⚠ Renting a Commodity Built on Someone Else's ChipsHigh threat

Beneath CoreWeave's engineering skill lies an uncomfortable fundamental: its product is GPU compute, a commodity built on chips that its supplier sells to everyone, which structurally limits how wide the moat can be. CoreWeave does not make the GPUs; it buys them from Nvidia, the same Nvidia that sells to the hyperscalers, the other neoclouds, and anyone else with the money. The value CoreWeave adds is real — in how it assembles, networks, operates, optimizes, and finances that hardware at scale — but it is reselling access to a resource that its customers and competitors can also acquire directly, and that fundamental fact caps the premium and the durability of its position.

A business that resells a commodity it buys from a single powerful supplier sits in a structurally weak spot. Nvidia captures the majority of the value in the AI-compute chain through its own enormous margins; CoreWeave earns a spread on top by operating the hardware well, but that spread is pressured from above by Nvidia's pricing and from the side by competitors reselling the same chips. There is no proprietary technology that only CoreWeave possesses, no unique input it controls — the underlying capability is available to anyone who can buy GPUs and hire engineers. CoreWeave's operational excellence and scale let it do this better than most, and that is a genuine, if thin, advantage. But an investor should hold the structural truth clearly: CoreWeave is a highly-skilled operator of a commodity resource it does not control, dependent on a supplier that also arms its rivals, earning a spread that the supplier and the competition both press on — a position that can be lucrative in a boom — $5.13B of revenue lucrative¹ — but that lacks the control over a scarce, proprietary input that a truly wide moat requires.