“Yikes.” That was my reaction to Wednesday’s US CPI report, which showed inflation across the world’s largest economy ran far hotter than expected last month.
“Yikes” was also Elon Musk’s reply to SPAC king Chamath Palihapitiya, who spent part of his Wednesday ogling a chart of CoreWeave’s beleaguered 2032s, which traded below 78 cents on the dollar midweek.
CoreWeave’s a controversial stock, to put it mildly. Along with Oracle, it’s the poster child for debt-funded data center buildouts, but unlike Oracle, it’s not too big to fail, it lacks Oracle’s connections in Trumpland and to the extent it has revenue streams other than renting out access to compute, they aren’t material.
There are the bonds. As you can see, they’re burning and now yield in excess of 15%.
The fundamental bear case is simple enough: CoreWeave’s borrowing heavily at increasingly punitive rates to fund its expansion, which means it’s just a giant, leveraged bet on AI demand growth.
If assumptions about AI demand prove overly optimistic, or even if it just turns out that the vast majority of AI-related compute needs can be met with less sophisticated equipment than what CoreWeave rents out, the business model could run into trouble.
As we’ve seen time and again this year, merely restating that bear case is often enough to send the stock tumbling. On Wednesday, it plunged 17% intraday, a move CEO Michael Intrator described as “irrational” in an internal memo.
The selloff counted among the five largest in CoreWeave’s short history as a public company.
In the same company e-mail mentioned above, Intrator said CoreWeave’s confident in its funding and insisted the backlog “more than supports” its debt obligations.
Not helping matters on Wednesday was CoreWeave bear Gil Luria, who showed up on Bloomberg TV to accuse the company of “destroying value at every turn.” Separately, he said the shares could go to zero in the event “the AI cycle sputters.”
There’s the clip. “CoreWeave’s gonna be the first to go,” Luria told The Prof G podcast, of a hypothetical Wile E. Coyote moment for AI demand. “The equity value’s gonna be zero.”
That’s not what you want to hear. The other two publicly-traded neoclouds fell sharply in sympathy, and Oracle was on track for its worst day in months, down more than 9%.
This was insult to injury for all things AI-related coming as it did amid a snowballing correction for semis. The SOX fell a sixth day and is now down more than 11% since last Tuesday.
Oh, and a report from advisory firm Gartner suggested eight out of 10 neoclouds could disappear within four years. “Neoclouds will be among the most exposed to the imbalance between revenue and capital spending,” a report dated Wednesday warned.
While the hyper-scalers “have massive and highly diversified revenue sources that could plausibly sustain AI investments for years,” the neoclouds “must soon earn returns” if they want to stay in business.
By 2030, the report went on, “cost pressure and a lack of a differentiating business model will force over 80% of neocloud providers to cease independent operations.”
CoreWeave rolling over would be quite a spectacle. I don’t know that it would amount to a “black swan” event, but it would certainly hurt the AI narrative at a time when it is already facing increased scrutiny over further investment. If the government were to refuse help — and remember, AI is a big part of their plan to boost GDP and reduce the debt — wouldn’t Microsoft or Meta have to step-in just to keep their own build-outs running smoothly?
I suspect one of the big boys will step in to shore them up by offering financing themselves in exchange for very favorable terms. At the end of the day, CoreWeave’s liabilities are a drop in the bucket for the hyperscalers and are at least backed by hard assets (even if those assets don’t end up holding their full value at today’s prices).
They might even loop in Papa Trump as the super intelligence dealmaker to stroke his ego. It’s a win-win-win!
Yeah, the risk of getting too bearish on this name is, among other things, Nvidia’s thinking. Huang’s in for $2 billion at $87.20. CoreWeave’s a major strategic partner of the most important company on Earth. Microsoft and Meta are obviously huge customers and Google has a multi-faceted partnership with them. So… could it go to zero? Sure, but I gotta think one or more of the big guns would just buy it before then. If not, they’d basically be saying the capacity they’ve contracted for isn’t worth anything, which seems a little odd. I mean what does CoreWeave actually have? Like $50 billion in debt and lease liabilities? The hyper-scalers can sell IG debt at a clip of $25 billion any time they want at still-reasonable rates. It’s hard to imagine companies that together plan on spending ~$2.2 trillion between them on compute over the next two years chancing the domino effect of a CoreWeave failure over a $50 billion debt load on a company with a market cap of a measly $36 billion. Letting it go would be far costlier than just buying it.
It does seem reasonable it is hard to imagine they will not double down. Your take on this helps me to erect a signpost for my investment.
” The hyper-scalers can sell IG debt at a clip of $25 billion any time they want at still-reasonable rates. ”
Meanwhile, those same hyper-scalers are actively going through contortions to move datacenters off of their balance sheets, no?
That said, who are the largest creditors? Who do they owe money to? Without any research whatsoever, I assume it is to Nvidia for current and future GPU purchases. Good old-fashioned vendor finance. Nvidia seems to have cash to spread around to everyone, so aren’t they the most likely white knight here?
Admittedly — and just as a general word of caution — this does have a distinctive “start of something” feel to it. I wouldn’t be surprised if CoreWeave is a mainstay of the news cycle for a while. There are going to be shakeouts and it’s certainly vulnerable.
I wouldn’t be surprised if we have a correction, but I don’t think we’re at the big shakeout stage. There’s still too much money floating around and the Fed has plenty of ammunition if things start looking dicey.
My guess is that a twin supply glut in semis and oil is what triggers a significant downturn in the next year or two. Those are the two industries that have experienced windfall profits but are most susceptible to boom/bust cycles. Supply might catch up in both of those industries sooner than people expect and profits can evaporate quickly.
I could see an oil glut happening sooner if Papa Trump and Iran decide to make nice after the midterms and the deprived royal families in the Middle East decide to play catch up on their lost revenue from Trump’s misadventures.
Nvidia now holds 47.2 million shares of CoreWeave stock, or around 10% of the company.
“FUN” #
Along with the revenue growth, CoreWeave also nearly tripled its debt over the last year and now has $21 billion on the balance sheet, with an average interest rate of roughly 11%. That means the company is paying about 25% of its top-line revenue just to cover the interest on its loans.
Maybe reorganize, become a country?
this is fun on several levels
Coreweave Cl A (CRWV)
ATH 6/20/25 at $ 187
7/29 anniversary overshoot LOL was ‘celebrated’ with a low of $ 60.55
If that breaks tomorrow good night
This is apt and quite good in Foreign Affairs: https://www.foreignaffairs.com/united-states/global-ai-bust-ruchir-sharma
Thanks for the link, excellent and eye opening. If I’m Warsh and read this, I have one more reason to raise rates.
Meanwhile, since I have some chunky options positions expiring on Friday (which I should have rolled last week) I’m in a back to the past mode watching overnight markets. I know enough not to read much into overnight futures trading, but I’m curious if there is anything behind the drop in 10-year yields I’m seeing. And a bit of a bid in the yen spot rate. Hell, even gold managed to crawl up 50 cents or so.
Maybe I need to go out and buy some smokes and revert back into my old life tonight….
Well, think about it: AI debt jitters could mean the beginning of a slowdown in AI-related corporate supply, which would mean less competition for USTs. Also, it’s probably just a risk-off, flight-to-quality dynamic.
but 2Y yield seems over excited?
What you say makes sense. But the fact that the 10-year is moving so much in the thin overnight market rather than during the more liquid normal hours is “curious”. I’ve been told that big players sometimes pushed around thin overnight markets with some intent. Sometimes pushing a price down in order to set up a market to buy once the trading moved to more liquid hours. Or to simply run a bunch of stops. Horrible market manipulation!
But with so much volume being in volatility-driven and other mechanical models, I suppose it’s possible that some big fish hopes to move prices enough to trigger those models to follow. “Sharks sighted! Everyone out of the water immediately!”
Or just an old man’s delusions. I’ll just go to bed.
Ah well, glad I went to sleep!
Is CoreWeave the Bear Stearns in this story? I guess within a year we should know the answer.
Who’s Countrywide Financial then? Nvidia I suppose?
That would make Jensen Huang the Angelo Mozilo of this story, which is actually kind of fitting.
And by “fitting” I mean “hilarious”.
“Admittedly — and just as a general word of caution — this does have a distinctive “start of something” feel to it. I wouldn’t be surprised if CoreWeave is a mainstay of the news cycle for a while. There are going to be shakeouts and it’s certainly vulnerable.”
It definitely feels like some of the shine has come off the entire space at least for the time being. The cost of borrowing and the drain on overall liquidity is bringing more scrutiny now, and rightfully so. Keep an eye on semis and the KOSPI.
Now, sir, you are talking. Data centers are, in truth, mostly undifferentiated enterprises selling a commodity-type products. The economic outcome of such firms will be a zero economic profit. While various AI models are somewhat differentiated, the outcomes they are intended to create are not all that different.