Coreweave (CRWV): A Bomb Ticking

CoreWeave: The AI darling that's up 90% since IPO... but drowning in $14B debt. Is this a ticking time bomb or your next 10x stock?

What does CRWV do?

The simplest way to explain what CRWV does is the following:

I am a company that builds AI (that includes language models, image models, agents, you name it). What do I need to build this?  – data centers and compute power, like a lot of them, with a lot of Nvidia GPUs inside. 

Not only do I need a lot of data centers and power, but I also need them NOW. If I wait to build such infrastructure myself, I have to spend so much on capex and by the time I make it, it will be too late. So, I turn to Coreweave (CRWV). I can also turn to Nebius (NBIS), Iren (IREN), and Lambda, but CRWV is the biggest of them all, and this article (video) is about them. All these companies are called “Neoclouds”, and they are probably the ultimate AI stocks, providing a direct exposure to the AI theme.

Anyway, CRWV are like: “Here, these are my data centers, these are my Nvidia GPUs, you can rent them and use them for computing. I can also provide you with some cooling, software tools”, but the main product here is the GPUs.

And this is basically how CRWV makes money, and people also call it Infrastructure-as-a-Service (IaaS).

Bull case – current demand exceeding supply

The main bull case for CRWV is the fact that demand for their services far exceeds the supply at this stage, and this is reflected in their stellar growth of revenue.

Source: Company Presentation

For the financial year 2025, the company is expected to generate roughly $5.1 billion of revenue (or 161% more from 2024). And not just that, CRWV currently has a backlog (contractual obligations, not yet recognized as a revenue) of $55.6 billion (over 10 times the revenue they will generate this year), 40% of which will be recognized as a revenue in the coming two years. That means, if there is perfect execution and no complications, we will see a total of $22 billion of revenue in 2026 and 2027.  This number may actually be a conservative estimate, as CRWV may strike deals that can further push the backlog up.

Source: Company Presentation

This fact didn’t remain unnoticed by the investors. After the IPO earlier this year, the stock has returned over 90%, a mouthwatering performance, even after the 40% correction in the past month.

Source: TradingView

Other bull cases that contributed to the rally are the fact that they are very close to Nvidia. If Nvidia has new chips, Jensen will send them the new product right away – a real VIP client. Also, CRWV computing performance is perhaps better than other general cloud providers, as they have experience with intensive computing tasks – in fact, as a company, they started as a crypto miner back in 2017, and they were able to transfer their expertise in high-level computing into AI.

What about the other side?

Every story has a flipside; the CRWV story also. As we mentioned, the 40% correction, behind the impressive growth, there are several red flags that make investors anxious. So, what are the red flags?

Capex-led growth

In the last 5-6 quarters, CRWV's gross margin has been around 70-75%, which is high, perhaps implying solid profitability and cash generation. But...the reality is different.

70-75% gross margin implies cost-of-goods-sold (COGS) to be just 25-30% of the total revenue. However, this cost does not include the cost they pay for the GPUs. Think as if you are a car reseller. You buy a car for $40,000 and sell it for $60,000, but you don’t record the cost in the income statement. This makes the profit look deceptively high.

In fact, for CRWV, the GPUs are recorded in the cash flow statement as a capital expenditure. Even if the gross profit is high, in accounting terms, the actual cash flows that end up within the firm are much smaller. 

And even if the company scales up in the future, they still need to spend a lot on chips, either when building new data centers or to replace the old GPUs with the existing ones, making the revenue very closely tied to capex.

This is the exact opposite of, let’s say, software firms, where the costs are rather fixed in nature and as the software product revenue scales up, costs remain relatively stable while the profitability skyrockets.

Margin Vulnerability

Another peculiar aspect is the fact that the cost of goods sold (COGS) does not even include the depreciation of the GPUs. There is already plenty of discussion regarding the depreciation timeline, but we won’t touch this topic now. What we want to mention is that neither Alphabet nor Oracle records GPU depreciation the way CRWV does, as they put it within COGS. Fishy, right?

So, how come CRWV's gross margins have been stagnant for quite a while at around 75%? Actually, CRWV's gross margin is quite unlikely to go up just because a lot of the COGS would go for rent for the data centers, as they don’t own them. The landlords of these properties (such as Equinix) have way more bargaining power and can easily re-negotiate higher rents as the fight for data center properties becomes more intense. This means CRWV is highly likely to pay more on rent and will make the gross margin go down. On the other hand, it is hard for CRWV to negotiate higher prices for their services with clients, so to put it simply, CRWV will be squeezed between the landlords and the AI scalers. 

Debt burden

So, okay, CRWV needs capital to grow, but where will the capital come from? As of September, CRWV has $1.89 billion in cash and cash equivalents, $597 million in restricted short-term deposits and $478 million in restricted long-term deposits. That makes a total amount of around $3 billion. As their contractual obligations are around $55 billion, they need to build a crazy amount of data centers, and $3 billion of the current cash they have will be far from sufficient.

Some may say: “Borrow money! Banks and investors will be happy to finance this great AI story”. I would say: “Sure, but they have already borrowed a lot”. Currently, CRWV has $3.7 billion in short-term debt and $10.3 billion in long-term debt (or a total of $14 billion) owed to banks and investors.

These borrowings incur interest expense, and the last thing you want when you desperately have to expand your business is to spend money on heavy interest payments. In the first 9 months of 2025, CRWV paid $840 million in interest, compared to $212 million in the same period last year. Or in other words, 24% of the revenue generated in 2025 so far has gone towards interest payments. If they borrow more, the future loans would be with a higher interest rate, as lenders would see them as riskier, and hence, the interest expenses would further go up. That is the interest only, CRWV needs to repay the maturity of a bond worth $4 4billion in 2028.  

To add to the picture, CRWV leases a significant portion of their data centers, and that means in the future it will owe money to the data center operators. Currently, the lease liabilities are $4.7 billion.  

In addition, the debt we are talking about here might be just the tip of the iceberg. Through complex investment vehicles, CRWV may actually hide a lot more debt away from the balance sheet (perhaps as big as the one in the official balance sheet), and this may backfire if things go sour.

But even if they don’t go the debt-financing path, CRWV may choose to raise money through equity, which will dilute the share count and harm the current investors, as their proportional ownership in the company will go down.

Clients and Suppliers

The clients and the suppliers of CRWV also need special attention. In 2024, the top two customers of CRWV accounted for 77% of the revenue. 62% of the revenue came from Microsoft, but the final beneficiary is OpenAI. In 2025 Q3 earnings, CRWV management proudly announced how they are successfully diversifying their customer base, with the largest backlog customer being just 35%.

Why is customer diversification such a big deal, and why doesn’t CRWV just sell to OpenAI directly? It’s simple, the reason why CRWV avoids concentration in one client is the fact that this won’t be liked by debtors, and they will demand higher interest when CRWV wants to borrow more, especially when your largest client (OpenAI) still doesn’t have a solid revenue track but already amassed over a trillion dollars of future business commitments.

However, this well-celebrated decrease in single-customer reliance might be illusional as well. Google has been another major customer, but what Google essentially does is buy computing power from CRWV at a lower price and sell it to OpenAI at a higher price, so even though we see diversification of revenue, the ultimate beneficiary remains the same, which means no significant risk diversification.

How about the supplier side? Almost the entire cost related to the GPUs is actually going to Nvidia. But the more interesting thing is that Nvidia is also a customer, buying CRWV capacity for internal purposes (15% of the 2024 CRWV revenue is from Nvidia). Not only that, but they also own 5% share in CRWV, which makes Nvidia a supplier, customer and shareholder of CRWV – a very bizarre situation that resembles the circular transactions from the dot com bubble.  If a negative catalyst comes their way, this can create a lethal chain of reactions.

Source: Reuters

Credit Default Swaps of CRWV

To demonstrate the complexity of CRWV debt situation, we should take a look at how credit analysts view the stock.  Credit default swaps are financial derivatives that function like insurance against the default of a borrower's debt (e.g., bonds or loans). Spreads reflect market-perceived default risk: Low spreads (e.g., 50 bps) indicate low risk; high spreads (e.g., 700+ bps) signal elevated concerns.

Key CDS Metrics for CRWV (as of early December 2025) Based on recent market data:

Basically, there is 40-47% chance CRWV to default within five years and 50-60% for this to happen in 10 years.

Worth noting that the probability has gone significantly up in the last 1-2 months.

Conclusion

Is CRWV doomed? It is hard to say, but what we know is that the company is in a very complex situation and in order to survive, the demand for AI computing should remain at the current high levels, and there are no delays in GPU deliveries nor data center construction.