#美股财报季 #Gate股票畅聊季 CoreWeave’s high-leverage predicament
After the U.S. market closes on Nov 10, CoreWeave will release its third-quarter earnings. The importance of this report lies not in how good it is, but in revealing the conditions under which a business model can fail. As of June 30, CoreWeave had $104 billion in remaining performance obligations, up 246% year over year. In the same 10-Q: second-quarter revenue was $2.575 billion, up 112% year over year; capital expenditures in the first half totaled $14.117 billion, while operating cash flow was only $3.663 billion—a gap of $10.454 billion, against just $5.524 billion in unrestricted cash on hand. The problem is not a lack of orders; it is how those orders are financed.
I. The interest bill: more than tripled in a year
In one year, net interest expense rose from $267 million to guidance of $900 million, nearly three and a half times higher; revenue growth over the same period was 112%. The gap between these two curves captures the shape of CoreWeave’s entire problem. By management’s own guidance, the midpoint of Q3 interest expense is nearly 4 times adjusted operating profit. Even if operations go smoothly and leverage performs as planned, the company still won’t earn enough to cover interest—this isn’t a pessimistic analyst projection; it’s the math in the company’s own guidance.
CoreWeave’s average cost of debt is 8.3%, versus 2.3% for Nebius and 1.7% for IREN—the same business, but financing costs are 3.6 times higher.
The difference isn’t asset quality, but structure: Nebius received $6.958 billion in prepayments from its Microsoft contract, covering 40% of the contract value; customer money arrived upfront. IREN, meanwhile, has its own power and facilities. CoreWeave, by contrast, relies almost entirely on debt financing.
The cost is becoming visible: the $4.2 billion convertible notes issued in September have a conversion price of $97.85, while the closing price on Nov 6 was $46.40—53% below the conversion price.
More than $15 billion must be refinanced within three years, while the AI bond market has effectively shut to new neocloud issuance since September. If it can’t sell debt, it can only turn to equity—the company launched a 35 million-share ATM program on Sept 14, implying about 6% dilution.
II. Customers are being culled, and high investment doesn’t guarantee high returns
It’s true that computing capacity is in short supply. The problem isn’t demand; it’s the customer mix.
Microsoft is narrowing its third-party cloud partners to a handful of core operators. It has already declined to renew Oracle and Nscale, and required some IREN deliveries to be pushed back to February 2027 due to network upgrades. On the earnings call, Nadella mocked this kind of business as “like 4 customers who buy a lot of capacity from you”.
It’s not that customers no longer want to rent computing capacity; they’re no longer willing to pay a premium to intermediaries
First, revenue recognition timing: On Oct 28, Microsoft reported FY27 Q1 capital expenditures of $45.2 billion, below the expected $50.7 billion, with the shortfall explicitly attributed to “third-party capacity delivery delays”; CRWV fell 6% that day, then another 10.4% on Nov 2. Delays don’t mean cancellations, but for a company that burned $10.5 billion in six months, every quarter of delayed revenue recognition proportionally magnifies net losses.
Second, renewal rights: Microsoft accounts for 67% of its 2025 revenue.
Who can survive delays?
IREN pushed its $4 billion annualized revenue target to January 2027, citing customer design changes and supplier delivery timing. It has power, its own facilities, and customer prepayments.
Nebius is even more solid: in August, it confirmed that every phase of the planned capacity at its New Jersey campus would be delivered on schedule. It can survive delivery delays because it isn’t propped up by pure debt.
What CoreWeave lacks is the standing to be seen as a reliable counterparty in the financing market—Moody’s explicitly identifies Oracle and CoreWeave as facing credit stress.
High investment doesn’t guarantee high returns. Full-year capital expenditure guidance is $35–39 billion, about 2.9 times full-year revenue guidance, and interest starts accruing before the assets are built: Nvidia’s CFO acknowledged that a one-gigawatt data center costs about $50 billion and takes 4–6 months from shipment to going online, during which the buyer has already paid in full. CoreWeave has about $9.4 billion in construction in progress, with returns not expected until after 2027. Meanwhile, GPUs have product cycles of just 2–3 years, but CRWV depreciates them over 6 years. The company is betting everything on its estimate of residual value—creditors are pricing its debt at a 16% yield.
III. Rate hikes: not a risk, but an established fact
The Fed isn’t “likely to hike rates”—it has already done so, and is highly likely to do so again. On Sept 16, it hiked by 25 basis points to 3.75%–4.00% (the first hike since July 2023, unanimously approved by all 19 officials); on Oct 28, it held rates steady; CME pricing puts the probability of a cumulative 25-basis-point hike by Dec 9 at 64.1%.
The transmission path is direct: CRWV’s $2.6 billion DDTL 5.5 is priced at Term SOFR + 5.5%, so it moves with the policy rate; existing debt carries coupons of 8.5%–9.75%, while yields in the secondary market have already reached 16%. Every additional $1 billion in financing at 9.6% means $96 million more in annual interest—while the entire second-quarter GAAP operating loss was only $49 million.
More counterintuitively, AI data center construction itself is becoming a reason for rate hikes. The Sept FOMC minutes state that some participants believed “AI infrastructure build-out may cause aggregate demand to exceed supply in the medium term, putting upward pressure on inflation”.
This creates a feedback loop: computing expansion drives up inflation → rate hikes → higher financing costs for computing expansion → expansion is forced to slow → but inflationary pressures on the supply side remain.
Ultimately, the funding structure is wrong (pure debt leverage, while peers hedge with customer prepayments and their own power); bargaining power is eroding (the largest customer accounts for 67% of your revenue while publicly questioning your business model); and the macro environment is working against you (the rate-hiking cycle has just restarted, financing windows are narrowing, and equity financing is diluting shareholders).
The result: CoreWeave must expand as fast as possible to prove its debt is safe. But if the pace of expansion slows, interest and depreciation won’t slow with it. This is a model that allows no mistakes. The order book answers “is there business?” but not “who can afford to pay for this business?” It has proven that demand exists, but not yet that demand can turn into returns.
After the U.S. market closes on Nov 10, CoreWeave will release its third-quarter earnings. The importance of this report lies not in how good it is, but in revealing the conditions under which a business model can fail. As of June 30, CoreWeave had $104 billion in remaining performance obligations, up 246% year over year. In the same 10-Q: second-quarter revenue was $2.575 billion, up 112% year over year; capital expenditures in the first half totaled $14.117 billion, while operating cash flow was only $3.663 billion—a gap of $10.454 billion, against just $5.524 billion in unrestricted cash on hand. The problem is not a lack of orders; it is how those orders are financed.
I. The interest bill: more than tripled in a year
In one year, net interest expense rose from $267 million to guidance of $900 million, nearly three and a half times higher; revenue growth over the same period was 112%. The gap between these two curves captures the shape of CoreWeave’s entire problem. By management’s own guidance, the midpoint of Q3 interest expense is nearly 4 times adjusted operating profit. Even if operations go smoothly and leverage performs as planned, the company still won’t earn enough to cover interest—this isn’t a pessimistic analyst projection; it’s the math in the company’s own guidance.
CoreWeave’s average cost of debt is 8.3%, versus 2.3% for Nebius and 1.7% for IREN—the same business, but financing costs are 3.6 times higher.
The difference isn’t asset quality, but structure: Nebius received $6.958 billion in prepayments from its Microsoft contract, covering 40% of the contract value; customer money arrived upfront. IREN, meanwhile, has its own power and facilities. CoreWeave, by contrast, relies almost entirely on debt financing.
The cost is becoming visible: the $4.2 billion convertible notes issued in September have a conversion price of $97.85, while the closing price on Nov 6 was $46.40—53% below the conversion price.
More than $15 billion must be refinanced within three years, while the AI bond market has effectively shut to new neocloud issuance since September. If it can’t sell debt, it can only turn to equity—the company launched a 35 million-share ATM program on Sept 14, implying about 6% dilution.
II. Customers are being culled, and high investment doesn’t guarantee high returns
It’s true that computing capacity is in short supply. The problem isn’t demand; it’s the customer mix.
Microsoft is narrowing its third-party cloud partners to a handful of core operators. It has already declined to renew Oracle and Nscale, and required some IREN deliveries to be pushed back to February 2027 due to network upgrades. On the earnings call, Nadella mocked this kind of business as “like 4 customers who buy a lot of capacity from you”.
It’s not that customers no longer want to rent computing capacity; they’re no longer willing to pay a premium to intermediaries
First, revenue recognition timing: On Oct 28, Microsoft reported FY27 Q1 capital expenditures of $45.2 billion, below the expected $50.7 billion, with the shortfall explicitly attributed to “third-party capacity delivery delays”; CRWV fell 6% that day, then another 10.4% on Nov 2. Delays don’t mean cancellations, but for a company that burned $10.5 billion in six months, every quarter of delayed revenue recognition proportionally magnifies net losses.
Second, renewal rights: Microsoft accounts for 67% of its 2025 revenue.
Who can survive delays?
IREN pushed its $4 billion annualized revenue target to January 2027, citing customer design changes and supplier delivery timing. It has power, its own facilities, and customer prepayments.
Nebius is even more solid: in August, it confirmed that every phase of the planned capacity at its New Jersey campus would be delivered on schedule. It can survive delivery delays because it isn’t propped up by pure debt.
What CoreWeave lacks is the standing to be seen as a reliable counterparty in the financing market—Moody’s explicitly identifies Oracle and CoreWeave as facing credit stress.
High investment doesn’t guarantee high returns. Full-year capital expenditure guidance is $35–39 billion, about 2.9 times full-year revenue guidance, and interest starts accruing before the assets are built: Nvidia’s CFO acknowledged that a one-gigawatt data center costs about $50 billion and takes 4–6 months from shipment to going online, during which the buyer has already paid in full. CoreWeave has about $9.4 billion in construction in progress, with returns not expected until after 2027. Meanwhile, GPUs have product cycles of just 2–3 years, but CRWV depreciates them over 6 years. The company is betting everything on its estimate of residual value—creditors are pricing its debt at a 16% yield.
III. Rate hikes: not a risk, but an established fact
The Fed isn’t “likely to hike rates”—it has already done so, and is highly likely to do so again. On Sept 16, it hiked by 25 basis points to 3.75%–4.00% (the first hike since July 2023, unanimously approved by all 19 officials); on Oct 28, it held rates steady; CME pricing puts the probability of a cumulative 25-basis-point hike by Dec 9 at 64.1%.
The transmission path is direct: CRWV’s $2.6 billion DDTL 5.5 is priced at Term SOFR + 5.5%, so it moves with the policy rate; existing debt carries coupons of 8.5%–9.75%, while yields in the secondary market have already reached 16%. Every additional $1 billion in financing at 9.6% means $96 million more in annual interest—while the entire second-quarter GAAP operating loss was only $49 million.
More counterintuitively, AI data center construction itself is becoming a reason for rate hikes. The Sept FOMC minutes state that some participants believed “AI infrastructure build-out may cause aggregate demand to exceed supply in the medium term, putting upward pressure on inflation”.
This creates a feedback loop: computing expansion drives up inflation → rate hikes → higher financing costs for computing expansion → expansion is forced to slow → but inflationary pressures on the supply side remain.
Ultimately, the funding structure is wrong (pure debt leverage, while peers hedge with customer prepayments and their own power); bargaining power is eroding (the largest customer accounts for 67% of your revenue while publicly questioning your business model); and the macro environment is working against you (the rate-hiking cycle has just restarted, financing windows are narrowing, and equity financing is diluting shareholders).
The result: CoreWeave must expand as fast as possible to prove its debt is safe. But if the pace of expansion slows, interest and depreciation won’t slow with it. This is a model that allows no mistakes. The order book answers “is there business?” but not “who can afford to pay for this business?” It has proven that demand exists, but not yet that demand can turn into returns.
