The MoatThin moat

CoreWeave is the purest and most extreme example in this collection of a company riding a historic tailwind on a thin, fragile, borrowed moat. It is the leading 'neocloud' — a specialized cloud-computing provider that rents Nvidia's AI chips, at enormous scale, to the companies building artificial intelligence. In a few years it has gone from a failed crypto-mining venture to a roughly $48 billion public company¹ whose revenue is growing in the triple digits and whose contracted backlog approaches $100 billion. The demand is real, the growth is staggering, and the business is at the white-hot center of the AI build-out. But the honest rating is thin, because underneath the spectacular growth sits a business that has never earned a profit, is financed by a mountain of debt secured against rapidly-depreciating chips, depends on a handful of customers who are also its fiercest competitors, and is entangled in a web of circular financing with its own supplier. It is a real business with real scale — and a moat that is unproven, levered, concentrated, and borrowed.

What moat CoreWeave has comes from a few genuine sources. Start with specialized scale and performance: CoreWeave built its data centers and software from the ground up for AI workloads, and it can deliver GPU compute at a scale, utilization, and performance that the generic hyperscaler clouds have, at times, struggled to match — a real engineering edge that made it the destination of choice for AI labs needing vast compute fast. Next comes its extraordinary relationship with Nvidia, which gives it early and preferred access to the most sought-after chips in the world and which has taken an ownership stake of around 11%² — Nvidia is at once CoreWeave's supplier, its investor, and its backer. Sturdiest of all is the contracted backlog: some $104 billion of long-term, take-or-pay commitments³ from customers including OpenAI, Microsoft, and Meta, which lends the business years of revenue visibility. And behind it all runs a capital-raising machine that has pulled in roughly $28 billion of financing in a year⁴ to build faster than anyone else.

But every one of those strengths is shadowed by a matching vulnerability, and together they are why the moat is thin. The performance edge is contested: the hyperscalers CoreWeave outruns today are its customers, and they are building their own specialized AI infrastructure to insource what they now rent. The Nvidia relationship is a dependence as much as an advantage — Nvidia can favor others, standardize access, or find its interests diverging from CoreWeave's. The backlog is impressive but concentrated in a few enormous counterparties and, being a promise rather than cash, carries execution and credit risk. And the capital machine runs on debt: CoreWeave finances its GPU fleet with borrowings secured against the chips themselves, at high interest, so that interest expense (over $500 million in a single recent quarter)⁵ and depreciation (over $1 billion in a quarter) swamp its explosive revenue and keep it deeply unprofitable.

The concentration risk deserves its own emphasis, because it is the sharpest edge in the business. Microsoft accounted for roughly two-thirds of 2025 revenue, and the backlog leans heavily on a small number of giant customers — Microsoft, OpenAI, Meta — whose spending decisions can make or break CoreWeave and who each have their own reasons to build rather than rent over time. And the whole structure sits inside a circular financial web: Nvidia invests in CoreWeave, which uses debt and equity to buy Nvidia's GPUs, to serve customers like OpenAI, in which Nvidia is also investing — the same dollars appearing to circulate through the AI economy in ways that flatter every link in the chain and that no one has seen tested by a downturn.

The valuation reflects a market that has, for now, chosen to price the growth and the backlog over the fragility. At around $88 a share and roughly six times sales⁶ — with no price-to-earnings ratio at all, because there are no earnings — CoreWeave is valued as a company whose $104 billion backlog will convert into profitable, durable revenue and whose thin moat will hold as the AI build-out matures. It might. The bull case is that CoreWeave is the essential specialized infrastructure of the AI era, growing into a backlog that guarantees years of scale, backed by Nvidia and demanded by everyone. The bear case is that it is a levered, concentrated, unprofitable reseller of a depreciating commodity, whose customers are its competitors and whose financing is circular, priced as though its moat were wide when it is thin. An investor in CoreWeave owns explosive growth, a real backlog, and a genuine but fragile position at the heart of the AI boom — bought with debt, concentrated in a few hands, and dependent on a cycle no one has seen end. The growth is real, the backlog is real, and the risks are exactly as large as the opportunity. The number that decides which wins is the spread between what the backlog converts into revenue and what the debt and depreciation consume: if contracted dollars arrive faster than interest and chip-aging eat them, the thin moat thickens into a franchise; if they don't, the same leverage that built CoreWeave this fast will unbuild it faster.