CoreWeave bond rout puts debt-funded AI infrastructure under scrutiny
CoreWeave's bonds have slipped into distressed territory and its shares are down over 20 percent. What the credit scare means for India's own GPU build-out.
The News
Bonds issued by CoreWeave, the US cloud company that rents out Nvidia GPUs to AI developers, slid into distressed territory last week, with investors demanding more than 1,000 basis points of extra yield over government debt to hold them. The company's shares have fallen more than 20 percent since the close on Tuesday, according to a report in The Verge.
The sell-off followed a public post by investor Chamath Palihapitiya flagging the distressed debt and questioning whether the AI infrastructure boom has become a private credit bubble. On the same day, cofounder Brannin McBee told the Financial Times that CoreWeave wants its debt rated investment-grade so it can borrow from insurers, whose capital he described as "unlimited". Nervous bondholders read that as a signal of more borrowing to come.
The figures behind the worry are large. As of its second-quarter results, CoreWeave owed more than $35 billion and had lost more than a billion dollars in the first half of the year. Against that sits $103.7 billion in remaining performance obligations, of which about 40 percent falls due in 2028. Three customers, Microsoft, OpenAI and an undisclosed third widely assumed to be Nvidia, generated almost three-quarters of second-quarter revenue.
Oracle, another heavy AI borrower, saw its shares fall almost 10 percent over the same stretch, alongside a bond sell-off.
Why It Matters
CoreWeave pioneered the use of GPUs as loan collateral, which makes it the clearest test case for whether debt-funded compute can earn a return. The tension is timing: it pays interest today on fast-depreciating chips, while much of its contracted income lands years out.
The closest historical parallel is the fibre-optic overbuild of 1998 to 2001. Carriers borrowed heavily to lay capacity ahead of projected internet demand; when credit tightened, Global Crossing and WorldCom both filed for bankruptcy in 2002, even though the traffic eventually arrived. MetLife Investment Management's John Yovanovic made a similar point to Bloomberg, noting that the concern is the supply of new debt still to come rather than any change in existing credit profiles.
Indian Angle
India is in the early stages of its own GPU build-out. The IndiaAI Mission, cleared by the Union Cabinet in March 2024 with an outlay of Rs 10,371.92 crore, subsidises access to compute run by domestic providers such as Yotta Data Services, while conglomerates including the Adani and Reliance groups have announced large data-centre programmes. These operators finance expansion partly through dollar loans and bonds, so a global repricing of AI infrastructure risk raises their cost of capital too, and that ultimately shows up in the per-hour GPU rates Indian startups pay.
There is also a portfolio angle. Indian retail investors have built meaningful exposure to US technology through the Liberalised Remittance Scheme and international feeder funds, many of which are heavy in AI infrastructure names. A credit-led correction in that segment would hit those holdings directly.
For Indian lenders and regulators, CoreWeave is a useful warning on collateral quality. GPUs lose resale value quickly, and banks or NBFCs considering chip-backed or data-centre loans will want stress tests that assume shorter useful lives than standard IT equipment schedules.
FAQ
What does it mean for CoreWeave's bonds to be distressed?
A common market definition treats a bond as distressed when investors demand a yield at least 1,000 basis points, or 10 percentage points, above comparable government bonds. CoreWeave's debt crossed that line last week. It does not mean the company has defaulted, but it does mean any new borrowing is likely to cost it considerably more.
Is CoreWeave at risk of collapse?
Not imminently, on the available numbers. The company reported $103.7 billion of remaining performance obligations as of the second quarter, contracts that customers can cancel only in very limited circumstances. Its largest clients include Microsoft and OpenAI. The pressure is on profitability and refinancing costs rather than on near-term revenue visibility.
Why should Indian investors care about a US neocloud?
Many Indian investors hold US technology shares through the Liberalised Remittance Scheme or international mutual funds, and AI infrastructure names now make up a large slice of those indices. Repricing of AI debt can also raise borrowing costs for Indian data-centre operators that fund expansion with dollar loans and bonds.
This story was reported by The Verge. Read the full original coverage at The Verge.