Wall Street heavyweights including Goldman Sachs and JPMorgan have delivered bullish views and optimistic outlooks on CoreWeave, Inc. (CRWV), the stock at the very epicenter of the "AI compute selloff" and "AI bubble burst" narrative, even as AI NeoCloud listed companies continue to see their stocks and bonds plunge.
JPMorgan stated directly on Tuesday that most global commercial banks have very limited investment and allocation exposure to CoreWeave, that the related loans carry sufficiently adequate collateral documentation, and that funds are disbursed strictly in tranches tied to AI infrastructure project milestones.
Combined with Goldman Sachs upgrading the stock rating a day earlier amid CoreWeave's steep share price decline, the positive judgment from these Wall Street giants is becoming increasingly clear: as cutting-edge AI agents like Muse and Astra sweep across the globe, the explosive expansion of AI compute demand and project economics still have strong support, and the market is reassessing what valuation discount should correspond to AI infrastructure-related financing pressures; however, the specific safety margin of bank loans versus the investment returns of common equity still depends on different cash flow distribution priorities.
On Monday, despite the bullish boost from Goldman Sachs upgrading CoreWeave's stock rating to "Buy," CoreWeave shares rose about 5% intraday but ultimately closed down 0.44% at $61.12; its dollar bonds maturing in 2031 saw yields briefly breach 16% last week before retreating to 15.13% on Monday as market pessimism eased somewhat.
Goldman Sachs research shows that its analysts used a five-year contract as an illustrative model: initial capital expenditure of $6 billion, annual revenue of $2.6 billion, minus annual operating expenses of $585 million and approximately $213 million in cash taxes, yielding annual after-tax unlevered free cash flow of approximately $1.802 billion. Combining the initial $6 billion investment with the subsequent five years of cash flows, the after-tax unlevered internal rate of return (Unlevered IRR) is approximately 15.3%. This directly demonstrates that under the model's assumptions, the project can recoup its investment, cover operating costs and taxes, and generate returns.
Goldman Sachs' final calculation data shows that under assumptions of $13 million annualized revenue per megawatt and a 10% debt interest rate, Goldman measured a weighted average cost of capital (WACC) of 11.8%, significantly below the aforementioned 15.3% project return, creating a positive return spread of approximately 3.5 percentage points. Therefore, Goldman's latest calculation fully demonstrates that after covering investment and operating expenses, the project still earns returns exceeding what capital providers require, thereby creating economic value.
The latest bullish views from Goldman Sachs and JPMorgan together strongly reflect that after covering investment and operating costs, the project is still expected to earn returns exceeding the cost of capital; the financing needed for continued expansion is another problem that must be solved.
Goldman Sachs believes that CoreWeave primarily relies on asset-backed financing tied to customer contracts, GPU assets, and project cash flows, with new debt costs estimated at approximately 7.5% to 10.5%, with the model adopting 10%. Therefore, the 13.5% to 15% trading yields on unsecured bonds do not directly represent the financing cost of all new projects.
Additionally, Goldman Sachs stated that the approximately 15.3% calculation does not include GPU residual value after the five-year contract ends; under Goldman's separately modeled sixth-year residual value scenario, the corresponding return could increase to approximately 20% to 25%. This means its baseline argument is already built on first-contract cash flows, with subsequent asset monetization constituting an additional source of returns.
Philadelphia Semiconductor Index Ends Losing Streak, but AI Rebound Still Needs to Clear Corporate Credit Hurdle
On October 19, the Philadelphia Semiconductor Index rose 1.61%, ending an eight-day losing streak, with Micron and SanDisk gaining 2.97% and 5.53% respectively, as memory chips led the rebound. However, the SOX had fallen approximately 11.4% over the prior week, and as of Monday's close, was still down 10.4% for October and about 22.8% below its 52-week high. Despite still being up 59.8% year-to-date, the阶段性 correction has clearly crossed the threshold typically described as a technical bear market.
The Korean market has also failed to form a sustained recovery: after the KOSPI rebounded 1.76% on Monday, it fell another 3.73% on October 20, closing at 5,552.74 points. Bottom-fishing has occurred, but a full restoration of risk appetite remains to be confirmed.
CoreWeave, Inc. (CRWV), at the very center of this AI selloff superstorm, has borne more concentrated pressure. Its shares fell about 25.2% last week, and despite an intraday rebound on Monday, still closed down 0.44% at $61.12; the stock is down about 29.8% for October, down 14.7% year-to-date, and approximately 56.9% below its 52-week high.
The repricing in financing markets has also extended to unlisted peers: Nvidia-backed Nscale is seeking an IPO valuation of approximately $30 billion, but some potential investors are demanding pricing at $16 billion to $19 billion, equivalent to a discount of approximately 37% to 47% — notably, this is an investor demand for a lower price, not a completed valuation reduction.
The underlying mechanism of this stock-bond resonance is the repricing of the financing gap between front-loaded capital expenditure and multi-year repayments. GPU clusters must complete power supply, liquid cooling, high-speed interconnect, and delivery acceptance before they can continuously generate compute service revenue; elevated long-end interest rates combined with widening credit spreads raise new financing costs and压低 the present value of future cash flows. CoreWeave's 2031-maturing bonds saw yields briefly approach 16% last week, with CDS at approximately 1,000 basis points; Reuters cited LSEG data indicating a corresponding five-year cumulative implied default probability of nearly 63%, a figure that includes risk premiums and depends on recovery rate assumptions, and is not a direct prediction of actual default probability. On Monday, the bond yield had retreated to approximately 15.13%, indicating some easing of credit pressure, but the financing threshold associated with AI infrastructure remains very high. Although in Goldman's view the project is expected to earn actual investment returns exceeding the cost of capital after covering investment and operating costs, the financing needed for continued expansion is another problem that must be solved.
Will AI Financing Pressure Contagion Spread to Global Banks? JPMorgan: CoreWeave Bond Selloff Poses Limited Risk to Global Banks
A latest research report published by JPMorgan on Tuesday shows that most global commercial banks have limited exposure to CoreWeave, Inc. (CRWV). Previously, amid investor concerns about debt financing issues, shares and bonds of this Nvidia-backed AI cloud computing company fell sharply last week. AI infrastructure spending is making global institutional investors increasingly uneasy. Technology companies worldwide are raising large sums from private credit investors, banks, and bond markets to fund rapid expansion, while market doubts about the returns on these investments are growing, and concerns are mounting over whether AI infrastructure financing pressures could spread from bond trading markets across the entire global banking system, similar to the 2023 "Silicon Valley Bank crisis."
This major Wall Street financial giant said in its research report that large U.S. investment banks and Japan's Mitsubishi UFJ Financial Group may have potential exposure to CoreWeave in the range of $1 billion to $3 billion, while most European and Asian banks have exposure below $500 million. JPMorgan stated that a significant portion of this exposure comes from delayed draw term loans, which are primarily arranged through special purpose vehicles, with funds released as projects reach phased milestones.
Analysts led by Kian Abouhossein wrote in the report: "From the perspective of the banks providing financing, we believe that banking institutions have managed their exposure to CoreWeave quite well and have very adequate collateral." The firm said that because banks have limited disclosure of their exposure to the AI investment cycle, the above estimates are based on "simplified assumptions" calculated by the firm. It is understood that JPMorgan's equity analyst team gives CoreWeave an "Overweight" rating, with a target price of $125 by the end of 2027.
The firm's research report shows that this cloud computing company focused on the AI NeoCloud sector has obtained approximately $24 billion in delayed draw term loan commitments, of which $15 billion has been drawn and $9 billion remains undrawn. CoreWeave, known as Nvidia's "own child," focuses on providing high-performance computing infrastructure for AI and has signed tens of billions of dollars in contracts with large technology companies. The company's debt scale is approximately $50 billion.
As the earliest cloud rental adopter of Nvidia graphics processors (AI GPUs) in the data center sector, CoreWeave seized the wave of data center AI compute resource demand, won favor from Nvidia's venture capital arm, and can even obtain priority access to the extremely sought-after Nvidia H100/H200 and Blackwell series AI GPUs. It once forced cloud service giants like Microsoft, Google, and Amazon to lease cloud AI compute resources from CoreWeave, earning the title of Nvidia's "own child."
The most prominent feature of CoreWeave's AI cloud compute rental service is its focus on providing the most high-end AI GPU (especially NVIDIA GPU) clusters in large volumes and with high efficiency, allowing users to obtain high-performance AI GPU compute resources on demand through cloud service systems — i.e., cloud AI compute resources — for AI workloads such as machine learning, deep learning, and inference. CoreWeave supports large-scale elastic deployment, allowing users to quickly scale AI GPU counts up or down based on project needs, making it suitable for AI model training (such as large language models, computer vision systems, etc.) and massive inference workloads requiring real-time processing.
Last week, yields on CoreWeave's 2031-maturing notes briefly approached 16%, while the cost of buying default insurance on its debt rose to approximately 1,000 basis points. JPMorgan said the main risk facing banks is the timing mismatch between upfront AI infrastructure spending and revenue received gradually afterward. The firm added that it expects CoreWeave's free cash flow to remain negative through 2030.
Banks Look at Collateral Protection, Shareholders Look at Project Returns
The key to JPMorgan's judgment is that banks do not bear CoreWeave risk in the same way. It estimates that large U.S. investment banks and Mitsubishi UFJ may have potential exposure in the $1 billion to $3 billion range, while most European and Asian banks are below $500 million; a large portion of the related financing uses delayed draw term loans, secured by equipment, customer contracts, and project cash flows, with funds released gradually as projects reach milestones.
Company disclosures also show that some financing is undertaken by special purpose subsidiaries holding equipment and contracts, with the related claims enjoying collateral protection. However, most such loans still have parent company guarantees, and SPVs do not mean risk is completely isolated. The nearly 16% yield on unsecured bonds in the secondary market also cannot be directly regarded as the overall bank loan or company-wide cost of capital.
The equity bullish logic rests on whether the project can create economic value. JPMorgan's previously illustrative calculation showed that a typical five-year contract project return is in the "mid-teens" percentage range, above its estimated 9% to 11% cost of capital, meaning that under the relevant assumptions, the project can still create returns exceeding the cost of capital after covering initial investment and operating costs; GPU renewals after the initial contract ends could also bring additional cash flows.
Goldman Sachs is bullish on compute supply tightness and CoreWeave's delivery and performance advantages, believing there is room for improvement in revenue per unit of capacity and project returns. The shared judgment of both institutions is that free cash flow consumption caused by rapid expansion can coexist with good single-project economics; what truly needs continuous verification is whether financing can be bridged, whether capacity can be delivered on schedule, and whether contract payments can be realized.
Translating this into actual Wall Street valuations and ratings, Goldman Sachs upgraded its rating from "Neutral" to "Buy" while lowering its target price from $139 to $106, implying approximately 73.4% upside from Monday's closing price; JPMorgan maintained "Overweight" with a target price of $125 by the end of 2027, implying approximately 104.5% upside. Separately, according to S&P Global statistics presented by Stock Analysis, the composite rating from 41 analysts is "Buy," with an average twelve-month target price of $140.58, corresponding to approximately 130% potential upside. Target price horizons differ, but together they reflect that institutions are still willing to assign value to long-term cash flows; Goldman Sachs simultaneously lowering its target price also indicates it has already incorporated more expensive AI infrastructure financing conditions into its judgment.
Comments