CoreWeave, Inc.

Money comes from two linked arrangements: long-duration commitments in which a customer pays for a fixed amount of capacity over a period of years whether it uses all of it or not, recognized in even amounts across the term, and shorter on-demand access billed as it is consumed. Revenue has grown every year across the recent trailing period, and so has the amount customers owe but have not yet paid, so a growing share of revenue sits as a receivable rather than cash already collected.

The system scales mainly by adding large amounts of physical infrastructure capacity, and CompanyGraph reads its capital structure as leaning heavily on borrowed money to do so: debt is elevated against equity, against total assets, and against the cash the operating business generates, all at once, alongside a broader distress signal reading high and a recent trailing period in which profitability was not consistent. It is also one of a sizeable, bounded group of companies CompanyGraph currently maps as organized around this same pattern of subscriptions and multi-year commitments, so the way it scales is a shared pattern rather than a singular one.

The system depends on a small number of hardware suppliers, above all NVIDIA, whose processors its infrastructure is built around under current customer contract terms, and on other component suppliers that operate through individual purchase orders rather than long-term capacity guarantees. It also depends on outside operators of the data centers and telecommunications capacity it uses but does not own, on electricity availability, and on the pace at which new processor generations and data-center capacity can be delivered, drawing more broadly from a wide range of upstream industries beyond these named dependencies.

A small number of customers account for most of its revenue: one customer, Microsoft, alone accounts for a large majority of it, and two other named customers, OpenAI and Meta, hold large multi-year capacity commitments extending several years into the future. Its buyers are large AI developers and enterprises rather than individual consumers, and downstream it supplies a comparatively narrow set of industries relative to the broad base it draws from upstream.

No measurement of what rivals can or cannot replicate is available. What can be said is positional: CompanyGraph currently maps a bounded, moderate-sized group of other companies as organized around this same subscription-and-commitment pattern, so at that level, the way this system is built is a recurring pattern rather than a singular one.

Customers who sign committed contracts agree to pay for a fixed amount of capacity across a multi-year term regardless of how much they actually use, with terms that typically run several years and a large share of already-committed future revenue still to be delivered years out. That structure means a customer wanting to leave before the term ends would still owe for the capacity it committed to, which is a contractual form of lock-in rather than a technical one.

The broader category CompanyGraph places this company in generally expects growth to be limited by how well it retains customers against switching and against the cost of winning them, but the company's own account does not describe that as its limit. Instead, it names the availability of power, of the processors and components its infrastructure needs, of data-center space, of financing for continued capital spending, of permits, and of specialized staff as what limits how fast it can expand, pointing toward a physical-and-capital limit rather than a customer-retention one.

The company's own risk disclosures name several vulnerabilities first: the cost and availability of electricity, a limited number of suppliers for significant infrastructure components, possible failure, damage, interruption, or security breaches at the outside data-center providers it uses, the difficulty of managing fast growth, concentration in a small number of customers, and the scale of capital it must keep raising. It also discloses that one customer accounts for the large majority of its revenue and that its infrastructure currently runs on processors from a single named supplier under existing customer contract terms.

The company names government export-control and trade policy on advanced computing hardware as an outside force that can affect the availability and cost of the processors its infrastructure depends on, together with tariff exposure it names for the current policy environment. It also names a modest exposure to a handful of foreign currencies on the cost side, set against revenue that is mostly priced in a single currency.