Is CoreWeave's credit crisis more like the 2023 SVB panic rather than the 2008 subprime mortgage crisis?

The current key is not whether CoreWeave will collapse, but whether computing power pricing can support the financing logic of this cycle. The answer to this question will determine the ultimate destination of this panic.

The yield on CoreWeave's unsecured bonds has soared to around 15%, and the CDS spread has broken through 1,000 basis points, raising strong doubts in the market about the systemic risks of the entire AI capital expenditure cycle.

However, top Wall Street institutions are collectively speaking out: this is not a 2007 New Century Financial-style systemic collapse, but more like the phased panic triggered by Silicon Valley Bank (SVB) in 2023 — an extremely small and structurally fragile instrument is being treated by the market as a barometer of the trillion-dollar AI capital expenditure cycle.

Ajay Rajadhyaksha, head of global macro research at Barclays, pointed out that CoreWeave's approximately $1.5 billion in unsecured bonds are structurally subordinated in its total debt of about $35 billion. Using the price signal of this instrument to negate the entire AI cycle is no different from predicting a U.S. recession with a roughly $4 billion regional bank ETF in 2023. Goldman Sachs argues that the market has chosen the wrong reference frame — the unsecured bond market does not represent CoreWeave's real new financing cost.

Historical Precedent: SVB or the 2007 Financial Crisis?

The core of this debate lies in the choice of historical analogy.

In April 2007, New Century Financial, the largest U.S. subprime mortgage lender, filed for bankruptcy. Afterwards, the entire credit market gradually collapsed, and the S&P 500 index fell by about 40% in the following two years. In March 2023, Silicon Valley Bank was closed by the regulatory authorities, and the market immediately formed a narrative: regional banks are the lifeblood of credit for small and medium-sized enterprises, and the depletion of credit will trigger a recession. The bond market quickly priced in 75 to 100 basis points of interest rate cuts that year, and the VIX panic index even surged more than in 2007.

As a result, the Federal Reserve did not cut interest rates at all in 2023, but continued to raise them instead. The U.S. economy's full-year growth far exceeded expectations, and the recession never arrived. The regional bank ETF KRE recouped all its losses, and the S&P 500 rose by about 40% in the following two years.

Bloomberg columnist John Authers pointed out that both crises seemed severe in real time, but their final outcomes were completely different. SVB was a panic that was over-extrapolated, not a precursor to systemic risk; the bankruptcy of New Century Financial revealed a structural defect that ran through the entire financial system. CoreWeave's current situation is morphologically closer to SVB.

Rajadhyaksha summed up the core mistake of the SVB panic: "The mistake was not recognizing that Silicon Valley Bank had collapsed. It did collapse. The mistake was extrapolating from a small, highly volatile, structurally fragile instrument to the entire U.S. economy."

Barclays: The $1.5 Billion "Tail" Is Wagging the Entire "Dog"

Rajadhyaksha directly analogizes the logical structure of the current market panic to that of the SVB period.

In 2023, that "tail" was an equal-weighted regional bank ETF with an asset management scale of about $4 billion — KRE, many of whose constituent stocks are thinly traded. It was the decline of this small ETF that drove the repricing of trillions of dollars in interest rates, investment-grade credit and stock positions, forming a self-reinforcing panic cycle.

The 2026 "tail" is CoreWeave's roughly $1.5 billion in structurally subordinated unsecured bonds.

Rajadhyaksha sorted out CoreWeave's three-tier debt structure: The most important part is the secured delayed-draw term loan (DDTL), held in an independent special purpose vehicle (SPV), and each SPV is backed by specific GPU hardware, customer contracts and data center leases — when Meta signs a multi-year "pay-or-play" contract and assigns it to the SPV, the substantial collateral for the loan is Meta's payment commitment; followed by about $9 billion in convertible bonds with coupons ranging from 1.75% to 2.875%; the bottom layer is the unsecured notes, with tens of billions of dollars in claims queuing ahead during liquidation.

"A 15% yield tells you something about this piece of paper, but it cannot tell you that this company is about to cease operations," Rajadhyaksha wrote, using an analogy: "If a person has a $35 million mortgage plus a small credit card, the high credit card interest rate only indicates a problem with the credit card, not that all mortgage borrowers in this city are in trouble."

The digital comparison is equally intuitive: in March 2023, the market priced in 100 basis points of interest rate cuts due to the decline of a $4 billion ETF; in 2026, the market began to question a trillion-dollar capital expenditure cycle because roughly $1.5 billion in structurally subordinated bonds are trading at 80 cents on the dollar.

Goldman Sachs: The Market Has Chosen the Wrong Reference Frame for Financing Costs

The core argument of Goldman Sachs analysts Gabriela Borges and Maura Hager also focuses on the misinterpretation of the debt structure.

Goldman Sachs pointed out that CoreWeave's main financing instrument is DDTL — an asset-backed project financing secured by specific customer contracts, GPU infrastructure and related cash flows. Lenders underwrite the cash flow quality of specific contracts, not CoreWeave's overall credit status. This structure shifts financing risks from the company's balance sheet to the underlying assets and reduces refinancing risks. The unsecured bond market has been closed to new issuances by CoreWeave for several months, and measuring financing costs based on the yield of this market is equivalent to referring to an invalid price signal.

Based on the annualized interest expense of $2.56 billion in the second quarter divided by total debt of about $30 billion, Goldman Sachs estimates that CoreWeave's current weighted average cost of debt is about 800 to 900 basis points. For new financing, Goldman Sachs assumes that DDTL pricing for investment-grade customers (such as Meta) is about SOFR + 300 basis points, and about SOFR + 600 basis points for high-yield customers, with a mixed cost range of 750 to 1050 basis points. The conservative benchmark is set at 1000 basis points, which means that the new financing cost will only increase by 100 to 200 basis points from the current level.

Goldman Sachs also cited market data: both DDTL 5.0 and 5.5 are trading at prices no lower than the original issuance discount, indicating that the asset-backed financing market's confidence in CoreWeave has not collapsed.

It is worth noting that Goldman Sachs stated in the report's disclosure appendix that it is acting as the lead or co-lead underwriter for a pending underwriting business of CoreWeave, holds more than 1% of CoreWeave's shares, has received investment banking remuneration from CoreWeave in the past 12 months, and expects to continue to seek related business in the next three months.

The Fulcrum and Fragility of Goldman Sachs' Bullish Logic: Revenue Per Megawatt Needs to Rise by 30%

The other half of Goldman Sachs' bullish argument is built on the upward expectation of CoreWeave's computing power pricing capability.

Goldman Sachs estimates that CoreWeave's current mixed monetization level is about $10 million in annual recurring revenue (ARR) per megawatt. The supporting factors for the upside include: short-term contracts (3 to 6 months) have been signed at a price of $40 million per megawatt; CoreWeave raised prices by 25% in July this year and another 10% in October; the profit margin contributed by contracts signed in the second quarter is 5 to 10 percentage points higher than that of recent contracts. Goldman Sachs takes $13 million per megawatt as the "reasonable starting assumption" for newly signed long-term contracts.

Substituting into the example model (5-year term, 200-megawatt contract, 80% debt financing, 10% financing cost, zero residual value), we can get an unleveraged internal rate of return of 15%, forming a positive spread of about 3 percentage points with the weighted average cost of capital of 11.8%.

However, Goldman Sachs' sensitivity analysis also reveals the fragility of this logic: if calculated based on the current actual mixed level of $10 million per megawatt, the unleveraged internal rate of return will drop to 6.3%, and the spread will become negative 6 percentage points under the same financing cost. The entire bullish logic depends on CoreWeave's revenue per megawatt increasing by about 30% on the existing basis.

Goldman Sachs' own model data also reveals how long this path will be: free cash flow will be negative $28.2 billion in both 2026 and 2027, negative $15.8 billion in 2028, long-term debt will rise from about $44 billion this year to about $91 billion in 2028, and free cash flow is expected to turn positive in 2030.

Supply Chain Data: The Reality of Demand Is the Real "Backstop" of the AI Cycle

Rajadhyaksha looks deep into the supply chain, using hard data to refute the narrative of systemic collapse.

TSMC recently reported revenue that exceeded its own upper guidance by $1 billion, setting a profit record for the sixth consecutive quarter, with high-performance computing revenue accounting for 68%, and the company raised its fourth-quarter growth guidance. ASML recorded record sales in the same period, with operating profits more than doubling. CEO Christophe Fouquet said that the planned supply in 2027 and 2028 will not be able to fully meet customer demand, and some customers have already started trying to book equipment for 2029 and 2030.

The five major hyperscale cloud vendors (Microsoft, Meta, Alphabet, Amazon, Oracle) will spend a total of about $7.25 trillion to $8 trillion on AI infrastructure in 2026, and the estimate for 2027 exceeds $1 trillion. CoreWeave's 2026 capital expenditure guidance is about $35 billion to $39 billion, accounting for about 4% to 5% of the above total.

"Even if CoreWeave stops spending entirely, we will only lose 4% to 5% of the entire AI capital expenditure," Rajadhyaksha said. "The spending of hyperscale cloud vendors mainly relies on free cash flow, not the unsecured bond market. Extrapolating from an emerging cloud vendor to one of the most profitable companies in the world, this logical chain is far too long."

Barclays also acknowledges that there is a key difference between this time and 2023: there is no equivalent central bank tool to act as a safety net. In 2023, the Federal Reserve launched the Bank Term Funding Program (BTFP), allowing banks to pledge underwater bonds at par in exchange for liquidity, which became the key to breaking the panic cycle. There is no similar policy tool for AI infrastructure financing. But Rajadhyaksha believes that the real demand of hyperscale cloud vendors for computing power itself constitutes the "backstop logic" of this cycle.

Rajadhyaksha admitted: "The market has a tendency to always find the smallest, most volatile instrument in a complex system, and then treat it as a barometer of the whole. It was KRE in 2023, and CoreWeave's unsecured bonds in 2026. Both signals are real. The extrapolation in 2023 was wrong, and we think the extrapolation in 2026 is equally wrong."

The Real Risk Trigger: Can Computing Power Rent Prices Hold Steady

Although top Wall Street institutions collectively "put out the fire", all parties have clarified a common risk trigger condition.

Rajadhyaksha pointed out that whether the AI capital expenditure cycle repeats the mistakes of the subprime mortgage crisis depends crucially on whether computing power rent prices can be maintained at a level sufficient to cover the cost of capital. "If the rent is not as high as expected, a real crisis may be triggered. CoreWeave is an alarming situation, but it cannot provide evidence of this."

Goldman Sachs' sensitivity analysis also points to this core fragility: if ARR per megawatt cannot rise from the current $10 million to $13 million, the entire positive spread logic will be completely reversed. Once supply catches up with demand and pricing falls, CoreWeave is still locked in data center leases of 7 to 16 years, with a high exit cost.

Brian Stutland of Equity Armor Investments believes that the core of CoreWeave's problem lies in its own business model, not the overall prospect of AI, but he also admits that this is a "two-sided coin" — whether it is an isolated company-level problem or reveals a broader structural vulnerability, there is still a "50-50 probability" at present.

For investors, the current key is not whether CoreWeave will collapse, but whether computing power pricing can support the financing logic of this cycle. The answer to this question will determine the ultimate destination of this panic.

This article is from "Wallstreetcn", author: Xu Chao, authorized for release by 36Kr.