CoreWeave doubled its revenue and earned less doing it — one line on the income statement explains both

CoreWeave rents out AI computing power, and business is booming: revenue more than doubled to $2.58 billion and the shares jumped 14%. Yet the company earned less profit than a year ago on twice the sales. That sounds impossible, and it has a single, unglamorous explanation — the chips wear out. CoreWeave buys GPUs with borrowed money, and the cost of writing those chips down over their useful life is now growing faster than the revenue they generate. Everything else in this report — the debt, the interest bill, the leases — follows from that one fact.

Valuation — the market cap is not the price

At $90.87 CoreWeave’s equity is worth $50.1B. Its enterprise value is $96.2B — nearly double, because the debt is roughly half the enterprise. Any multiple quoted on market cap alone describes a levered slice rather than the business.

That first pair is the whole valuation argument in two rows. On the equity, CoreWeave looks like a cheap hypergrowth name at 2.6× forward sales. On the enterprise, it is at 12.7× trailing sales with $51.6B of total debt and negative $13.66B of trailing free cash flow.

Two facts sit oddly together and both are true: the stock rose 14% on this print, and it is down 6.1% over 52 weeks in which revenue more than doubled. Short interest is 11.1% of shares outstanding — by a wide margin the most-shorted name this section has covered. The sell-side average target is $143.26. Nobody is neutral on this one.

The depreciation question, which is the whole report

CoreWeave charged $1,393M of depreciation and amortisation this quarter, up 149% against 112% revenue growth. That single line does three things at once.

It is 92% of Adjusted EBITDA — the $1,510M figure is the $(626)M net loss plus D&A, interest, stock comp and tax. It is 99.2% inside adjusted operating income, which is why that measure reads $128M rather than $1.5B, and why it fell 36%. And it is the thing that flips the sign on operating leverage: excluding it, operating expenses grew 94% against 112% revenue; including it, the company posted a $(49)M operating loss.

The forward path is worse before it is better, for two disclosed reasons. Accumulated depreciation is $5.9B against $52.6B of gross PP&E — 11%. On a six-year life that is a young fleet with most of its depreciation ahead of it. And $11.9B of construction in progress, 23% of gross PP&E, is currently earning nothing and depreciating nothing; when it lands it hits the income statement with no grace period.

The correction our own skeptic pass insisted on. A first draft called the six-year life "unusually aggressive." It is not. Alphabet depreciates servers and network equipment generally over six years, Microsoft states two to six, Amazon five to six. CoreWeave is at the norm. What is true is that Amazon moved the other way — cutting a subset of servers from six years to five effective January 2025, citing "the increased pace of technology development, particularly in the area of artificial intelligence and machine learning," at a cost of $1.4B in extra depreciation. One large buyer of the same equipment concluded it wears out faster than six years.

The capital structure

$35,068M of debt against $5,024M of equity. Of that, 89.6% is recourse — a direct obligation of, or unconditionally guaranteed by, CoreWeave, Inc. Only $3,663M is non-recourse.

The rates are the part that does not appear in any summary: the first delayed-draw term loan carries 15%, the Magnetar facility 12%, recourse OEM arrangements 11%, and four of the five senior notes 10%. $10.6B matures inside roughly eighteen months against $5.5B of unrestricted cash. Since quarter-end the company has added a $2.6B facility at SOFR+5.5% and drawn $1.2B on its revolver.

A distinction worth keeping straight, because the skeptic pass corrected us on it. "Low non-recourse debt" is not the same as "no ring-fencing." Note 10 discloses that beyond the non-recourse facility, further borrowings sit in bankruptcy-remote special-purpose subsidiaries whose assets "may be used only to settle the obligations of those entities" — covering $18.2B of non-current debt. There is a great deal of structural separation here. It simply runs alongside a parent guarantee rather than instead of one.

Then the item that is not on the balance sheet at all. Note 8 discloses $35.5B of estimated future undiscounted payments on leases that have been executed but had not commenced at 30 June, running 2026–2029 on seven-to-sixteen-year terms. CoreWeave states plainly that these "were not included in our condensed consolidated balance sheet." That figure excludes two cost-plus arrangements: one site with 393 MW still undelivered, capped at $14.7B over sixteen years, and others with 355 MW undelivered.

So the obligations are roughly $35.1B on the balance sheet and $35.5B more beside it, against $5.0B of equity.

Risk — each isolated, do not blur

1. Interest against profit (dominant). $640M of quarterly interest against $128M of adjusted operating income is 0.20× coverage, and negative on GAAP. Q3 guidance pairs $200–260M of adjusted operating income against $860–940M of interest — the gap widens in dollars, not narrows.
2. The depreciation assumption. Six years is the norm; if AI silicon is economically obsolete sooner, the charge is understated and the debt was raised against faster-decaying collateral.
3. Refinancing. $10.6B due inside eighteen months against $5.5B of cash, at a company whose existing paper runs to 15%. This works while capital markets stay open.
4. The off-balance-sheet leases. $35.5B commencing over four years, plus capped arrangements excluded from it.
5. Bookings deceleration. ~$7.4B of gross bookings in Q2 against ~$40B in Q1. One quarter is not a trend, and it is the first below its year-ago comparison in six.
6. Unit costs. Cost of revenue 26% → 34% of revenue. The 25% July price increase reads partly as pass-through rather than pure pricing power.
7. Concentration and counterparties. The customer roster is a handful of AI labs and hyperscalers whose own capex plans set CoreWeave’s revenue.

Horizon and sizing (kept separate)

Horizon. The next two quarters are about mechanics: whether the July price increase holds, whether Q3 bookings recover from ~$7.4B, and whether the $10.6B of near maturities refinances at better than 15%. The thesis resolves over 3–5 years on the depreciation question — does a GPU bought in 2026 still earn in 2032?

Sizing considerations (not a recommendation). This is the most levered expression of the AI-capex cycle in this section, and leverage cuts both ways: the equity is a thin slice on top of a large, expensive, largely recourse debt stack, which is why the stock can rise 14% on a print and still be down over a year of doubling revenue. An 11% short interest means the price carries mechanical squeeze risk in both directions that has nothing to do with the filings. Position size here is a statement about the capital markets staying open, not only about AI demand.