您的浏览器禁用了JavaScript(一种计算机语言,用以实现您与网页的交互),请解除该禁用,或者联系我们。 [伯恩斯坦]:CoreWeave:GPU、GW及看不到的自由现金流……如何评估新型云服务公司? - 发现报告

CoreWeave:GPU、GW及看不到的自由现金流……如何评估新型云服务公司?

2026-03-24 伯恩斯坦 John
报告封面

US Communications InfrastructureCoreWeave, Inc. Underperform Price Target CRWV CoreWeave: GPUs, GWs, and no FCF in sight...How should onevalue a Neocloud? CRWV (UP, $56) is tricky to value given its aggressive growth, high capex and debt levels,and lack of clear comparables. We’ve analyzed 5 different valuation methodologies (DCF, EV/Revenue, EV/MW, EV/EBITDA and EV/EBIT), giving our rationale for choosing EV/EBIT. DCF-based valuation is out of the question.While we appreciate that DCF keeps thefocus on cash flows, we do not expect CRWV to generate positive FCF for the next 4 years.Most of the DCF value therefore would hinge on terminal value - no thanks. EV/Revenue might seem better given CRWV’s growth trajectory, but fails to accountfor capex.Traditionally, EV/Revenue is used for early stage, high-growth companies withthe assumption that the growth provides visibility into future earnings and cash flows. WithCRWV’s capex burden, this misses half the story. Many have suggested using power, but unfortunately neither EV/Active Power norEV/Contracted Power works cleanly as a primary valuation lens.Active Power suffersfrom the same issues as EV/Revenue; Contracted Power is even weaker given nonuniformterms and no assurance of revenue conversion. Comps here are also limited. EV/EBITDA would typically be our pick, but we just can’t get comfortable ignoringCRWV’s depreciation.We estimate CRWV will spend $80B+ in capex over the next 5years. Even for the hyperscalers, AI expansion has made historical capex trends unreliable.Considering the asset-heavy nature of CRWV, we believe ignoring the depreciation impactmakes EV/EBITDA inappropriate. While still imperfect, we believe EV/EBIT solves most of these valuation obstacles. Itis dynamic enough to adapt to changes in CRWV’s growth and cape trajectory and enablesus to value CRWV at the enterprise level, incorporating the expected debt issuance. We are at28.4x FY 2027 EBIT and view 25-30x as reasonable. Investment Implications We rate CoreWeave Underperform with a $56 price target. DETAILS Overview of CRWV’s Business Model CRWV operates as a Neocloud, offering rental of fully-deployed GPU clusters. Most neoclouds, including CRWV, are sourcingtheir power and data center capacity from other providers (largely private), though some may own or self-build (CRWV doesnot, at present). They then signlarge, take-or-pay contracts with tech players, often hyperscalers or LLM developers, andfinance GPU purchases against those contracts. The neocloud then collects revenue for the term of the contract, whilecontinuing to pay lease fees on its data center real estate. Like other neoclouds, CRWV’scapital requirements are high, leaving deals cumulativelyFCF negative for the first ~4yearsof each contract. CRWV has been clear about the shape of these economics, with big upfront investments and revenuerecognized over the duration. What CRWV is banking on is the outer years of the contract plus the chip value post-contract,when fully-owned and depreciated chips can be used to generate high-margin, on-demand revenue. We agree that GPU lifespans are likely to exceed 6 years, though we believe the utility erodes with time (the actual amount ishotly debated - we’re at 10%/yr). When these deals and the on-demand revenue are rolled together, the FCF trajectory starts tobe somewhat concerning -we model CRWV’s FCFE trajectory as cumulatively negative until 2030. Despite our forecastof ~$4B in operating cash flow in 2026, we anticipate CRWV will burn through~$26B of FCF, driven by ~$30B of capitalexpenditures to fund GPU purchases. Almost all of this is likely to be funded by incremental debt issuance, creating a veryprecarious structure, and extracting value from residual equity. This fragile capital structure dynamic combined with CRWV’sevolving and uncertain competitive position in the AI data center space makes the choice of an appropriate valuation metricmore significant than usual. There is no perfect approach, but we definitely have strong opinions on the relative merits of each. Valuation Metric I: Discounted Cash Flow (DCF) A DCF is a helpful secondary validation of value for companies with predictable cash flows and growth trajectories and theplace where we always like to start. But in this case, CRWV is expected to burn $26B, $12B, and $7B in Free Cash Flow (FCF)over the next 3 years respectively, and only generate positive FCF from FY 2030 onward. Important note: our FCF figures aremeaningfully higher than consensus as we are skeptical of CRWV’s revenue trajectory and therefore assume far fewer futureGPU purchases. As a result,almost all of CRWV’s value in a DCF is derived from its terminal value, a metric notoriously sensitive to inputassumptions about the terminal multiple (if using a multiples-based valuation) or perpetual growth rates and discount rates.These assumptions take on further uncertainty due to the rapidly changing competitive landscape of the neocloud