CoreWeave: A Thermometer for the AI Market
CoreWeave is a unique player in the U.S. stock market. It doesn’t make GPUs. Instead, it borrows heavily to buy Nvidia chips, then rents them out to major companies like OpenAI, Meta, and Microsoft.
Since becoming a “sublandlord” of AI computing power, CoreWeave has been thriving.
A few years ago, it was just an Ethereum mining company. Since going public last March, it has become a major force in the AI market.
A company with a market cap of $34 billion is seeing bond investors demand yields of nearly 15%, while the five-year default risk implied by its CDS is also surging.
That was the reading CoreWeave gave the U.S. stock market last week.
On October 16, the stock closed at $61.39, down about 25% for the week—its biggest weekly drop in nearly two months.
At the same time, yields on bonds maturing in 2031 briefly topped 16%, while bond prices fell to around 79 cents on the dollar.
The stock was falling. The bonds were falling, too.
1. The bond market is starting to ask: Can AI actually generate a return?
CoreWeave has total debt of about $50 billion.
Five-year CDS briefly rose to around 1,195 basis points.
A CDS with a notional value of $1 million would cost about $119,500 a year in premiums.
Based on this, the market estimates the five-year probability of default at more than 60%. But that is not the actual probability of default; it also depends on model assumptions such as recovery rates.
The risk signal: creditors are demanding greater compensation for taking on risk.
JPMorgan’s central concern is that AI infrastructure requires huge upfront investment, with returns coming in gradually, making it difficult to turn free cash flow positive in the near term.
If demand for computing power continues to grow, debt-fueled expansion may remain sustainable.
But if customer demand, lease prices, or financing conditions change, high leverage could become a risk.
Max Frumes of 9fin called it “the canary in the coal mine.”
That doesn’t mean CoreWeave is certain to run into trouble. Rather, it could be an early warning of risks in the AI credit market.
2. Could banks get caught up in the fallout?
Estimates released by JPMorgan on October 20 show that the potential exposure of major U.S. investment banks and Japan’s Mitsubishi UFJ Financial Group is roughly $1 billion to $3 billion each. Most European and Asian banks have exposure of less than $500 million.
Related loan commitments total about $24 billion, of which around $15 billion has been drawn and about $9 billion remains undrawn.
These figures do not mean that banks will ultimately lose that much money.
Some financing is ring-fenced through special-purpose entities and project assets, and loans are disbursed in stages as projects progress.
JPMorgan described the related financing arrangements as “well managed and very well secured.” In other words, they are managed carefully and backed by ample collateral.
Looking at CoreWeave alone, we cannot conclude that the banking system faces a major risk. But the market may not have a clear picture of how much debt risk has accumulated across the entire AI infrastructure cycle.
3. The bullish-bearish divide is itself a market reading.
On October 19, Goldman Sachs upgraded CoreWeave from Neutral to Buy, with a price target of $106.
Its reasoning: not all new cloud-computing companies are alike, and CoreWeave has advantages in its contract economics and competitive position.
HSBC maintained its Reduce rating, with a price target of $39.
Bernstein also upgraded its rating from Underperform to Market Perform.
For the same company, some see the sharp selloff as an opportunity, while others believe the risks are far from fully priced in.
On October 15, Nvidia CEO Jensen Huang publicly said he was optimistic about CoreWeave’s prospects, adding that its performance was improving every day.
But he did not promise a bailout.
Cathie Wood of ARK also bought CoreWeave shares during the steep decline.
Being bullish on a company’s long-term business does not mean its debt structure is risk-free.
Being bullish on AI demand does not mean every AI infrastructure company is worth buying.
4. Does Tuesday’s rebound mean the risks are gone?
On October 20, CoreWeave closed at $62.13, up 1.65%, and gained another 4.7% or so after hours.
One possible catalyst was SpaceX’s release of its third-quarter results.
Revenue was $14.58 billion, up 161% year over year and above market expectations.
AI business revenue was about $7.56 billion, up 830% year over year and exceeding Starlink for the first time.
More notably, of the roughly $19.6 billion SpaceX spent on capital expenditures that quarter, $16.44 billion went toward AI infrastructure.
These figures show that demand for AI computing power and investment in the sector remain enormous.
SpaceX’s growth does not directly prove that CoreWeave’s business model is safe, much less offset its debt risk.
If you follow U.S. stocks, AI, or CoreWeave, three things will matter going forward:
1️⃣ Whether yields on the 2031 bonds can fall back below 15% and stabilize.
2️⃣ Whether CDS spreads continue to widen or begin to narrow.
3️⃣ Tesla’s and Meta’s earnings reports—especially Meta’s AI capital spending and cash flow.
The market needs to look not only at how much AI companies are willing to spend, but also at whether those investments can be sustainably supported by their own businesses.
CoreWeave is a thermometer. A thermometer measures temperature, not illness—the illness lies in the narrative around AI capital spending.
CoreWeave is a unique player in the U.S. stock market. It doesn’t make GPUs. Instead, it borrows heavily to buy Nvidia chips, then rents them out to major companies like OpenAI, Meta, and Microsoft.
Since becoming a “sublandlord” of AI computing power, CoreWeave has been thriving.
A few years ago, it was just an Ethereum mining company. Since going public last March, it has become a major force in the AI market.
A company with a market cap of $34 billion is seeing bond investors demand yields of nearly 15%, while the five-year default risk implied by its CDS is also surging.
That was the reading CoreWeave gave the U.S. stock market last week.
On October 16, the stock closed at $61.39, down about 25% for the week—its biggest weekly drop in nearly two months.
At the same time, yields on bonds maturing in 2031 briefly topped 16%, while bond prices fell to around 79 cents on the dollar.
The stock was falling. The bonds were falling, too.
1. The bond market is starting to ask: Can AI actually generate a return?
CoreWeave has total debt of about $50 billion.
Five-year CDS briefly rose to around 1,195 basis points.
A CDS with a notional value of $1 million would cost about $119,500 a year in premiums.
Based on this, the market estimates the five-year probability of default at more than 60%. But that is not the actual probability of default; it also depends on model assumptions such as recovery rates.
The risk signal: creditors are demanding greater compensation for taking on risk.
JPMorgan’s central concern is that AI infrastructure requires huge upfront investment, with returns coming in gradually, making it difficult to turn free cash flow positive in the near term.
If demand for computing power continues to grow, debt-fueled expansion may remain sustainable.
But if customer demand, lease prices, or financing conditions change, high leverage could become a risk.
Max Frumes of 9fin called it “the canary in the coal mine.”
That doesn’t mean CoreWeave is certain to run into trouble. Rather, it could be an early warning of risks in the AI credit market.
2. Could banks get caught up in the fallout?
Estimates released by JPMorgan on October 20 show that the potential exposure of major U.S. investment banks and Japan’s Mitsubishi UFJ Financial Group is roughly $1 billion to $3 billion each. Most European and Asian banks have exposure of less than $500 million.
Related loan commitments total about $24 billion, of which around $15 billion has been drawn and about $9 billion remains undrawn.
These figures do not mean that banks will ultimately lose that much money.
Some financing is ring-fenced through special-purpose entities and project assets, and loans are disbursed in stages as projects progress.
JPMorgan described the related financing arrangements as “well managed and very well secured.” In other words, they are managed carefully and backed by ample collateral.
Looking at CoreWeave alone, we cannot conclude that the banking system faces a major risk. But the market may not have a clear picture of how much debt risk has accumulated across the entire AI infrastructure cycle.
3. The bullish-bearish divide is itself a market reading.
On October 19, Goldman Sachs upgraded CoreWeave from Neutral to Buy, with a price target of $106.
Its reasoning: not all new cloud-computing companies are alike, and CoreWeave has advantages in its contract economics and competitive position.
HSBC maintained its Reduce rating, with a price target of $39.
Bernstein also upgraded its rating from Underperform to Market Perform.
For the same company, some see the sharp selloff as an opportunity, while others believe the risks are far from fully priced in.
On October 15, Nvidia CEO Jensen Huang publicly said he was optimistic about CoreWeave’s prospects, adding that its performance was improving every day.
But he did not promise a bailout.
Cathie Wood of ARK also bought CoreWeave shares during the steep decline.
Being bullish on a company’s long-term business does not mean its debt structure is risk-free.
Being bullish on AI demand does not mean every AI infrastructure company is worth buying.
4. Does Tuesday’s rebound mean the risks are gone?
On October 20, CoreWeave closed at $62.13, up 1.65%, and gained another 4.7% or so after hours.
One possible catalyst was SpaceX’s release of its third-quarter results.
Revenue was $14.58 billion, up 161% year over year and above market expectations.
AI business revenue was about $7.56 billion, up 830% year over year and exceeding Starlink for the first time.
More notably, of the roughly $19.6 billion SpaceX spent on capital expenditures that quarter, $16.44 billion went toward AI infrastructure.
These figures show that demand for AI computing power and investment in the sector remain enormous.
SpaceX’s growth does not directly prove that CoreWeave’s business model is safe, much less offset its debt risk.
If you follow U.S. stocks, AI, or CoreWeave, three things will matter going forward:
1️⃣ Whether yields on the 2031 bonds can fall back below 15% and stabilize.
2️⃣ Whether CDS spreads continue to widen or begin to narrow.
3️⃣ Tesla’s and Meta’s earnings reports—especially Meta’s AI capital spending and cash flow.
The market needs to look not only at how much AI companies are willing to spend, but also at whether those investments can be sustainably supported by their own businesses.
CoreWeave is a thermometer. A thermometer measures temperature, not illness—the illness lies in the narrative around AI capital spending.