to buy that low around 7,200.鈥?
How do financing costs transmit to chips?
Neocloud companies rely heavily on debt issuance to support spending. And as U.S. Treasury yields rise and inflation remains stubborn, financing costs are climbing. Luis Alvarado, co-head of global fixed income strategy at Wells Fargo Investment Institute, put it more bluntly: the wave of easy financing experienced in the first nine months of this year is beginning to fade, and 鈥渆very new dollar, especially every marginal new dollar, will now face stricter scrutiny,鈥?with investors carefully assessing how much additional yield compensation issuers are willing to offer.
CoreWeave is also looking for cheaper money 鈥?according to reports, the company is seeking investment-grade ratings for more chip-backed debt so it can tap insurance companies鈥?capital pools. Co-founder Brannin McBee called insurance-side capital 鈥渋nfinite鈥?and stressed that he does not view this type of debt as 鈥渢errifying debt.鈥?
Ohsung Kwon, chief equity strategist at Wells Fargo, pointed out that if neoclouds cut future capital expenditure plans to reassure investors, that would become a major headwind for the semiconductor sector 鈥?whose revenue comes directly from data center developers paying huge sums for chips. 鈥淚f they start to see risks on the financing side, and if the market resists their capex far more than it does today, if they cut capex next year, that is a risk for semiconductors,鈥?Kwon said in an interview. That concern is already weighing on the Philadelphia Semiconductor Index: it has fallen 15% since October.
More than tech: a 鈥榬eality check鈥?/b>
It is not just tech stocks. Several Wall Street strategists warn that the collapse in neoclouds is giving the broader market a 鈥渞eality check鈥?at a time when investors are already anxious about an AI bubble. 鈥淚f the AI bubble is going to burst, it will likely begin with a credit event: neoclouds and high-risk AI debt defaulting, triggering a chain of valuation shrinkage, spending cuts, and contraction 鈥?with both valuation multiples and growth shrinking,鈥?wrote Tom Essaye, founder of Sevens Report, in a note last week. He added that he is still closely watching stress signals in the credit market. A Barclays strategist team led by Venu Krishna noted that neocloud stocks are currently reacting 鈥渁symmetrically鈥?to credit pressure 鈥?spreads widening triggers a disproportionate decline, while the relief from spread tightening is very limited, highlighting market doubts about the durability of the neocloud business model.鈥?They wrote that in a report last week. However, Krishna also said that although he is looking for 鈥渟igns of contagion鈥?in the stock market, he has not yet seen evidence that the problem has spread beyond tech and growth stocks. Investors have been rotating from high-flying tech stocks into defensive sectors, which looks more like uncertainty about the AI trade than indiscriminate selling of all assets at once. But once the largest-weighted sector is sold, the index gets dragged down. Also on Thursday, the S&P 500 fell more than 1%, even though more stocks rose than fell that day.
鈥楾he tail wagging the dog鈥?or 鈥榗rying wolf鈥?/b>
The market remains divided on how to characterize this alarm. One view holds that CoreWeave is limited in size and highly leveraged, a secondary link in the industrial chain, and that it is now disrupting a trillion-dollar AI trade in reverse 鈥?a classic case of 鈥渢he tail wagging the dog,鈥?in which a real but localized credit event is being overextrapolated into a reversal of the entire AI trade. Another view emphasizes that pressure has already spread from the bond market to the stock market, with giants such as Nvidia and Meta also caught up in the selloff. Earnings season has been revealing: Meta beat revenue expectations but still fell sharply on massive capital expenditure and cash flow pressure; Super Micro Computer disclosed strong preliminary revenue data but still failed to dispel doubts; Tesla cut its capital expenditure plans, triggering concerns about insufficient AI investment. Even Western Digital鈥檚 flawless earnings report failed to win investor favor. In short, invest too much and you fall; invest less than expected and you also fall 鈥?the AI narrative that was once chased by capital has suddenly met a picky grader.
Gartner offers a longer-term warning: more than $3.5 trillion in AI infrastructure investment faces impairment risk, and by 2030 more than 80% of neocloud service providers may struggle to operate independently. There are still several hurdles this week: the Federal Reserve鈥檚 October 27-28 policy meeting 鈥?where the market expects no change 鈥?as well as earnings from Microsoft, Google, and Amazon, all of which will provide new evidence on technology spending and profitability. Reynolds鈥?choice is to wait on the sidelines for a few weeks 鈥?waiting for that 7,200 level.
Comments