CoreWeave (CRWV): The Business Almost Works. The Balance Sheet Might Not.

Thesis

CoreWeave is not a broken business, and any bear case that starts there is easy to dismiss. Revenue grew 112% year over year to $2.58 billion in the second quarter. Backlog stands at $104 billion, and roughly 93% of the revenue increase came from existing customers expanding usage, which is what real demand looks like. Adjusted operating income was positive at $128 million against expectations of $66 million. Operating cash flow in the first half was roughly $3.7 billion, and it was positive.

The problem is not the business. It is what sits on top of it.

Total debt has gone from roughly $2 billion in 2023 to about $35 billion as of the second quarter of 2026. In the second quarter alone the company raised more than $10 billion across unsecured debt, convertible bonds, a $3.1 billion delayed draw term loan, a $1 billion strategic investment from Jane Street, and an inaugural euro-denominated bond.

Key Metrics

The Interest Bill Is the Loss

This is the single clearest number in the entire story. Net interest expense in the second quarter was $640 million. The net loss for the same quarter was $626 million. The interest bill is, almost exactly, the loss.

Look at the progression by quarter over the past year: $267 million, then $311 million, then $388 million, then $536 million, then $640 million. It has increased every single quarter. Meanwhile GAAP operating income was negative $49 million, essentially breakeven.

So the honest framing is not that the business does not work. It is that the business is roughly at breakeven while the financing structure built to fund it costs $640 million a quarter and is growing. Strip out the debt and this is a company approaching profitability. With the debt, it lost $626 million in three months.

The Liquidity Mismatch

As of the first quarter, current assets were roughly $5.6 billion against current liabilities of roughly $17.8 billion. That is a current ratio near 0.3x, meaning short-term obligations exceed short-term resources by about three times. A ratio meaningfully below 1.0 means the company cannot cover near-term obligations without refinancing, raising capital, or drawing committed facilities.

Free cash flow has been negative every year since 2022, and the gap is widening because capex is enormous. In the first quarter, operating cash flow of $3.0 billion was met with $7.7 billion of capex, producing negative $4.7 billion of free cash flow in a single quarter. Full-year capital expenditure guidance is $35 billion to $39 billion against revenue guidance of $12.4 billion to $13.2 billion.

This is a company spending roughly three times its annual revenue on capital expenditure, funded by debt, while operating at breakeven.

What the Credit Market Is Saying

Equity investors and credit investors currently disagree about this company, and I think the credit market is the one worth listening to.

Corporate credit ratings sit below investment grade at Ba3, B+, and BB-. In the secondary market, CoreWeave bonds have traded at yields in the 11% to 13% range. That is deep high yield pricing. The average cost of debt is roughly 9%.

The reason this matters more now than it did three months ago is rates. Long-dated Treasury yields recently reached their highest levels in close to two decades. For a company rolling $35 billion of debt at 9%, every incremental move higher in the cost of capital is a direct hit to the equity, and it arrives through refinancing rather than through the income statement all at once.

The Backstop Assumption Just Got a Constraint

The most comforting argument in the bull case is that Nvidia will not allow a marquee GPU customer to fail publicly, because the reputational damage would spread through the entire neocloud financing ecosystem it depends on. That argument has always rested on Nvidia’s willingness. It now has to contend with Nvidia’s ability.

In July, Nvidia introduced a financing program offering credit support to AI cloud companies in exchange for a share of their cloud revenue. Critically, it included a provision to rent compute capacity back if those customers could not sell it, which is precisely the demand floor that makes aggressive GPU purchases financeable. According to a Wall Street Journal report in late August, Nvidia paused some of these deals after employees raised concerns that the structure could draw antitrust scrutiny. Nvidia stated that the underlying business model remains in place and continues to evolve, and the program could be revamped or folded into another initiative.

Two things follow. First, a rescue mechanism that attracts regulatory attention is harder to deploy than one that does not, and antitrust exposure is not a problem that can be solved with capital. Second, the reporting indicates Nvidia signaled a preference for capacity distributed among multiple smaller AI firms rather than concentrated with one large customer. CoreWeave is the largest independent neocloud. If that preference holds, it is positioned on the wrong side of it.

This does not eliminate the possibility of Nvidia support, but it makes the mechanism harder to deploy than it appeared a month ago.

The Asset Question

Property and equipment stands at roughly $46.7 billion of a $77.1 billion balance sheet, overwhelmingly GPUs and data center infrastructure. CoreWeave depreciates these on a useful life of roughly six years.

My own view, and this is a judgment rather than a disclosed company figure, is that the real economic depreciation is faster than the accounting schedule, because each Nvidia generation makes the prior one less competitive for frontier workloads. If the true economic life is meaningfully shorter than six years, the collateral backing $35 billion of debt is worth less than book, and it declines every year. The counterargument, which is legitimate, is that CUDA lock-in and a growing secondary market for older GPUs running lower-tier workloads extend economic life beyond what raw compute-per-dollar decay implies.

What I Would Have To Be Wrong About

The bull case is real and I would rather state it plainly than pretend it does not exist.

The backlog is $104 billion and growing, and the expansion is coming from existing customers rather than new logos, which suggests durable demand. The newest financing facilities are structured as project finance, secured against specific contracts and assets, and some carry investment grade ratings on that standalone basis. The weighted average cost of debt has reportedly come down about 300 basis points over the past year. Nvidia retains a strategic interest in the health of its largest customers, though as discussed above, the mechanism it built to express that interest is now under regulatory question. Truist raised its target to $165 from $155 in late August, arguing the shares could nearly double. And the second quarter beat on both lines while adding Anthropic and Meta as customers.

For me to be wrong, capital markets need to stay open to CoreWeave on reasonable terms for another two to three years while revenue compounds fast enough to grow into the debt. That is entirely possible. It is also a bet on financing conditions, not on the company.

How This Could Resolve

These are my own estimates of how a leveraged but strategically important company with real contracted revenue tends to resolve. They are judgments, not market-implied probabilities.

The outcome I consider most likely is a distressed exchange, where creditors extend maturities in return for higher yields or equity conversion. The company survives, there is no bankruptcy filing, and existing equity is heavily diluted. Second is a strategic acquisition, where a hyperscaler acquires assets, contracts, and capacity at distressed prices in a prepackaged or out of court process. Third is an Nvidia-led rescue through bridge financing, additional equity, or outright acquisition, preserving CoreWeave as a captive operator rather than an independent company. I weight that third path lower than I would have a month ago, for the antitrust reasons described above, which shifts probability toward the first two.

In both of the outcomes I consider most likely, the equity is severely impaired even though the business continues.

A Structural Note

Not every AI infrastructure company carries the same risk. Operators backed by government-linked or strategic capital sit in a different position than standalone GPU clouds funding themselves in high yield markets. If the credit cycle tightens, the independent neoclouds are first through the door, and CoreWeave is the largest of them.

Status

This is a Sell and Avoid for me. The stock has already fallen from above $110 in mid-August to roughly $84, so part of this has begun to play out, and I am not claiming to have called that move. My view is that the meaningful risk is still ahead, and it is a 2027 refinancing question rather than a next-quarter earnings question.

The thing I would watch most closely is not revenue or backlog. It is the cost and availability of new debt. As long as CoreWeave can keep raising capital on reasonable terms, this continues. The moment it cannot, the sequence resolves quickly.

Disclosure: I do not hold a position in CRWV as of publication. I may initiate a short position in the coming days.