Will Coreweave Trigger a Blowup in the US Junk Bond Market? Is Oracle Next?

Cracks in the AI infrastructure financing chain are spreading from the periphery to the core.

CoreWeave bonds are deeply distressed, and as a tenant, it underpins the repayment capacity of over $14 billion in high-yield data center bonds. According to Deutsche Bank’s comprehensive market credit mapping, the divergence in AI credit is accelerating—while AA-rated tech giants still enjoy quasi-sovereign financing costs, the CDS for Oracle, already at the lower end of investment grade, has widened sharply. A downgrade would trigger massive forced selling, pulling this “peripheral crisis” into the heart of the market.

CoreWeave’s bonds have now traded at distressed prices for three consecutive days. Its 9.625%, 2032 maturity notes closed at a yield of around 14%; Reuters reports that the yield on the 2031 notes exceeded 16%—a level typically viewed as a distress signal. The five-year CDS closed at 1,117 basis points, implying a default probability of about 63% within five years, equivalent to Lehman Brothers’ CDS levels right before its bankruptcy filing. Since their issuance in June, the 2032 notes have fallen below 80 cents on the dollar, breaching the conventional definition of a distressed range.

The core issue is not just about CoreWeave itself. CoreWeave is the sole tenant for seven out of the 20 project-financed data center bonds in the high-yield market, involving $14.1 billion in debt and accounting for 31% of the market. This means CoreWeave's capital cost crisis directly impacts the debt repayment capacity of issuers such as Applied Digital, Core Scientific, and Galaxy Digital.

Meanwhile, Deutsche Bank warns that Oracle’s five-year CDS has surged 36 basis points in two days to 288 basis points. Should its BBB rating face a downgrade, institutional investors restricted to holding investment-grade assets would be forced to sell, triggering a systemic domino effect.

AI Exposure in the Junk Bond Market: $14.1 Billion Hangs on CoreWeave

According to Morgan Stanley’s latest monthly report, “AI Financing Update: Pausing, Not Retreating,” at least 20 project-financed data center bonds exist in the high-yield market, amounting to $45.6 billion in debt and financing 4,063 MW of key IT capacity. Each bond is issued by a single-purpose entity, with debt repayment sourced solely from a single tenant’s rental payments.

CoreWeave is the largest tenant among them, and also the weakest link. Of the 20 bonds, CoreWeave occupies seven as tenant, covering $14.1 billion in debt and 1,512 MW capacity, accounting for 31% and 37% of the totals, respectively. This surpasses the Google-backed Fluidstack series ($11.9 billion) and Nvidia ($8.4 billion), making CoreWeave the single largest rent source in the high-yield data center market.

Specific issuers relying on CoreWeave’s rent include:

• Applied Digital (APLD): Two bonds totaling $3.94 billion ($2.35 billion 9.25% 2030 notes and $1.59 billion 7% 2031 notes, both at the Polaris Forge project in Ellendale, North Dakota), with another $2.15 billion Polaris Forge 2 bond tenanted by Oracle;

• Core Scientific (CORZ): $3.3 billion, financing 590 MW of capacity across Texas, Georgia, Oklahoma, and North Carolina. This is the largest single bond for CoreWeave-leased capacity in the tracker;

• Galaxy Digital (GLXY): $3.51 billion 9.875% notes for the Helios campus in Lubbock, Texas, with final delivery due March 2028;

• Edged ($1.3 billion, co-tenanted by CoreWeave and Alibaba), Elk Grove Village ($900 million), and DDC 01 Propco ($1.12 billion, Richmond, Virginia), the last of which priced on September 23—just three weeks before the recent market upheaval.

This is double exposure, not merely circular financing. Take Galaxy bondholders: they bear both the construction risk at Lubbock and the credit risk regarding CoreWeave’s rent payments post-2028.

Market Pricing by Tenant: Spread Differentiation Is Obvious

Morgan Stanley’s spread data makes the market’s assessment clear. As of September 28, the average spread on the seven CoreWeave-tenant bonds was around 400 basis points—twice that of the four Google-backed Fluidstack deals (Cipher, TeraWulf, Flash, Meridian, ~195 bps), and well above Amazon-leased bonds (about 150 bps). With the same asset class, structure, and similar ratings, the only variable is the tenant’s name and whether a hyperscale cloud service provider guarantees the lease.

The bonds with the widest spreads share common characteristics: Zenith Arc (Jane Street tenant, Oklahoma, final delivery October 2028) at 599 bps spread/yield of 11.09%; Galaxy at 559 bps/10.68%; Edged at 498 bps/9.92% (160 bps wider since issuance); and SE Cosmos (leased by SoftBank) at 451 bps. Weak tenants, lengthy construction cycles—or both—are shared risk factors.

Notably, the best-performing CoreWeave tenant bond is the soonest to be delivered: Applied Digital’s original 9.25% 2030 notes, with final delivery October 31, 2026, priced at 104.75 with a 218 bps spread. The closer a project is to energization, the less sensitive the market is to tenant credit; the longer the period to completion, the more weight market attaches to tenant credit.

Additionally, Morgan Stanley’s LTC (loan-to-cost) to yield-to-maturity scatterplot shows that the bonds issued in Q2 and Q3 2026 (peak of the boom) cluster at LTC 85%-95% and yields of 13%-17%—i.e., newly issued bonds have thinner equity cushions and demand higher returns, concentrating risk exposure.

Deutsche Bank’s Market Map: AI Credit Now Shows Three-Tiered Divergence

Deutsche Bank Research charted every bond in both the USD investment-grade and high-yield indices by yield and duration, highlighting AI-related names, to produce the most systematic AI credit panorama to date.

The conclusion: overall, AI financing costs remain under control, but divergence at the far end of the curve is widening.

Deutsche Bank divides AI credit into three tiers:

• AA-rated AI names (Meta, Alphabet, Nvidia, Amazon): Short-term funding costs are about 5%-5.5%—the low end of investment grade—indicating the market still treats them as quasi-sovereign borrowers.

• BBB-rated AI names (Oracle, SpaceX): For durations of 10–12 years, financing costs are around 8%-8.5%, topping the investment-grade range—1.5-2 percentage points higher than other bonds of similar duration, representing a clear “AI tax.”

• High-yield AI names: Mostly BB-rated, with yields from 7%-11%, roughly the middle of the high-yield market.

CoreWeave is the only outlier. Among all high-yield index bonds, yields of 15%-19% are usually unrelated to AI, but CoreWeave is the exception—its BB+ rated unsecured 2032 notes fell below 80 cents only four months after issue, with five-year CDS at 1,118 bps.

Deutsche Bank notes that CoreWeave rents out computing power on 4-6 year “take-or-pay” contracts, with outstanding orders exceeding $100 billion—about two-thirds from investment-grade customers such as Microsoft, Meta, IBM, and Caterpillar. But the bond market evidently cares more about leverage ratios than client lists.

The bank also flags an overlooked structural problem: the depreciation cycle for advanced GPU chips is about 4-6 years—matching contract terms, but the debt that finances them is longer-term.

This means collateral depreciates before the debt matures, which is the fundamental reason for widening spreads on longer-dated bonds. Adding in repricing of Fed rate hike expectations, term premium becomes an added burden for all borrowers needing to refinance.

Oracle: The Next Domino?

Deutsche Bank explicitly highlights Oracle as the next risk node to watch closely.

The bank warns that “such concerns could become self-fulfilling and risk triggering broader issues among other AI names,” calling Oracle a “bellwether”: its five-year CDS soared 36 basis points in two days to 288 bps, with Goldman Sachs data showing a 31 bps one-day rise on Wednesday.

Oracle’s systemic risk comes from its rating position. Currently at BBB—the lowest rung of investment grade. If downgraded, huge numbers of institutional investors restricted to investment-grade assets would be forced to sell. This is not orderly repricing, but a sudden liquidity shock.

In the bond market, the spreads of two Oracle-tenant bonds (Applied Digital’s Polaris Forge 2 and Yondr’s Loudoun County project) are already 185-220 bps wider than Oracle’s own unsecured bonds of the same tenor, with Yondr’s spread briefly widening to nearly 300 bps in early September. Morgan Stanley warned last week that data center delivery delays could severely impair Oracle’s debt situation.

Financing Gap and Divergence of Views

At the heart of CoreWeave’s dilemma is a $35 billion financing gap.

Deutsche Bank analyst Brad Zelnick, in a report titled “Dislocation Creates Opportunity to Buy,” reiterates a Buy rating and $150 price target on CoreWeave, noting $15.5 billion in liquidity at the end of Q2, 60%-65% of backlog from investment-grade customers, and that IRRs on new contracts have jumped from ~15% to “over 20%.” But he also concedes the key issue:

Undoubtedly, CoreWeave will need additional capital over the next 12-18 months to fulfill existing orders. Operating cash flow over 12 months is about $10 billion, while capital requirements are around $45 billion, leaving a $35 billion gap.

If even half this gap must be financed via current 16% unsecured bond yields, additional interest expenses would reach about $2.8 billion per year—exceeding analysts’ forecasts for CoreWeave’s total expected loss of $2.8 billion this year.

Goldman Sachs’ Rich Privorotsky believes the market “is overreacting in pockets,” that “there are reasonable financing constraints, but these do not negate the fundamental demand logic.” Morgan Stanley sees “high-quality high-yield looks attractive on a risk-reward basis once rates stabilize.”

Goldman and Morgan Stanley both preface with ‘high quality’ and ‘if rates stabilize’—but with 10-year US Treasuries still above 5% and rate hike expectations rising, neither condition is met for a BB- shell building in Lubbock whose tenant CDS already prices in default.

Transmission Path: From the Edge to the Center

Combining Morgan Stanley and Deutsche Bank’s reports, the current transmission path for AI credit risk is now quite clear:

The first layer is CoreWeave itself—16% unsecured bond yield, CDS-implied default probability over 60%, and a $35 billion funding gap;

The second layer is the $14.1 billion in high-yield data center bonds reliant on CoreWeave’s rent—issuers such as Galaxy, Core Scientific, Applied Digital, and Edged, with longer construction or delivery periods facing higher risks;

The third layer is Oracle—its BBB rating perched at the edge of investment grade, widening CDS; a downgrade would trigger forced selling and extend the crisis from high-yield into the core of investment grade;

The fourth layer is institutional investors such as life insurers—major buyers of such project finance bonds—now re-examining what it really means for a hyperscale cloud service provider’s name to be listed as a backer in the prospectus.

Deutsche Bank’s conclusion closely matches Goldman’s Privorotsky:

“As the capex cycle turns from free cash flow to debt finance, we must increasingly respect the price signals from the credit markets. Overall AI financing costs are not extreme, but troubling themes have emerged at the edges.”

Credit crises always begin at the periphery. In 2007, subprime was also believed to be ‘controllable’.