✦ The Five-Gigawatt BuildThin moat

The scale of what CoreWeave intends is easier to state than to finance: roughly five additional gigawatts of data-center capacity by 2030, with 2026 capital spending expected to roughly double¹. For context, that is a build-out comparable in power terms to what established hyperscalers are attempting — being undertaken by a company that has never reported an annual profit.

There is a coherent argument for going this fast. The contracted backlog is enormous, customers are asking for capacity years forward, and in a land-grab the returns accrue to whoever has megawatts energized when the demand lands. Building ahead of contracts would be reckless; building behind them means handing the growth to a competitor. CoreWeave is, at least, mostly building against signed agreements.

But the financing is the fragility. This build-out runs on GPU-collateralized debt and interest costs already measured in hundreds of millions a quarter, against depreciation on assets whose useful life is genuinely debated. A slowdown does not have to be dramatic to hurt: it only has to arrive before the capacity is contracted and the debt is serviced. Watch the capex figure against contracted backlog conversion, the weighted cost of new borrowing, and whether any of this build-out is being funded with equity rather than debt. The financing mix is the honest measure of how confident management really is.