NEW YORK — Global stock markets are sliding on fears that debt-fueled investment in artificial intelligence infrastructure could buckle under high interest rates. The alarm has spread across markets as the credit standing of CoreWeave collapsed. The company leases data centers, installs AI cloud and computing resources inside them and rents that capacity to Big Tech firms. With inflation showing no sign of easing and keeping upward pressure on rates, analysts increasingly warn that funding costs for AI infrastructure investment could snowball. If Big Tech's AI infrastructure projects fail to proceed as planned, the damage would ripple across the entire AI ecosystem — from Nvidia, which supplies graphics processing units, to the memory chipmakers that provide high-bandwidth memory to chip designers. The U.S. economy, which depends on AI-related investment for much of its growth, could also slow. With AI-linked shares in an unprecedented slump, Wall Street is split between those who see this as an early warning that the so-called semiconductor supercycle is breaking and those who call the concern overdone.
GPU infrastructure provider CoreWeave's bond yield at 15%; high-risk circular financing model in crisis
On the 14th, Wall Street was jolted by news of the credit deterioration at CoreWeave, the world's largest provider of GPU infrastructure. In the bond market that day, the yield on CoreWeave's bonds maturing in 2032 jumped to as high as 14.8%. According to Bloomberg, the bonds, sold at face value four months earlier, had fallen to 78% of that price. Wall Street typically treats corporate bonds trading below 80 cents on the dollar as effectively distressed debt.
The probability of default within five years implied by CoreWeave's credit default swap prices also topped 60%. On the 15th, CoreWeave's five-year CDS premium rose above 12 percentage points during trading, an all-time high. A CDS is a form of insurance purchased against the possibility that a particular company defaults. The higher the CDS premium, the greater the market judges that company's default risk to be.
CoreWeave began as Atlantic Crypto, a cryptocurrency mining firm founded in 2017 by three people: Chief Executive Michael Intrator, Chief Strategy Officer Brian Venturo and Chief Development Officer Brannin McBee. In its early days, the company mined Ethereum with a single Nvidia GPU out of an office in Manhattan, New York. After cryptocurrency prices crashed in 2019, it changed its name to CoreWeave and repurposed its GPUs for cloud computing. Demand for AI servers surged after OpenAI unveiled ChatGPT on Nov. 30, 2022, and CoreWeave's value climbed vertically. In 2023, it raised $100 million in Series B funding from Nvidia and began locking up the newest GPUs that even Big Tech firms struggled to obtain. From 2024, it signed large infrastructure supply deals with top global Big Tech firms including Microsoft and OpenAI, putting it on a steep growth path.
In March last year, it listed on the Nasdaq and raised about $1.5 billion. It has since operated more than 50 AI data centers across North America and Europe, securing GPU infrastructure of more than 250,000 units.
The problem is that CoreWeave's business model depends entirely on rapid growth in the AI industry. Its basic structure is circular financing: it pledges GPUs already obtained from Nvidia as collateral to borrow heavily from banks, then uses that money to buy the newest GPUs and build additional data centers to lease to Big Tech. If the AI investment boom continues, capital can expand explosively, but even a slight slowdown can bring the business to a halt — a textbook ultra-high-risk model.
Second-quarter revenue reached $2.575 billion, up 112.4% from a year earlier, but the net loss came to $626 million. Debt that CoreWeave must repay in cash, principal and interest alike, stood at $35.6 billion, and interest expenses alone reached $604 million in the second quarter. CoreWeave's free cash flow has also been in net outflow since 2022.
Tech stocks fall day after day on AI funding fears; Samsung Electronics and SK hynix market value shrinks sharply
The CoreWeave episode immediately hit the stock market. On the 14th in New York, CoreWeave shares plunged 15.46%, while Nebius and Oracle, neo-cloud firms with similar business structures, dropped 10.08% and 8.83%, respectively. Large tech names also fell across the board: Nvidia (-3.16%), Microsoft (-1.80%), Amazon (-1.74%), Google parent Alphabet (-1.74%), TSMC (-3.40%), SpaceX (-2.80%), Facebook parent Meta (-3.27%), Broadcom (-3.10%), Tesla (-2.15%), SK hynix (000660.KS) (-6.06%), Micron (-4.08%) and AMD (-5.73%). The Philadelphia Semiconductor Index fell 4.30%.
Intrator sent a memo to employees that day arguing that "there are times when the market misunderstands and behaves irrationally" and that the order backlog is more than enough to cover all of the company's debt. It was not enough to turn sentiment around. SpaceX Chief Executive Elon Musk wrote "yikes" about the CoreWeave situation on X, formerly Twitter, the same day.
CoreWeave shares temporarily rebounded 5.94% on the 15th, helped by comments from Nvidia Chief Executive Jensen Huang, who is consistently optimistic about AI. At an event in Stockholm, Sweden, that day, Huang said CoreWeave's financial capability is improving every day and that Nvidia does not need to bail the company out financially. CoreWeave shares recovered some ground, but most other tech names, including Nvidia (-1.84%), kept falling. The Philadelphia Semiconductor Index dropped 0.88% that day even as Taiwanese foundry TSMC posted a record quarterly net profit for the third quarter. TSMC said third-quarter net profit rose 68.7% from a year earlier to NT$763.1 billion (about 31.7 trillion won), a sixth straight quarterly record.
The aftershocks continued on the 16th. CoreWeave shares fell as much as 11.91%, while Nebius (-12.18%), IREN (-11.19%) and Oracle (-4.64%) also dropped sharply. With most tech shares including Micron (-5.63%) falling steeply, all three main indexes declined: the Dow Jones Industrial Average (-0.34%), the Standard & Poor's 500 (-1.09%) and the Nasdaq Composite (-2.15%). The Philadelphia Semiconductor Index plunged 3.83%, an eighth consecutive session of losses — its longest losing streak since 2008, during the global financial crisis. Over that period, the index lost a total of 15.74%.
As capital flowed out of AI-related shares, money rotated into health care, consumer and financial stocks. Among tech names, only Apple — known for lagging in AI — benefited, topping $5 trillion in market value for the first time on the 14th.
The shock from CoreWeave spread beyond New York to Asian markets including South Korea and Japan. According to CNBC on the 15th, the combined market value of the world's 20 most valuable AI and semiconductor stocks evaporated by more than $1 trillion in the four trading sessions from the 12th through that day. Over that stretch, Nvidia shares fell 5.3% and its market value shrank by about $291 billion, while SK hynix lost about $24.4 billion and Samsung Electronics (005930.KS) about $23.7 billion. The three largest memory chipmakers — Samsung Electronics, SK hynix and Micron — accounted for about 40% of the decline across the 20 companies.
Meanwhile, according to Bloomberg on the 15th, CoreWeave's three co-founders — Intrator, Venturo and McBee — have sold $2.4 billion worth of shares so far this year. CoreWeave shares are now more than 60% below their peak. Compared with June last year, when the share price peaked, the paper wealth of the three founders has fallen by about $15 billion.
China's memory push adds to risk of a later payback; a Fed rate increase would be a second shock
The pullback in AI-related investment driven by surging U.S. Treasury and corporate bond yields is not new. With no sign of a letup in inflation or in the expansion of government debt, corporate funding costs are already near their limit. According to Britain's Financial Times on the 12th, Morgan Stanley put new AI-related debt raised worldwide last month at $23 billion (about 31 trillion won), more than half below the August figure. New AI-related debt issuance has fallen steadily since reaching $113 billion (about 152 trillion won) in June. In the U.S. investment-grade bond market, where Big Tech firms raised $306 billion (about 410 trillion won) from January through August, AI-related corporate bond issuance effectively stopped last month. So far this year, AI-related debt totals $466 billion (about 624 trillion won), more than four times the $101 billion (about 135 trillion won) raised in the same period last year.
Broadcom Chief Executive Hock Tan said at the DBS Asia Future Forward conference in Singapore on the 12th that the current memory chip shortage could look different a year from now, while warning that the biggest constraint may be capital. Intel Chief Executive Lip-Bu Tan said the most pressing issue is the rise in the cost of capital, adding that investors' desire to see a return on invested capital is also a challenge. CNBC noted that as higher rates raise infrastructure investment costs, the cost of building data centers and buying advanced chips also rises, which can severely damage the profitability of AI projects.
Externally, China's advance in chip technology has become another drag on AI-related share prices. According to Chinese outlets Yicai and STCN on the 12th, CXMT President Cao Kanyu said at a recent event that the company plans to release DDR5 RDIMM, a next-generation memory chip for servers, by year-end — a declaration of intent to seize technology on a par with Samsung Electronics and SK hynix. CXMT raised as much as 66.61 billion yuan (about 13.2 trillion won) through an initial public offering in July. Its shares surged from the first day of trading, making it the most valuable company on China's mainland market. Musk wrote on X on the 15th that China will produce an enormous amount of RAM in 2028, suggesting that a memory chip glut could arrive sooner than expected.
The possibility of further rate increases by the Federal Reserve could, depending on timing, prove disastrous for AI-related shares, since it could push already-high bond yields even higher. According to CME FedWatch, the federal funds futures market raised the implied probability that the Fed holds rates steady at its Oct. 27-28 Federal Open Market Committee meeting to 84.5% on the 16th from 72.3% on the 15th. Steven Zeng, U.S. rates strategist at Deutsche Bank, told Bloomberg on the 15th that even with strong readings on the consumer price index, the producer price index and retail sales, the market has been reluctant to predict that the Fed will rush to raise rates, adding that if the bad news around AI snowballs, concern about a slowing economy could outweigh worries about inflation.
On the 16th, Fed Chair Kevin Warsh, in a conversation with International Monetary Fund Managing Director Kristalina Georgieva at the IMF-World Bank annual meetings in Bangkok, Thailand, identified AI-driven investment as the most important upside factor for the global economy. Warsh nonetheless said he does not prejudge the outcome of the meeting, firmly declining to hint at the path of monetary policy. Cleveland Federal Reserve Bank President Beth Hammack, regarded as one of the Fed's leading hawks, said at a Goldman Sachs event in New York on the 15th that investment in the technology sector is proceeding at full speed even as borrowing costs rise, and that if expectations take hold that inflation pressure will become increasingly persistent, rates may have to be raised by a much larger margin.
Whether the semiconductor supercycle riding on AI demand is truly nearing its end cannot yet be determined. The stock market has overcome several rounds of AI bubble talk since the second half of last year and has kept trending higher. Still, if high rates persist, the odds look considerable that the debt-financed AI project timelines of each Big Tech firm will be disrupted. In that case, the payback that Wall Street has been counting on would inevitably be pushed back. For now, there is a chance that another cold wind blows through the KOSPI market on Monday the 19th, Korea time.
※ "Trump Stocker" is a column delivering on-the-ground reporting and analysis of U.S. markets, companies, policy, politics and diplomacy that may help investors during the presidency of Donald Trump. Subscribe to receive useful news from the United States.