CoreWeave Hit a 2026 Low. Goldman Sachs Said Buy Anyway

A 25 Percent Drop in Five Days — And an Upgrade

CoreWeave’s shares tumbled to $59.31 on October 19, the lowest level since May 2025 and a brutal end to a stock that had doubled earlier in the year. Five trading days accounted for a 25 percent decline. The market closed at $61.12.

And Goldman Sachs upgraded the stock from neutral to buy on that exact day.

The move is the kind of signal that makes you look closer at what analysts think they know versus what the bond market is already pricing in. Because on the same session that Goldman published its positive revision, CoreWeave’s outstanding debt — $52 billion — was trading at a yield that suggested investors were far less convinced than the equity-side research was letting on.

Who Is Actually Buying the Story Here?

Goldman’s upgrade came after a summer where shares climbed as high as $137, making the current level look like a steep reversion. JP Morgan analysts, speaking to Bloomberg, took the same tone: the business model is sound despite the debt load and rising bond yields. Their reasoning hinged on two things — sustained demand for AI compute and long-term customer contracts that lock in revenue far out.

Nvidia’s CEO Jensen Huang added weight to that camp during the week of October 12. He noted that capital costs for CoreWeave were actually decreasing and described the company’s competitive position as strengthening daily. That is not an ordinary friend of the company to have inside the boardroom; Nvidia owns roughly 10 percent of CoreWeave and supplies the vast majority of the GPUs running its facilities.

So you have three bullish voices — Goldman, JP Morgan, and Nvidia’s chairman of the board — all talking up the same name at the same moment the stock is getting crushed. That alignment is either exceptionally well-informed or it is the financial industry doing what it always does in a pullback: buying the dip in a name it cannot afford to miss.

The Bond Market Is Not Enthusiastic

Here is what the equity analysts are not telling you about: CoreWeave’s 2031 bonds, bearing a coupon of 9.75 percent, are now yielding 15.7 percent on a final-yield basis. That spread tells a story the press releases do not. Investors are demanding roughly six percentage points of extra compensation over the coupon rate because they fear something will go wrong between now and maturity — not because of the coupon itself, but because of the scale of the debt pile and the pace at which it has accumulated.

A 15.7 percent yield is not a healthy number for a company that is still scaling. It means borrowing new money to refinance old money at increasingly punitive rates, and it means every delay in data-center delivery or every renegotiation of a customer contract compresses margins in real time.

The Numbers Are Bigger Than the Headline

On the face of it, CoreWeave looks like the winner in the AI infrastructure boom. Revenue in the second quarter hit $2.5 billion. Its order backlog — future revenue locked into customer contracts — sits at approximately $104 billion. That is extraordinary. The named customers alone read like a who’s who of the generative-AI economy: Anthropic, Google, Meta, and Perplexity.

But $104 billion in backlogged revenue is not cash in the bank. It is a promise contingent on construction timelines, equipment deliveries, and the willingness of lenders to keep funding expansion. When debt service costs rise, the gap between promise and reality widens fast.

A year ago, CoreWeave shares were trading roughly 55 percent higher. The summer high of $137 now looks like the peak of a very public bet that demand would outpace debt indefinitely. Today’s price implies the market is recalibrating that bet — and recalibrating it downward by a lot.

The Gartner Impairment Question

Japanese financial analysts covering CoreWeave have flagged a Gartner projection that AI infrastructure companies face a cumulative impairment risk of up to $550 trillion over the coming decade. That number sounds abstract until you consider that CoreWeave has already borrowed $52 billion and is likely to borrow far more before its current expansion cycle ends. If even a fraction of Gartner’s estimate proves accurate — meaning massive asset write-downs across the data-center and semiconductor ecosystem — CoreWeave’s debt structure becomes untenable at today’s valuations.

JP Morgan’s counter-narrative is simple: the demand is real, the contracts are signed, and the yield on CoreWeave’s bonds is simply overpriced relative to the fundamental credit story. Goldman’s upgrade follows the same logic. The question is whether either institution is underestimating how hard it is to service $52 billion in debt when the cost of capital moves against you.

Who Wins, Who Loses

The institutional buyers who stepped in at $100-plus are underwater by roughly half. Retail investors who bought the Nvidia hype thread are still holding the bag. The bondholders who bought at par are sitting on unrealized losses of nearly ten percentage points per bond.

CoreWeave itself is in a strange position: too big to ignore, too leveraged to relax, and with enough customer demand to keep the lights on even if the balance sheet takes a hit. The company can keep building — the backlog will fund the next round of borrowing — but every new dollar of debt comes at a higher price.

If Goldman and JP Morgan are right, this is a temporary dislocation and the stock will recover as the backlog converts to revenue and the bond yield narrows. If the Gartner scenario has any traction, the opposite happens: impairments force write-downs, the debt becomes unmanageable, and the companies that bet on CoreWeave’s durability learn a costly lesson about how quickly infrastructure-scale debt can turn toxic.

What Happens Next

The next few quarters will be decisive. Any sign that CoreWeave’s construction pipeline slips, that a major customer renegotiates terms, or that refinancing costs spike beyond 15 percent will pressure the stock further. Conversely, a steady conversion of backlog to reported revenue and a tightening of bond spreads would vindicate the bullish analysts and turn this dip into a textbook buy.

The market is currently splitting between those two futures. The fact that Goldman upgraded on the day of the stock’s worst five-day drop in over a year suggests the institution believes the fear is overdone. But the bond market has a different timeline, and it is the one that gets paid first if things go wrong.