GPU Fleet Scale & UtilizationNarrow moat

The physical embodiment of CoreWeave's scale is its GPU fleet — one of the largest deployments of Nvidia AI chips in the world outside the hyperscalers, spread across a rapidly-growing footprint of data centers. This scale is a genuine competitive asset: it lets CoreWeave serve the largest AI customers, whose training runs require tens of thousands of GPUs working together; it gives CoreWeave the volume to command Nvidia's attention and its scarcest chips; and it lets the company spread its fixed costs over an enormous compute base while running the fleet at high utilization — keeping expensive GPUs busy is the whole game in this business, and CoreWeave's operational skill at doing so is real.

The scale and utilization are why CoreWeave could grow revenue at triple-digit rates and win contracts the smaller neoclouds could not. But the honest and crucial caveat, developed in the accompanying threat, is that this scale is built on assets that depreciate rapidly and are financed with debt. A GPU is not like a fiber network or a warehouse that holds its value for decades; it is a fast-obsolescing chip whose worth falls sharply as Nvidia ships each new generation, and CoreWeave has bought its fleet largely with borrowed money secured against the chips. So the scale is simultaneously a real advantage and a wasting, leveraged one — an enormous fleet that must be constantly refreshed with more capital, whose value is eroding even as it earns, and whose financing carries a heavy interest burden. Scale in GPUs is a genuine edge in the moment, but unlike scale in durable infrastructure, it is a treadmill that consumes capital — ~$28B raised in a single year to keep it turning¹ — rather than a fortress that compounds.