According to Woofun AI, the CoreWeave (CRWV.US) credit default swap (CDS) price fluctuates sharply, with an implied default probability exceeding 60%. This market signal does not directly point to the bursting of the AI industry bubble, but rather provides a key revaluation anchor for the on-chain core income asset USD.AI and its token susDAI. The debt financing structure of CoreWeave (CRWV.US) is highly synonymous with USD.AI's lending model in the underlying logic. The difference is only the counterparty's credit quality: the former relies on giants such as Meta (META.US) and OpenAI, while the latter mainly targets small and medium-sized neocloud vendors.
This structural similarity means that traditional financial markets' risk pricing on CoreWeave (CRWV.US) is redefining the true risk premium for on-chain GPU credit through mapping effects. An in-depth analysis of this transmission mechanism is the key to understanding the current valuation split in the Web3 credit market.
CDS is an insurance tool that measures the risk of debt default. The price is measured in base points, and 100 basis points correspond to an annual premium rate of 1%. CoreWeave (CRWV.US)'s 5-year CDS peaked at 1194.5 basis points in the intraday session on October 15, meaning investors holding its bonds were required to pay an annual premium of 11.945% to hedge against the risk of default.
This figure is nearly 20 times the CDS Index for investment-grade companies (approximately 60 basis points). Based on this, the market introduced an implied probability of default of 60%, but this is not a direct probability of bankruptcy, but rather a risk-neutral probability based on a specific recovery rate assumption.
Notably, the indicator is dynamic, and as of October 19, the CDS price had fallen back to 1,082 basis points. This fluctuation reveals the market's concerns about CoreWeave (CRWV.US)'s short-term liquidity pressure, rather than a final verdict on its long-term viability. Understanding the nature of this data is a prerequisite for removing market noise and understanding real risk stratification.
CoreWeave (CRWV.US)'s total liabilities as of the first half of 2026 were approximately US$52 billion, of which total debt was approximately US$35 billion, operating leases were approximately US$16.3 billion, shareholders' equity was only about US$5 billion, and the leverage ratio was as high as 10 times. Its $35 billion debt structure is clearly hierarchical. The first tier is DDTL (Delayed Withdrawal Term Loan), which accounts for the largest share. Approximately $15 billion has been withdrawn and approximately $9 billion has not been withdrawn.
The core risk control logic of DDTL lies in bankruptcy isolation: the capital is not directly loaned to CoreWeave (CRWV.US) parent company, but is injected into an independent project company (SPV), and each SPV binds specific GPU assets, customer contracts, and data center leases. Bank evaluations focus on the credit of contractual counterparties (such as Meta (META.US)) rather than CoreWeave (CRWV.US) itself. The second tier is convertible bonds, with a size of about $11 billion, with coupon interest between 1.75% and 2.875%. The third tier is unsecured notes, which have caused market panic. The scale is about 1.5 billion US dollars. They are structured subordinated debts, which rank after tens of billions of dollars of secured claims in the liquidation order.
This structure suggests that the CDS spike mainly reflects the risk of the most subprime debt rather than the entire debt system.
Barclays (BCS.US) analyst Rajadhyaksha graphically pointed out that the current market turmoil is like questioning the solvency of mortgage borrowers across the city due to high interest rates on small credit cards, which is a mismatch of reference systems. Based on this logic, Goldman Sachs (GS.US) upgraded the CoreWeave (CRWV.US) rating from 'neutral' to 'buy' on October 19, setting a price target of $106 (the stock price was $61.39 at the time). Goldman Sachs (GS.US) revisions are based on the fact that CoreWeave (CRWV.US)'s true weighted financing costs should be based on second-quarter annualized interest expenses of $2.56 billion divided by about $30 billion in total debt. The result is about 800 to 900 basis points, far less than 15% implied by market panic. The 15% yield only applies to that $1.5 billion subprime note and does not represent CoreWeave (CRWV.US)'s overall financing costs.
This revision of institutional views provides an important benchmark reference for reevaluating similarly structured on-chain credit products.
Comparing the DDTL model of CoreWeave (CRWV.US) with the loan structure of USD.AI, it can be found that the two are highly consistent in terms of risk control logic, and both follow the principle of “not looking at the creditworthiness of the borrower or the credit of the counterparty in the contract.” USD.AI divides loans into three tiers: Tier 1 requires the purchaser to be investment grade, with a yield of 7-9%; Tier 2 is non-investment grade but has a multi-year contract with a yield of 10-12%; and Tier 3 is spot scattered leasing with a yield of 12-15%.
This classification essentially encodes DDTL's risk control logic. The key difference, however, is the quality of the opponent. CoreWeave (CRWV.US) tenants include hyperscale cloud vendors such as Meta (META.US), OpenAI, and Microsoft (MSFT.US), while USD.AI publicly disclosed that most of the borrowers were small to medium neocloud vendors such as QumulusAI, Corvex, Hydra Host, Duos Edge AI, and Sharon AI. J.P. Morgan Chase (JPM.US) made it clear in its October 20 report that “not all project companies have hyperscale cloud vendor endorsements,” and this risk is even more significant in USD.AI's asset portfolio. As a result, USD.AI is not more secure because it is collateralized, but rather more risky because of the lower quality of the contract.
A deeper systemic risk is that USD.AI's 13 borrowers, although seemingly scattered, are all exposed to the same GPU rental curve. This isn't 13 separate risks; it's 1 systemic risk multiplied by 13. Once the rent curve declines, it may trigger centralized defaults, which in turn will cause a large number of GPUs to pour into the second-hand market, causing the liquidation price to collapse, creating a vicious cycle of “default → falling liquidation price → insufficient collateral → more defaults.” Currently, the GPU rental market is clearly divided: on the contract side, CoreWeave (CRWV.US) raised the price of the entire product line by about 25% in July, reflecting tight supply and demand in the long-term market; on the spot side, the H100 hourly rent fell 63% from $3.5 in early 2025 to $1.29 in early 2026; and the B200 hourly rent fell from $6.1 in early June to $4.37 in July, down 31% in three weeks. Since collateral settlement can only be carried out at spot prices, this rift forms the weakest link in the entire credit structure. Assuming that the liquidation price is 30% of the spot price, this circular effect will be further amplified.
According to data compiled by Woofun AI, as of October 17, USD.AI had a total deposit of US$574.5 million, 15 active loans totaling US$310.7 million, a weighted average loan interest rate of 10.46%, and a net APY paid to depositors of 6.25%. The 4.2 point spread between 10.46% on the asset side and 6.25% on the debt side is mainly due to the difference in capital deployment rates. Only $310.7 million (deployment rate of 54.1%) was invested in interest-bearing loans. Of the remaining $263.8 million, some stayed in Wilmington Trust escrow accounts awaiting hardware installation verification (up to 7% of the official caliber), and some were invested as PYUSD reserves in 5-year US bonds (yield 4.99%). After calculating 10.46% × 54.1% + 7% × 45.9%, gross profit of 8.87% was obtained, which matched 6.25% of net income after deducting about 2.6% of agreement fees and operating costs.
This means that more than one-third of the 6.25% return earned by depositors comes from bridge interest on 'undeployed fund' rather than a true GPU credit risk premium. As the deployment rate rises to a level corresponding to the official target of 12.4% return, risk exposure will nearly double, while revenue growth is limited. This structural characteristic is worth being wary of.
On the credit loss path, USD.AI has set up three buffering mechanisms: first, to hold first-line guarantee claims and have recourse authority to specific assets within the project company; second, borrowers need to invest 20-25% of their own capital and establish 10% debt repayment reserves and residual value insurance. The collateral needs to drop by more than a quarter before it affects depositors; third, loans use linear amortization, and the loan balance for the 12th month is reduced to 50%. Sensitivity analysis showed that even assuming a 50% default rate, a single loss of 20% (50% of the loan balance minus 30% of the settlement price), the total loss was approximately $31.07 million, or 5.41% of the total deposit. This indicates that the credit loss threshold is higher.
However, the risk of liquidity crushing is more hidden. SusDai is not a stablecoin; its net value fluctuates, and redemptions use a 30-day cycle, first-come-first-served queuing mechanism. This structure may evolve into bank overruns during times of stress: expectations of falling net worth trigger discounts in the secondary market, discounts exacerbate redemption fears, and the queue extends to 60 days or more, creating a self-reinforcing negative feedback cycle. In this process, there is no need for any loans to actually default; simply running out of liquidity can cause SusDai to be drastically discounted.
Derivatives strategies further amplify this risk. On the Pendle platform, SusDai's PT (principal token) transaction yield is often higher than 9%. Note, however, that PT's expiring promise returns 1 susDAI, not 1 dollar. Buy PT at a discount of $0.913, and if the net value of susDai is still 1.00 at the end of the day, you will achieve a 9.5% gain; if the net value falls to 0.913, it will be just flat; if it falls below this value, you will lose money.
As a result, 9.5% of fixed income can only absorb 8.7% of the net worth downside risk, and PT lacks liquidity before it expires, so it is impossible to withdraw in time. On the Morpho platform, revolving collateral strategies using susDAI as collateral (such as Frax usDAI Ecological Vault, Bitwise RWA Treasury) had the highest annualized yield of 27.13%. The arithmetic logic is: susDai's profit of 6.25% is multiplied by 5x leverage, minus 5% of borrowing costs multiplied by 4 times, to obtain a net return of 11.25%.
However, in the LLTV range of 86% — 91.5%, the risk increased simultaneously. Re7 Labs caused $27.4 million in bad debts due to the de-anchoring of xUSD and DeUSD in the November 2025 Stream Finance incident. The lesson was that it did not recognize the circular collateral structure and delayed price feedback. SusDai also faces this risk: market discounts can trigger liquidation without actual credit loss. The revolving strategy on Morpho is essentially betting that SusDai won't experience a liquidity crisis, rather than a bet that borrowers won't default.
In summary, the economy of holding susDAI is essentially selling credit insurance to a basket of small to medium neocloud vendors. They usually charge 6.25% of the premium (of which only 5.66 points actually come from credit; the rest is bridge interest), and in extreme cases they bear losses. Compared with the CoreWeave (CRWV.US) CDS seller charging an 11.945% premium, susDAI holders' risk compensation is clearly insufficient, and the counterparty quality is weaker.
Despite buffers such as 20% own capital and residual value insurance, the 10 percentage point spread revealed a serious split between open market and on-chain credit pricing. The reason this split has been around for a long time is because on-chain GPU credit has yet to go through a full credit cycle test. Since USD.AI issued its first loan of $620,000 in July 2025, it has taken 15 months without any default and only one early repayment. The market lacks real data on the liquidation price of next-generation GPUs such as the B300 in the second-hand market, causing pricing to be based on untested assumptions.
The probability of credit loss in the short term (3-6 months) is low, but the trade-off risk may be realized by expectations; the medium term (6-18 months) is the real stress testing window, and the first half of 2027 is expected to face the triple combined risk of rising deployment rates, late amortization of loans (collateral values falling to 31% and 21%), and a downward rent curve. Investors are advised to pay attention to the four major indicators of deployment rate, weighted loan interest rate, SusDAI secondary market discount, and changes in active loan balances.
If participating, Morpho's revolving leverage should be limited to no more than 3 times (net worth falls 22.5% before clearing). Pendle only allocates short-term PT, treats susDai as an equity position rather than a cash position, and uses a dumbbell strategy to manage risk. CoreWeave (CRWV.US)'s CDS fluctuation provided an open price signal for GPU mortgage credit for the first time, prompting the industry to re-examine the collateral rate (recommended reduction to 60-70%) and amortization period (recommended reduction to 24-30 months) to cope with potential valuation restructuring.