NEW YORK — The sharp deterioration in CoreWeave’s stock and credit markets is casting a harsh light on the financial underpinnings of the artificial intelligence infrastructure boom. While demand for AI computing power remains extraordinary, the company’s aggressive expansion, fueled by billions in borrowed capital, is now facing the unforgiving reality of increasingly expensive money. The situation at CoreWeave, a specialized cloud provider, reveals a paradox: even with a massive backlog of contracted revenue, the economics of supplying that demand can unravel if the cost of financing the necessary infrastructure becomes too high.
CoreWeave operates as a dedicated platform for high-performance, GPU-intensive workloads, primarily serving AI innovators and major enterprises like Microsoft, OpenAI, Meta Platforms, and Anthropic. Its business model involves signing large, often multi-year, fixed-price contracts—some extending up to six years—to provide AI compute capacity. To fulfill these commitments, CoreWeave requires enormous upfront capital to acquire vast quantities of Nvidia GPUs, secure data-center capacity, power these facilities, and construct the necessary infrastructure. For instance, building a 100-megawatt GPU cluster can involve over a billion dollars for the physical data center, with CoreWeave responsible for purchasing the GPUs.
This aggressive expansion has been financed through a combination of corporate debt, GPU-backed lending, and data-center/project financing, often utilizing delayed draw term loan (DDTL) facilities. These structures allow CoreWeave to access capital in stages as it deploys infrastructure, with the underlying GPUs, infrastructure, and contracted customer cash flows serving as collateral. In a simple hypothetical, a $1 billion multi-year compute contract might require hundreds of millions upfront for infrastructure. Lenders provide this capital, expecting future customer payments to cover operating expenses, interest, and principal, leaving CoreWeave with a return. This model is powerful when capital is cheap, enabling rapid expansion. However, it becomes precarious when borrowing costs rise.
Social Capital’s analysis, now echoed by broader market scrutiny, points to a critical vulnerability: the interaction between CoreWeave’s fixed-price customer contracts and its increasingly expensive, often floating-rate, financing. CoreWeave cannot simply raise contract prices because its borrowing costs increase if it agreed to provide compute at economics established when financing conditions were favorable. This creates a significant squeeze on profit margins. For illustration, if original economics saw $100 revenue, $65 infrastructure/operating costs, and $15 financing costs, leaving a $20 cushion, higher financing costs of $28 would shrink that cushion to just $7. These numbers are illustrative, not CoreWeave’s actual unit economics, but they highlight the problem: AI demand can remain extraordinarily strong, with CoreWeave boasting a backlog of approximately $104 billion in contracted future revenue, while the economics of supplying that demand deteriorate due to rising borrowing costs. The company’s Q3 interest expense was guided as high as $940 million, significantly overshadowing its Q2 operating loss of $49 million, on a total debt of $51.6 billion.
CoreWeave’s borrowing costs have seen a sharp increase. While some GPU-backed facilities, such as the $8.5 billion DDTL 4.0 facility closed in March 2026, were secured at favorable rates (SOFR + 2.25% floating, roughly 5.9% fixed) and even received investment-grade ratings, the overall picture has shifted. Reuters and LSEG data estimate CoreWeave’s total debt at approximately $50 billion to $52 billion. Brad Gerstner of Altimeter Capital, an early investor, alleged that about 80% of this debt is relatively inexpensive GPU-backed financing, with the company paying around 9% on its existing unsecured bonds. However, the cost of newly issued debt and the market’s perception of CoreWeave’s creditworthiness have deteriorated significantly. Yields on some of CoreWeave’s bonds have recently exceeded 15-16%. For instance, its 9.75% bond due in 2031 reached a yield to maturity of 15.78%, and its 9.625% notes maturing in 2032 traded for less than 80 cents on the dollar, equating to a yield of nearly 15%. These levels are typically associated with distressed debt.
The cost of insuring CoreWeave’s debt against default has also soared. Its five-year credit default swap (CDS) spreads reached a record level of over 1,000 basis points (10%). On October 20, 2026, CoreWeave’s CDS spreads were over 1,100 basis points, and on October 21, 2026, they stood at nearly 1,067 basis points, representing a 164% increase year-over-year and 30% in October alone. LSEG data indicated this implies a nearly 63% chance of default in five years, up from about 30% a year ago. CoreWeave faces significant refinancing requirements, with approximately $4.4 billion of debt maturing through 2026 and another $6.2 billion in 2027. Refinancing at current market rates of 15% or higher is likely not economically feasible. The company still needs billions for future data-center construction, with expected 2026 capital expenditure of $30-$35 billion, more than doubling 2025 levels. While rising bond yields in the secondary market do not automatically increase interest payments on all existing debt, the marginal cost of the next dollar CoreWeave borrows matters immensely for its growth story, impacting its ability to fund ambitious expansion and eroding future profitability.
A critical aspect of CoreWeave’s financial health is the distinction between contracted or planned capacity and what is actually operational. CoreWeave has promised customers enormous amounts of future compute, evidenced by its $104 billion contracted revenue backlog. However, delivering on these contracts requires substantial, ongoing infrastructure build-out. The company more than quadrupled its data center presence from three to 14 in 2023 and anticipated doubling its footprint to 28 globally by the end of 2024, with plans to invest $978.6 million in UK data centers by 2024. Its expected capital expenditure for 2026 is between $30 billion and $35 billion. This indicates that a significant portion of CoreWeave’s future growth is already contracted but not yet built or fully financed. The future profitability of these projects depends heavily on the price at which this infrastructure can be financed. If borrowing costs remain high or continue to climb, the economics of delivering on these pre-existing contracts could become severely strained, even with strong customer demand.
While CoreWeave is a private company, its debt instruments trade in the market, and their performance offers a stark indicator of investor sentiment. CoreWeave’s stock has experienced significant volatility, with shares falling by nearly 50% since April 2026, including a 15% decline on one day in October 2026 and a 29% drop for the month. However, the credit market signals are arguably more critical. When CoreWeave bonds trade significantly below par, it means investors are willing to pay less than face value, driving up the effective yield and reflecting a higher perceived risk of the company not meeting its obligations. A sharp rise in CDS spreads, soaring to over 1,000 basis points, indicates a dramatically increased cost of insuring CoreWeave’s debt against default. Credit investors are effectively pricing in a higher probability that something could go wrong years before an actual default might occur.
This creates a potential feedback loop: higher broader market interest rates and risk premiums lead to CoreWeave’s new borrowing becoming more expensive. This, in turn, causes future data-center projects to produce lower returns, leading investors to perceive greater financial risk. CoreWeave bonds fall, and CDS spreads rise, prompting lenders to demand even higher returns, making future financing still more expensive. Ultimately, equity investors lower the value they assign to CoreWeave’s future growth. This loop appears to be actively occurring, with reports from October 2026 indicating that CoreWeave’s debt selloff is raising concerns over private credit’s exposure to AI infrastructure financing, with borrowing costs climbing to distressed levels. The market is clearly hedging against potential financial instability in the AI sector, with CoreWeave as a prominent example.
Another potential vulnerability lies in the nature of CoreWeave’s collateral: highly valuable Nvidia GPUs. CoreWeave has borrowed heavily against infrastructure containing these chips, with facilities like the $2.3 billion debt financing in August 2023 and the $7.5 billion facility in May 2024 explicitly using Nvidia H100 GPUs as collateral. The question arises: what happens to lenders' collateral assumptions if newer generations of GPUs make existing chips economically obsolete faster than expected? This is a distinction between physical depreciation and economic obsolescence. A GPU may function perfectly, but its value can diminish if a newer chip delivers significantly more compute per watt or per dollar. CoreWeave was among the first cloud service providers to receive shipments of Nvidia's new Blackwell hardware in November 2024, implying a continuous upgrade cycle. AI infrastructure could begin behaving financially like aircraft or ships: enormously expensive assets financed with debt whose economics depend on remaining productive long enough to repay the financing. However, some lenders are reportedly willing to underwrite renewal risk, financing shorter-dated customer contracts with five-year debt maturities, signaling confidence in the long-term value of Nvidia GPUs.
CoreWeave’s dependence on a few major customers presents a dual-edged sword. Having technology giants like Microsoft, OpenAI, and Meta Platforms sign enormous, long-term contracts makes CoreWeave highly attractive to lenders due to reliable and substantial future revenue, evidenced by its $104 billion contracted backlog. However, this creates extreme customer concentration. According to company filings, Microsoft accounted for about 67% of CoreWeave's 2025 revenue, up from 62% in 2024. OpenAI has signed deals worth up to $22.4 billion, and Meta Platforms outlined plans to spend $14.2 billion on cloud services from CoreWeave through December 2031, later expanded to a $21 billion agreement through December 2032. Nvidia itself committed to purchase any of CoreWeave's unsold compute capacity through 2032 for $6.3 billion. While these contracts provide strong revenue visibility, extreme customer concentration means that losing, renegotiating, or experiencing problems with even one major customer could have an outsized and potentially devastating effect on the company’s financial stability. OpenAI's new $4 billion cloud deal with CoreWeave in May 2025 was also seen as a move to diversify its cloud infrastructure beyond Microsoft and ease CoreWeave's dependence on Microsoft.
Despite the recent credit market turmoil, a strong bullish case for CoreWeave persists, rooted in the fundamental drivers of the AI boom. AI compute demand remains extraordinary and insatiable, with major technology companies spending hundreds of billions on AI infrastructure. CoreWeave’s specialized focus positions it directly in the path of this massive spending wave. The company boasts long-term take-or-pay contracts with exceptionally strong counterparties, providing unusually strong revenue visibility. CoreWeave also possesses scarce access to Nvidia GPUs, power, and data-center capacity, a significant competitive advantage in a constrained market. Nvidia itself is a shareholder and key partner. Furthermore, falling financing costs could rapidly improve the economics again. If interest rates stabilize or decline, the pressure on CoreWeave’s margins from expensive debt could ease. Some investors, like Brad Gerstner, argue that the market’s focus on the high yields of a small portion of CoreWeave's debt misses the lower average cost of its GPU-backed financing. Nvidia CEO Jensen Huang also publicly stated in October 2026 that CoreWeave's cost of capital has actually declined. Newer GPUs could also generate substantially more revenue per megawatt, enhancing profitability. CoreWeave could therefore emerge as one of the most important AI infrastructure companies, leveraging its strategic partnerships, massive demand, and specialized infrastructure to become a cornerstone of the AI revolution.
The initial phase of the generative AI boom captivated investors with the marvels of Nvidia chips, the rapid adoption of ChatGPT, and the extraordinary capital expenditure of tech giants. This era was largely defined by technological breakthroughs and seemingly limitless equity funding. However, the sharp deterioration in CoreWeave’s credit markets reveals that the next phase of the AI boom may depend just as much on something far less glamorous: the price of money. CoreWeave represents an extreme example because it combines extraordinary AI demand with extraordinary capital intensity and leverage, accumulating over $50 billion in debt and financial obligations. The question therefore isn't simply: “Is AI a bubble?” It is: “Can the financial system fund the infrastructure promised by the AI boom at a cost that still makes the economics work?” CoreWeave’s situation underscores that even with a massive backlog of contracted revenue and undeniable demand for AI compute, the economics can unravel if the cost of financing the underlying infrastructure becomes too high. If the answer depends on borrowing at 7–9%, then a world where marginal financing costs 12–15% could produce very different winners and losers—even if demand for AI continues growing. The current market signals, with CoreWeave’s bond yields reaching distressed levels and CDS spreads soaring, suggest that the financial system is repricing the risk of this capital-intensive build-out. This shift means that access to affordable capital, rather than just access to GPUs or customer contracts, could become the ultimate bottleneck for the AI infrastructure boom.