A company called CoreWeave (NASDAQ: CRWV) has been quietly imploding and if you own AI stocks, it’s worth understanding exactly what’s happening and why.
CoreWeave’s share price has been cut in half from its May 2026 highs. It’s down approximately 27% since October 6 alone. The interest rate on its debt has surged above 15%, which is a level that signals serious distress in credit markets. Rating agencies have a junk rating on its bonds.
This is a company with $5 billion in revenues that still managed to lose $1.2 billion last year. And its entire business model is built on one single bet: that demand for AI computing power will keep rising indefinitely.
That bet is under pressure. And the knock-on effects could be wider than most investors realise.
What CoreWeave does
CoreWeave isn’t a software company. It isn’t building AI models. It’s doing something simpler and more specific. It owns enormous clusters of Nvidia GPUs and rents the computing power to AI companies that need it.
Think of it as a landlord for artificial intelligence. Except instead of renting office space, it rents the processing capacity that AI companies need to train their models, run their systems and serve their customers.
The business model sounds straightforward. Buy GPUs. Rent them out at a margin. Collect the spread.
The problem is the leverage. CoreWeave borrowed heavily to build its GPU infrastructure, and when you borrow at scale to fund capital-intensive assets, the interest rate on that debt becomes unsustainable.
At 15%, CoreWeave is paying a distress-level rate that consumes an ever-larger share of its revenue. As rates rise, investors panic-sell on fears the debt can’t be repaid. As investors sell, rates rise further. The spiral becomes self-reinforcing.
Here’s what makes this more than just one company’s problem.
CoreWeave has contracts with OpenAI, Microsoft and Intel. Nvidia doesn’t just supply CoreWeave with chips. It holds an 11% stake in the company. CoreWeave is embedded in the AI ecosystem in ways that make its difficulties genuinely contagious. When CoreWeave sneezes, the companies connected to it feel it.
The canary in the coal mine – but which coal mine?
The CoreWeave story is being read in two very different ways right now.
The bearish reading: CoreWeave is the first visible crack in an AI infrastructure boom that has been built on leverage, optimism and the assumption that demand growth would outrun capital costs indefinitely.
Sentiment about AI company finances has turned – OpenAI and Anthropic’s financials ahead of potential IPOs have raised questions – and CoreWeave is the company where those questions are manifesting most painfully. If this is the beginning of an AI valuation reset, the stocks that have been keeping the rest of the market afloat are about to become the thing dragging it down.
The bullish reading: CoreWeave’s problems are specific to CoreWeave. A leveraged, pre-profit infrastructure rental business with junk-rated debt and a single-theme business model isn’t representative of the AI economy broadly. It’s the weakest link in a chain – the first to feel pressure precisely because it has the least resilience. Its collapse, if it comes, clears out the speculative excess without undermining the genuine structural growth story.
My answer?
Both readings contain truth. The question is which one dominates in the near term.
Then look at what ASML and Taiwan Semiconductor just reported
If CoreWeave is the bearish signal, ASML and Taiwan Semiconductor just provided the most powerful bullish counter-signal available.
ASML, the Dutch company that makes the only machines capable of printing the world’s most advanced semiconductor circuits, beat Wall Street’s estimates for earnings and revenue and raised the low end of its full-year guidance for the third time this year.
More significantly, it’s already fully booked for 2027 and 2028. Chipmakers are already placing orders for 2029 and 2030. ASML’s current capacity cannot keep up with demand.
Taiwan Semiconductor, which controls more than 90% of all advanced chip manufacturing in the world, reported net profit up 69% year on year and record profits for the sixth consecutive quarter. It’s ramping up capital expenditure to expand capacity because demand is outpacing what it can currently produce.
ASML and TSMC are the two most critical chokepoints in the entire global semiconductor supply chain. They are near monopolies. Their order books don’t lie.
When the company that makes the chip-printing machines is booked through 2028 and already taking 2030 orders, the underlying demand for
AI computing is accelerating.
How to read the contradiction
The AI boom is real. The demand for chips is real. The infrastructure build-out is real. What isn’t real (or not indefinitely sustainable)is the idea that every company that hitched itself to that boom deserves to survive.
A leveraged GPU rental business with junk-rated debt and $1.2 billion in annual losses isn’t the same thing as a company with genuine competitive advantages serving real, growing demand. CoreWeave bet that the spread between what GPUs cost and what customers would pay would stay wide and stable. That bet is being tested. And the debt structure that allowed it to scale quickly is now the mechanism through which the test becomes about survival.
The AI companies most likely to survive (and compound) through this period of sharper scrutiny aren’t the ones built on borrowed infrastructure and thin spreads. They’re the ones with genuine earnings, pricing power and technology that customers cannot easily replace.
ASML has a literal monopoly on the machines that make advanced chips possible. Taiwan Semiconductor makes chips that no other company on earth can replicate at scale. They’re the irreplaceable physical infrastructure that AI requires and they are reporting record profits and record demand simultaneously.
CoreWeave is the canary in the coal mine. But the coal mine it’s warning about is not “AI is over.” It’s “pre-profit, leveraged, single-theme companies with junk debt are going to get repriced.”
That’s a different warning. And it’s a useful one, because it tells you exactly where not to be, and exactly where the durable money continues to be made.
The next AI winners may not be the companies you think…
And this brings me to something we’ve been examining closely in South African Investor.
Because CoreWeave’s problems point to a much bigger change taking place across the AI industry.
For the past two years, investors have been obsessed with growth.
Who is building the biggest data centres? Who is buying the most chips? Who is spending the most billions?
But there’s a question that’s becoming far more important.
Who can actually afford to keep doing it?
Earlier this month, two of the semiconductor industry’s most influential CEOs – Broadcom’s Hock Tan and Intel’s Lip-Bu Tan – raised precisely this concern.
Their warning was striking.
AI’s next big bottleneck may not be chips, electricity or computing power.
It may be money.
And that changes the investment equation completely.
Because when capital becomes scarce and expensive, the companies that can fund their own growth have an enormous advantage over those dependent on lenders and investors to keep the lights on.
But here’s where it gets particularly interesting.
There isn’t just a divide between strong AI companies and weak ones.
There are actually three very different groups emerging from this boom. And one of them could be in an especially powerful position as financial pressure builds.
In the latest issue of South African Investor, Joshua Benton breaks down these three groups and explains why the next phase of the AI boom could reward a very different set of companies from the last.
He also reveals the three questions investors should now be asking before buying any AI stock.
Because the biggest mistake investors could make today is assuming that booming demand for AI automatically means booming profits for everyone supplying it.
It doesn’t.
And as CoreWeave is discovering, getting that distinction wrong can be enormously expensive.
If you own AI shares – or you’re considering buying into the sector – I’d strongly encourage you to read Joshua’s analysis before making your next move.
What to do with this
The lesson from CoreWeave isn’t to abandon AI stocks.
Far from it.
The demand for computing power is still growing. The world’s biggest technology companies are still investing enormous sums in infrastructure.
And companies like ASML and Taiwan Semiconductor are demonstrating that the underlying boom remains very real.
But the days when investors could buy almost anything associated with AI and expect to make money may be coming to an end.
From here, financial strength could become just as important as technological leadership.
That means looking much more carefully at how companies fund their expansion, how much debt they’re carrying and whether their profits can justify their investment ambitions.
And it means recognising that some of the biggest beneficiaries of the next phase of AI may not be the companies attracting the most headlines today.
CoreWeave is a warning.
Not that AI is finished.
But that the market is beginning to distinguish between companies that can profit from the boom and companies that are simply spending billions to participate in it.
And I suspect that distinction will determine some of the biggest winners and losers in the market over the next few years.
Not a subscriber to Money Morning?
You can get free daily recommendations like these with Money Morning eletter. Just sign up here.
