CoreWeave reports second-quarter results on Tuesday after the close, and the setup is unusual: almost nobody is arguing about demand. At the end of March the company’s remaining performance obligations — contracted revenue signed but not yet delivered — stood at $99.4 billion, close to four times the year-earlier figure, with Nvidia, Microsoft, Meta and OpenAI among the counterparties. Meta alone committed $21 billion in March.
Wall Street expects roughly $2.55 billion of revenue for the quarter, better than double a year ago, and a loss of about $1.21 to $1.22 a share depending on whose consensus you use, against a $0.27 loss in the same quarter last year. Management has guided 2026 capital spending to $31–35 billion, roughly double last year, and has raised more than $20 billion of debt and equity so far this year to pay for it.
That is the entire story in three numbers. A $99 billion order book. A bill of up to $35 billion this year to fill it. And a business that does not yet make money. Shares traded near $90 in early August against a 52-week high of $153.20 — the market has already reframed this as a financing question rather than a demand one.
Our take: Backlog is the easiest number in the AI trade to admire and the hardest to bank. Management has said roughly 36% of the $99.4 billion converts to revenue inside two years, which means most of it is a promise about 2029 and later — contingent on power arriving, chips landing on schedule, and counterparties still wanting the capacity when it does. The spending, meanwhile, is happening now. That mismatch is why the interesting disclosure on Tuesday is not revenue growth, which will be enormous either way. It is what the money costs.
Why this print reads across
CoreWeave is the cleanest public proxy for a question hanging over the rest of the market: is AI infrastructure being funded out of cash flow, or out of capital markets? Microsoft, Alphabet, Amazon and Meta can absorb vast capital budgets because they have vast operating profits sitting underneath them. CoreWeave does not. It borrows, it takes customer prepayments, and it does both against contracts that mostly pay out years from now.
Moody’s put combined 2026 capital spending across Microsoft, Amazon, Alphabet, Meta, Oracle and CoreWeave at roughly $785 billion, rising toward $1 trillion in 2027. CoreWeave is the smallest of those six and the only one where a shift in credit conditions shows up in the model immediately. It also reports first this week, ahead of Cisco and Applied Materials — a run of results the week-ahead previews are treating as a checkpoint on whether AI spending is still accelerating. Everything else on the calendar, including Wednesday’s July CPI, is about rates. This one is about whether the buildout can still be financed at a price that works.
What to watch
- Backlog additions, not backlog size. The $99.4 billion is already known. What matters is how much new contracted revenue was signed in the quarter, and at what duration. Cantor Fitzgerald’s Brett Knoblauch has said he expects a sizeable beat on that line.
- Interest expense against revenue. The ratio of the two is the health check. If borrowing costs are climbing faster than delivered revenue, the model tightens regardless of how large the order book gets.
- Capex guidance. Raised, held or trimmed. A trim would be read as discipline by some and as slipping deployment by others, and management’s framing will decide which.
- Delivery-timing language. Power interconnects and chip availability are the two constraints that turn signed contracts into delayed ones. Listen for hedged phrasing on when capacity comes online.
- Customer concentration. A backlog anchored by a handful of names is a different risk profile from a diversified one, and the counterparties here are themselves the biggest spenders in the sector.
The bull case and the bear case start from the same fact, which is what makes Tuesday worth reading properly. Demand for AI compute is real, contracted and enormous. Serving it requires spending money the company has to raise first, against revenue that arrives later. Tuesday tells you which half of that sentence is currently winning.