CoreWeave's Growth Is Real. The Financing Is the Risk That Isn't Priced.

By 2026, CoreWeave has become the fastest-revenue-growth story in public markets — and the market can't decide what to do with it. Last week the stock traded hands near $95, down roughly half from its mid-2025 peak, even as the company reports 112% year-over-year revenue growth and keeps beating Wall Street. The consensus answer to that gap is easy: the stock fell because the business is overvalued and unprofitable. I think that answer has the clock wrong.

CoreWeave is what a rent-a-GPU-cloud looks like. It buys Nvidia's most powerful chips, builds data centers around them, and signs multi-year contracts to rent the capacity to companies training large AI models. It went public in March 2025 at $40 a share, and at its June 2025 peak the stock poked above $180 before the retreat. What matters now is not the demand — that is genuinely enormous. As of June 30, 2026, the company held $104 billion of contracted revenue backlog, up 246% year over year, and raised its full-year 2026 revenue forecast to $12.4–13.2 billion after a $2.58 billion second quarter. When a stock sells off despite that, the debate has already left the demand question behind.

The real bet on CoreWeave is about financing, not demand. Here is the single number that carries the whole case: in the second quarter, CoreWeave paid $640 million in net interest while its adjusted operating income was $128 million. Interest alone ran roughly five times the operating income the business actually produces after charging itself for its own hardware. That is not a profitable company with a cash-flow hiccup; that is a machine that must borrow faster than it earns just to stand still.

Walk the balance sheet and the shape becomes clear. CoreWeave ended the quarter with roughly $35 billion in debt — against a stock market value of about $52 billion — and guided to a 2026 capital-spending plan of $35–39 billion. It built the backlog by spending first and collecting later, and it pays for that build with tens of billions of new borrowing each year. The interest bill does not merely rise; it compounds as each new data center adds depreciation and each new bond adds coupons. The top line is growing, but the fixed financing bill is growing on top of it.

So where does the cheap money come from? That is the part the growth story does not advertise. CoreWeave borrows at rates far below what its own credit would justify — a funding package in early 2026 priced at an implied cost under 6%, down from roughly 15% for its 2023 debt — because its biggest customers back those loans with offtake agreements. And that brings the second hidden concentration: Microsoft alone accounted for about two-thirds of fiscal-year 2025 revenue and the bulk of the backlog. CoreWeave is, in the most literal sense, a levered toll road whose biggest toll-payer is also the collateral that keeps its borrowing cheap.

That is the reflexive loop that makes this a wager rather than a forecast. The market is paying for growth, and rightly so in revenue terms. But valuation and credit now travel together: at about $82 billion of enterprise value on roughly $35 billion of debt, lenders are not bystanders. They already signal the pressure point. Fitch rates CoreWeave's debt BB- with a negative outlook and has said it could downgrade if EBITDA leverage, excluding leases, is sustained above 4x or lease-adjusted leverage above 5x — and on current numbers debt sits at multiple times that.

Here is my contract, dated and falsifiable. Subject: CoreWeave common equity. Direction: it re-rates on the financing constraint, not on demand. Leading indicators: by the full-year 2026 report, total debt will have crossed about $40 billion and quarterly net interest will exceed roughly $700 million while adjusted quarterly operating income stays below $200 million. Mechanism: a $35–39 billion capex year needs tens of billions in new financing, and each dollar raises a fixed bill that income has not yet outstripped. Break condition: if a hyperscaler injects large, cheap permanent capital, or if adjusted operating income grows faster than the interest bill so coverage durably improves, the thesis dies.

I am not calling the backlog a fraud — it is the strongest part of the story. I am calling the valuation a bet that the funding loop never tightens. The stock market, with street targets clustered near $140 and AInvest's aggregate signal still labeling CoreWeave a Buy, is priced as if growth answers every question. The financing clock says the question that matters has a deadline, and it is not the demand report — it is the next filing, where the interest line tells you whether the treadmill is getting faster or the runner is finally pulling ahead.

Watch one thing: the coverage ratio, stated as quarterly operating income against quarterly interest. If it stays thin while debt balloons, the equity is a call option on a business whose economics are set by its lenders. If it turns, I am wrong. That is the point of writing the threshold down in public.