$51.6 billion. That is the debt Coreweave (CRWV) currently carries, and it is actually more than the company's market capitalization of $45.8 billion on the same day. To start with my conclusion, my investment recommendation is Hold, and the 12-month baseline target price is $88. Let me explain, step by step, why the financial structure, rather than demand, is the key factor for this stock.
The analysis reference date is September 15, 2026, and the stock price is the closing price of the US market on September 14. The Korean Won amounts in the text were calculated at 1,347.3 Won per Dollar (based on the Seoul foreign exchange market on September 14).
Table of Contents
- What kind of company is Coreweave (CRWV), and why should we look at the financials first?
- Coreweave (CRWV) Recent 5-Quarter Trend — Revenue Doubles, Debt Doubles
- I took a closer look at the $51.6 billion debt.
- A structure where depreciation and interest eat up EBITDA
- Coreweave (CRWV) Valuation — How to Turn a Backlog of $104.2 Billion into Value?
- Coreweave (CRWV) Target Price Scenarios: $88, $150, $45
- Recent Issues and Coreweave (CRWV) Risk — Four Bear Triggers
- Personally, this is a point that bothers me.
- Coreweave (CRWV) Investment Opinion Conclusion
- Frequently Asked Questions
- References
What kind of company is Coreweave (CRWV), and why should we look at the financials first?
First, let me give you my three-sentence conclusion. First, Coreweave's business is currently running very well. Revenue for the second quarter of 2026 was $2.575 billion (approx. 3.4693 trillion KRW), an increase of 112.51 TP3T compared to a year ago, and the backlog—contracts that have been signed but not yet recognized as revenue—has accumulated to $104.2 billion (approx. 140.4 trillion KRW).
Second, however, that growth was supported not by equity but by debt. Total borrowings increased 3.5-fold in just one year, rising from $14.5 billion to $51.6 billion (approximately 69.5 trillion won). Third, therefore, my 12-month baseline target price is $88 (approximately 118,562 won), and my investment rating is Hold (Low confidence).
Coreweave is a company that purchases NVIDIA GPUs (Graphics Processing Units, semiconductors used for AI learning and inference) in bulk, installs them in data centers, and rents out that computing power on an hourly basis. Its major clients are AI companies like OpenAI. In the industry, companies like this are referred to as neoclouds.
There is an analogy that comes to mind whenever I look at this company. It is like a landlord who maximizes bank loans to buy a building and then rents it out. It is a very good business if tenants (customers) are lining up, but the problem is that the value of the building converges to nearly zero after about six years. A new generation of GPUs comes out every 18 months. Therefore, for this stock, rather than the story of demand You need to look at the financial statements first. The picture is captured properly.
Coreweave (CRWV) Recent 5-Quarter Trend — Revenue Doubles, Debt Doubles
When I look at companies like this, I place the quarterly income statement and balance sheet side by side on a single table. Revenue alone looks impressive, but you can only see the real picture when you see what was expanded to generate that revenue right next to it. This is a table I created myself.
| Item (Unit: million dollars) | 2Q2 of 2025 | Q3 2025 | Q4 2025 | Q1 of 2026 | 2Q of 26 |
|---|---|---|---|---|---|
| sales | 1,212 | 1,365 | 1,572 | 2,078 | 2,575 |
| Gross profit margin | 74.2% | 73.0% | 67.6% | 65.5% | 65.9% |
| Depreciation expense | 560 | 630 | 821 | 1,147 | 1,393 |
| Interest expense | 267 | 311 | 388 | 536 | 640 |
| Net profit/loss | -290 | -110 | -452 | -740 | -626 |
| Total borrowings (ending period) | 14,562 | 18,808 | 29,822 | 35,147 | 51,608 |
It also stands out that the gross profit margin dropped by 8.31 points, from 74.21 TP3T to 65.91 TP3T. Gross profit margin is the profit margin calculated by subtracting direct costs from 100 won in sales; this means that the profit for every 100 won sold has fallen from 74 won to 66 won. This appears to be due to the increased proportion of colocation (renting data centers from others) and higher electricity costs.
However, the line where I really stopped is the very bottom line. Total borrowings, which stood at $14.562 billion at the end of June 2025, reached $51.608 billion in just one year. That amounts to borrowing an additional $37 billion—approximately 50 trillion won—over twelve months. Around the same time Oracle (ORCL) also had quarterly capital expenditures that exceeded quarterly revenue., Oracle is different in that it is a company that generates a steady stream of cash from its core business.
I took a closer look at the $51.6 billion debt.
“You can't make a judgment based solely on the statement that "there is a lot of debt." So, I analyzed the composition based on the end of the second quarter of 2026.
| composition | amount | Korean Won Conversion |
|---|---|---|
| long-term borrowings | $27.555 billion | Approximately 37.1 trillion won |
| short-term borrowings | $7.513 billion | Approximately 10.1 trillion won |
| Financial lease liabilities | $16.54 billion | Approximately 22.3 trillion won |
| Total borrowings | $51.608 billion | Approximately 69.5 trillion won |
| Cash and short-term financial assets | $5.539 billion | Approximately 7.5 trillion won |
| Net borrowings | $46.069 billion | Approximately 62.1 trillion won |
To get a sense of just how heavy these numbers are, you only need to look at three ratios. The debt-to-equity ratio is 1,0271 TP3T. This means that for every 1 won of shareholder money, there is 10 won in debt. Dividing net borrowings by the last 12 months' EBITDA results in a ratio of 12.2 times. Generally, a ratio exceeding just 5 times is considered high risk.
The third point worried me the most. The current ratio is 0.455. This means that while the amount owed within one year is $20.917 billion, the assets available for liquidity within one year are only $9.52 billion. Simply put, the next credit card bill is more than double the bank balance, meaning the structure can only keep running by constantly taking out new loans.
Borrowing conditions are not improving either. Coreweave accepted a rate of 5.501 TP/3T points above the base rate (SOFR) when entering into a $2.6 billion loan commitment in August 2026. In the same market, its competitor Nevius raised funds in July at a rate of 2.501 TP/3T points above the SOFR. The 31 TP/3T point difference signals that the bond market views the risks of the two companies completely differently.
Of course, there is good news as well. The company finalized an $8.5 billion GPU-backed loan in March 2026 with an investment-grade credit rating. As this is the first time in the industry that a loan secured by GPUs has received an investment grade, it is proof that the door to financing has not been closed. Original text of the company announcementIf you look at it together, the conditions are explained in more detail.
A structure where depreciation and interest eat up EBITDA
The figure most frequently cited by Coreweave bulls is the EBITDA margin. For the second quarter of 2026, EBITDA was $1.469 billion (approximately 1.9792 trillion KRW), with a margin of 57.01 TP3T. EBITDA refers to earnings before interest, taxes, and depreciation. Looking at this alone, the profitability rivals that of software companies.
However, I believe you shouldn't take this company's EBITDA at face value. There are two reasons for this.
First, depreciation is a real expense. In the second quarter, depreciation expense was $1.393 billion, representing 54.11 TP3T of revenue. Since it was 46.21 TP3T a year ago, its proportion continues to grow. With buildings, you use them for 30 years, so there is room to pass depreciation off as a "book value." However, GPUs come due for replacement every six years. Therefore, this depreciation is akin to a warning that the same amount of money must be spent again every six years.
Let's verify this with the numbers. As of the end of the second quarter, net tangible assets (fixtures) stood at $63.331 billion. Dividing this by six years yields approximately $10.6 billion annually. This calculation suggests that the $841 TP3T of the $12.6 billion 2027 EBITDA estimate I will write about later will disappear due to depreciation. This is precisely why there isn't much left in your hands even if EBITDA increases.
Second, interest is eating into gross profit. The $640 million in interest expenses in the second quarter represents 37.71 TP3T of the $1.696 billion in gross profit for the same quarter. This compares to 29.71 TP3T in the same quarter a year ago. This means the portion of earned profit paid to banks has increased by 81 TP3T points. Annualized, interest alone amounts to approximately $2.6 billion.
So, the operating profit for the second quarter was minus $49 million, and the net loss was $626 million (approximately 843.4 billion won). The interest coverage ratio (operating profit divided by interest expense) was 0.12 times; a ratio below 1 means that the money earned from operations is not even enough to cover the interest. The free cash flow (FCF, the actual cash remaining after deducting capital expenditures from cash earned from operations) for the past 12 months totaled minus $13.655 billion (approximately 18.4 trillion won). Iron Mountain (IRM), a data center REITIf you compare it with companies that have a long asset lifespan like that, you will definitely feel the difference.
Coreweave (CRWV) Valuation — How to Turn a Backlog of $104.2 Billion into Value?
I have only shared the bleak news so far, but we must look at the other side fairly. The backlog stood at $104.2 billion at the end of the second quarter, and the company announced that it secured over $25 billion in new contracts in the third quarter alone. Annual revenue guidance for 2026 is $12.4 billion to $13.2 billion, with capital expenditures ranging from $35 billion to $39 billion. The company projected third-quarter revenue of $3.525 billion on a median basis.
Since it is a loss-making company, I cannot use the P/E ratio (Price-to-Earnings Ratio, how many times the stock price is compared to annual earnings). So, I divided the Enterprise Value (EV, the sum of market capitalization and net debt; the cost of acquiring the entire company including debt) by EBITDA. With the current EV at $91.8 billion, it is 24.2 times based on the last 12 months' EBITDA and approximately 14 times based on the estimated 2026 EBITDA.
The important thing is 2027. Taking into account the rate of backlog exhaustion and the year-end power operation target (over 1.85GW), if we project 2027 revenue at $22.5 billion and an EBITDA margin of 56%, the EBITDA comes out to $12.6 billion. If we divide the current EVs by this, 7.3 timesIt is. If you only look at the numbers, it looks cheap.
Then, what would be the appropriate multiplier? I would say the appropriate multiplier is 10.0 timesWe set it at that level. Data center REITs like Equinix and Digital Realty trade at around 20 times, but those companies have asset lives of 25 to 30 years and investment-grade credit ratings. You can't apply the same multiple to a company whose assets are depreciated every six years and whose net debt is 12 times EBITDA. So, we considered half the level of REITs to be the appropriate benchmark. The current 7.3 times is discounted by 27.11 TP3T compared to the fair value of 10 times.
| Comparison items | Coreweave (CRWV) | Nevius (NBIS) | Data Center REITs (EQIX·DLR) |
|---|---|---|---|
| EV / Sales over the last 12 months | 12.1 times | About 41 times | — |
| EV / EBITDA | 24.2 times | — | About 20 times |
| Financial status | Net borrowings of $46 billion | Net cash | Investment grade |
| New mortgage interest rates | SOFR+5.50%(August 26) | SOFR+2.50%(July 26) | Corporate bonds 3~5% range |
| Useful life of major assets | GPU about 6 years | GPU about 6 years | 25~30 years of building |
Coreweave (CRWV) Target Price Scenarios: $88, $150, $45
There is a catch here. Just because the EV is cheap doesn't mean the stock is cheap. For Coreweave, more than half of the $91.8 billion in EVs—$46 billion—is debt. The shareholder share is just a thin layer sitting on top of that, so if the EV moves 101 TP3T, the stock price fluctuates by more than 201 TP3T. This is why I set a wide range for my scenario.
| scenario | 27 years of sales | EBITDA margin | EV/EBITDA | Expected net borrowings | Target price | Compared to the current price | weight |
|---|---|---|---|---|---|---|---|
| stress | $26 billion | 58% | 11.5 times | $78 billion | $150 | +80.8% | 25% |
| basic | $22.5 billion | 56% | 10.0 times | $76 billion | $88 | +6.0% | 45% |
| Weakness | $19 billion | 53% | 8.5 times | $66 billion | $45 | -45.8% | 30% |
Let me be honest with you. When I ran my model exactly as is, the bullish result was $166 and the bearish result was $34. Since moving just one multiplier by 0.5x caused the result to spike by more than 301 TP3 T, I trimmed the ends slightly inward and set them at $150 and $45. Even so, the upper and lower limits still spread out by more than three times, and this very width is the nature of this stock.
Let me also explain the rationale behind setting the weights at 451 TP3T for base, 251 TP3T for bullish, and 301 TP3T for bearish. Looking solely at the EV multiple, there is a discount of 271 TP3T, so I would have liked to increase the bullish weight. However, due to the structure where net debt exceeds market capitalization and the fact that the GPU residual value has not yet been verified, I could not set the bearish weight lower than the base. Therefore, I raised the bearish weight to 301 TP3T.
The weighted average of the three scenarios yields an expected stock price of $90.6. Compared to the current price of $82.98. Expected return +9.21 TP3TThat is it. For reference, the average target price of 35 Wall Street analysts is $144.46 (approximately 194,631 won), and the median is $150. There is quite a significant difference from my numbers.
I will compare myself to myself as well. When I first covered Coreweave in June, I set the baseline target price at $140.. The reason it dropped to $88 in three months is not due to worsening demand. In the meantime, revenue actually increased by 241 TP3T, and the backlog rose from $99.4 billion to $104.2 billion. What changed is total borrowings. It increased by $16.5 billion in just one quarter, rising from $35.1 billion to $51.6 billion.
Recent Issues and Coreweave (CRWV) Risk — Four Bear Triggers
The stock price dropped 6.81 TP3T in a single day on September 14. This was triggered by Bernstein's warning that CoreWeave is more exposed than its competitors as the center of gravity of AI demand shifts from training to real-time inference. The logic is that outlying data centers optimized for large training clusters struggle to handle the demand for inference, where latency is critical.
Even though Michael Burry stepped forward on the same day to refute the theory that AI is slowing down, AI-related stocks fell together. I believe this scene showed which side the market is currently on in this debate. The four bear triggers I have identified are as follows.
- Decline in GPU residual value → Impairment of collateral value → Violation of loan agreement terms. It is pointed out that since a significant portion of the loans are secured by GPUs, the clause could be triggered around 2027 if used GPU rental rates drop further. The indicator to check is the debt-to-net tangible asset ratio in the quarterly report.
- Rise in funding costs → Surge in interest expenses → Widening net loss. If SOFR+5.50% becomes the new standard, the interest burden will increase. You can check the quarterly interest expense trend in the income statement.
- Customer Focus → Backlog damaged if even one spot is missed. Revenue is concentrated in a few large AI companies. Please check the 10-Q disclosure regarding the proportion of major clients.
- Shift from learning to inference → Decline in utilization → Margin pressure. Whether the gross profit margin drops further below the 65% range is a signal.
On the other hand, there are also reassuring aspects. The fact that the backlog is $104.2 billion means that a significant portion of future revenue is secured by contracts. With deferred revenue (money paid in advance by customers) alone amounting to $9.692 billion (approximately 13.1 trillion won), it effectively means that customers are providing a portion of the working capital. When I looked at IREN, which is also a NeoCloudThe key issue was the time lag between the contract balance and actual revenue recognition, but Coreweave differs in that it bridges that gap with liabilities.
Personally, this is a point that bothers me.
I paused for a long time at one point while reading the second-quarter earnings report. In the same quarter, operating cash flow was $679 million, but $6.422 billion went out for capital expenditures. That is a difference of nearly tenfold. This gap was bridged by $13.457 billion newly borrowed in that quarter. I would like to call this structure a state where growth is subordinate to financing.
One more thing. I created a table to compare the loan interest rates between Coreweave and Nevius, and I felt that a gap of 31 points was the most honest signal. The stock market groups both companies together under the label of "AI infrastructure" and evaluates them based on their stories, but the people actually lending money had already drawn a line between them. Nvidia (NVDA), which sells GPUs, is the side accumulating cash.And the contrast that the side buying it and lending it is the one accumulating debt also kept lingering in my mind.
So, personally, I view this stock not as a "position to bet on AI demand," but as a "position to bet on the AI financing market." Even if demand is strong, shareholders' profits are the first to be cut if interest rates rise or used GPU prices collapse. Conversely, even if demand slows slightly, this stock could rise quite significantly if financing conditions improve. Honestly, that judgment is beyond my capabilities, so I have listed the confidence level as "Low.".
Coreweave (CRWV) Investment Opinion Conclusion
My investment opinion is Hold (Low Confidence), the 12-month basic target price is $88 (approx. 118,562 won)no see.
The process by which I reached this conclusion consisted of three steps. First, I looked at valuation signals; the 7.3x multiple of the 2027 estimated EBITDA was 271 TP3T cheaper than my estimated fair value of 10x, clearly signaling undervaluation. Next, I calculated the expected rate of return, which came out to +9.21 TP3T. By my criteria, this was +121 TP3T or higher, so it fell one notch short of a Buy rating. Finally, I examined financial signals; I determined that the combination of net debt at 12.2x EBITDA and a current ratio of 0.455 represented a risk worth sacrificing one notch in the expected rate of return.
The Wall Street consensus is a Buy with a target price of $144.46. There are two reasons why I took a different view. First, the consensus simply translates the low EV multiple into the stock price, but I believed that the shareholder equity would be thinner than expected after deducting the increasing net debt. Second, my calculation shows that the six-year depreciation of GPUs will consume 841 TP3T of 2027 EBITDA. I viewed this as a recurring expenditure rather than a "temporary expense.".
To summarize, I suggest looking at it this way. If you already hold the stock, tracking just three figures each quarter—net debt, current ratio, and gross profit margin—is sufficient. However, if you are looking to enter the market, I believe the first step is to calculate whether your investment weighting allows you to withstand a decline to $45. The next earnings announcement is scheduled for November.
Frequently Asked Questions
What is the target price for Coreweave?
The baseline 12-month target price is $88 (approximately 118,562 KRW). The target is set at $150 for the bullish scenario and $45 for the bearish scenario. The weighted average of the three scenarios is +9.21 TP3T relative to the current price, so the investment rating has been set to Hold.
Is Coreweave's debt at such a dangerous level?
While the absolute amount is large, the structure is more important. Total borrowings of $51.6 billion exceed the market capitalization of $45.8 billion, and the amount due within one year is 2.2 times the current assets. However, with a contract backlog of $104.2 billion providing support, there is no immediate need to worry about bankruptcy; the key is whether financing can continue.
Isn't an EBITDA margin of 571 TP3 T good?
EBITDA is a figure before depreciation and interest are deducted, so it is easy to overestimate this company. In the second quarter, depreciation expense was 54.11 TP3T of revenue and interest expense was 37.71 TP3T of gross profit, resulting in a net loss of $626 million. Since GPUs reach a replacement cycle of six years, you should not view depreciation solely as a book value.
Why has Coreweave's stock price dropped recently?
The stock fell 6.81 TP3T in a single day on September 14 after Bernstein pointed out that CoreWeave's outlying data centers would be at a disadvantage as AI demand shifts from training to real-time inference. It is currently down 45.81 TP3T from its 52-week high of $153.2. Debates regarding procurement costs for AI infrastructure are also playing a role.
References
- SEC EDGAR — CoreWeave Q2 2026 Earnings Press Release
- CoreWeave IR — Q2 2026 Earnings Announcement
- CoreWeave Q2 2026 Earnings Conference Call Full Text (August 11, 2026)
- Stock Analysis — CRWV Analyst Target Price Consensus
This article was drafted using AI research tools based on public disclosures and data, and I personally verified the numbers and logic one by one before publication. Stock prices, exchange rates, and estimates are based on the time of writing and are subject to change. The purpose is to share information and is not a recommendation to buy or sell specific stocks. Please remember that investment decisions and their consequences are the responsibility of the investor.