CoreWeave, Inc. Class A Common Stock · Moat analysis

Report previewCRWV.USReport date Last updated
← Back to reports
Report contents
01

Industry landscape

Where the company competes: a capacity race in rented GPU time

CoreWeave sells contracted GPU capacity, and the thing it sells is close to interchangeable. That single fact defines the arena more than any other. The company builds data centers filled with Nvidia accelerators and rents clusters to AI developers on multi-year terms. It does not sell a software platform with switching costs built into the product; it sells hours on machines that every rival buys from the same vendor. William Blair's technology analysts, in a September 30, 2026 industry report on what they call the neocloud segment, make the point directly: standardized Nvidia hardware limits product-level differentiation, and advantage instead comes from power and GPU access, cost of capital, deployment speed, reliability, customer relationships and software layered on top.

Value chain: suppliers upstream and a narrow customer set downstream

Leverage sits at both ends of CoreWeave's value chain. Upstream is Nvidia, which allocates accelerators and sets their price, alongside utilities and grid operators that control interconnection, data-center developers, and the debt and equity markets that fund construction. Downstream is a short list: frontier AI labs, hyperscalers buying overflow capacity, and, more recently, enterprises and governments.

CoreWeave's own slice is a spread. It collects a contract price per megawatt of live capacity and pays for the shell, the chips, the power and the financing. William Blair estimates an effective revenue yield of roughly $12.5 million per active megawatt per year across neoclouds, which gives a rough ceiling on what a signed gigawatt is worth. Costs are largely fixed once a lease is signed, so utilization decides whether that spread becomes profit or loss; the same report identifies utilization as the key driver of neocloud returns. Two further variables sit inside the spread and are easy to miss: the cost of the debt and leases used to fund the build, and the depreciation schedule applied to GPUs whose residual value nobody can yet price with confidence.

Geography: power is local, customers buy globally

The competitive boundary is not a country. Power, land and grid interconnection are intensely local, and the relevant map is a set of power-constrained data-center markets where a gigawatt contracted by one operator is largely unavailable to another. Customers, by contrast, procure globally and will place workloads wherever capacity exists. Scale therefore should be read in secured megawatts and contracted commitments, not in national share of a measured market. William Blair counts more than 18.5 GW of disclosed contracted or secured power across the leading neoclouds, against an assumption of about 20 GW powered and online by 2031 — a forecast, and one that implies the land grab is closer to its end than its beginning.

Market structure: a short list of scaled sellers and an open door behind them

General-purpose cloud remains a hyperscaler business. The neocloud layer sits beside it and is still small: William Blair estimates neocloud revenue at roughly $25 billion in 2026 and forecasts $200 billion to $300 billion by 2031, with a base case near $250 billion. The named field is CoreWeave, Nebius, IREN, Nscale, Crusoe and Lambda, and the same report ranks CoreWeave and Nebius as the best positioned, citing their record of securing power and data-center capacity, financing it, and delivering at scale.

Concentration among the scaled players is real. Four of them — CoreWeave, Nebius, IREN and Nscale — have together amassed about $275 billion of customer commitments, a figure that speaks to how badly buyers want capacity locked up. No published series allocates that total by company, so CoreWeave's precise share of the segment cannot be measured from public data; what can be said is that the leaders number fewer than five while at least six credible claimants chase the same customers, and three hyperscalers build competing capacity of their own.

Apply the practical test — can the leaders be counted on one hand? — and the answer is yes at the top, which normally signals meaningful barriers. Here the barriers are financial and contractual rather than technological. A new entrant needs capital at a scale few can raise, a power contract in a market where queues run years, an Nvidia allocation, and enough credibility to sign decade-long take-or-pay commitments. That combination is hard to assemble and easy to lose. The standardized hardware cuts the other way: no operator can charge much of a premium for the same chip, so competition runs through availability, price and financing terms.

The most important adverse fact is that the scarcity underpinning this structure is being relieved by CoreWeave's own largest potential competitors. As hyperscalers bring purpose-built AI capacity online, the overflow demand that seeded the neoclouds should shrink, and William Blair expects subscale players to be consolidated or squeezed out. Whether CoreWeave's contracted commitments translate into a durable position once power stops being the binding constraint — or whether they simply reflect a shortage that is now being fixed — is the question the rest of the analysis turns on.

02

Competitive advantage

Does the company have a moat? Moderate at best, and the rent is already shrinking

CoreWeave earns a landlord's gross margin and almost no operating profit. Gross margin ran at 66.6% in the first nine months of 2026 against operating income of $0.02 billion for the same nine months. That gap frames the whole question. What reads as pricing power is a scarcity rent on powered, Nvidia-equipped data-center capacity, and the rent is being competed down as that capacity gets built. CoreWeave rates Moderate.

A moat means above-normal returns protected by something a rival cannot buy. CoreWeave's returns are not above normal, and its protection, capital plus power contracts plus an Nvidia allocation, is purchasable by anyone with a cheaper balance sheet. The seller list is short enough to look concentrated: William Blair's September 30, 2026 report on the neocloud segment names CoreWeave, Nebius, IREN, Nscale, Crusoe and Lambda, and counts about $275 billion of customer commitments across four of them. Six scaled sellers, plus three hyperscalers building competing capacity of their own, is not a structure that keeps newcomers out for long.

Market position: Fairly Strong on ranking, unmeasured on share

Market position: Fairly Strong. William Blair ranks CoreWeave and Nebius as the two best-positioned neoclouds, citing their record of securing power and data-center capacity, financing it and delivering at scale. Company-reported revenue reached $5.13 billion in FY2025 and $8.32 billion in the nine months to September 30, 2026. Against William Blair's estimate of roughly $25 billion of neocloud revenue in 2026, an annualized run rate near $11 billion implies a share in the low-to-mid 40s. Treat that as an estimate: it assumes both measures cover the same population, and the segment definition may not match the company's reported revenue.

USD bnFY2023FY2024FY20259M 2026
Revenue0.231.915.138.32
Operating income(0.01)0.32(0.05)0.02
Net income(0.59)(0.86)(1.17)(2.08)
Operating cash flow1.832.753.065.64
Capital spending (est.)2.948.7010.3123.56
Total assets4.9817.8349.3091.48

Nine months to September 30, 2026, not annualized. Capital spending estimated as operating cash flow less free cash flow.

Stability: Moderate, and the price trend is the evidence

Stability: Moderate. The company listed in March 2025 and its reported history runs from FY2022 through FY2025 plus nine months of 2026, so five-year share stability cannot be confirmed from public figures. The pricing record is clearer than the share record. Revenue in the first nine months of 2026 was more than double the $3.55 billion booked in the same period a year earlier, while gross margin over those nine months fell from 73.8% to 66.6%. Growth of that speed alongside margin of that direction is what a market with real entry looks like: buyers are paying up for capacity today, and the price of capacity is already softening.

Profitability: Weak, with interest absorbing the spread

Profitability: Weak. In FY2025 the company reported operating income of −$0.05 billion and a small tax benefit, so NOPAT was at or below zero before any adjustment. Invested capital at December 31, 2025 comes to about $21.2 billion: equity of $3.33 billion plus debt of $21.84 billion less cash of $3.95 billion, excluding lease obligations that would enlarge the denominator. Estimated after-tax ROIC is therefore negative, against a benchmark of 15%-25% for clear evidence of advantage and 6%-8% for none. Return on assets, the cleaner measure given how thin equity is, ran at −11.9% in FY2023, −7.6% in FY2024 and −3.5% in FY2025, then about −3.0% over the first nine months of 2026 on average assets, not annualized. ROE says nothing useful here: equity was negative at the end of FY2024 and stood at $4.03 billion on September 30, 2026 against $87.46 billion of liabilities.

The operating leverage that scale is supposed to deliver has gone to lenders. Interest expense was $2.03 billion in the first nine months of 2026 against operating income of $0.02 billion, or about 24 cents of every revenue dollar. Depreciation is larger still: EBITDA of $2.47 billion against operating income of −$0.05 billion in FY2025 implies roughly $2.5 billion of depreciation and amortization, about half of that year's revenue (estimated).

Margins: Moderate, wide gross line, negative net line

Margins: Moderate. The gross line is high because renting expensive machines is a pass-through business: revenue carries the cost of chips, power and depreciation, and 67% to 74% gross margin on that base is respectable. It is also where competition shows up first, and it peaked at 74.3% in FY2024, easing to 71.7% in FY2025. Net margin went from −45.1% in FY2024 to −22.7% in FY2025 and −25.0% over nine months of 2026, an improvement that stalled as the asset base and interest bill grew. Costs booked below the gross line came to about $3.7 billion in FY2025, roughly 71% of revenue (estimated from reported gross profit and operating income), which is why a 71.7% gross margin produced no operating profit at all.

Overall rating: Moderate, because returns do not show protection yet

Overall rating: Moderate. The ranking is real; the returns are not. With profitability rated Weak, the overall rating cannot rise above Moderate however strong the market position looks, and the two decisive figures point the same way: estimated ROIC at or below zero on roughly $21.2 billion of invested capital, and a gross margin that gave back 7.2 percentage points year over year while revenue doubled. The barrier CoreWeave holds protects access to capacity, not the price of it.

Where the advantage comes from and how long it can last: a purchased edge with a closing window

In a segment where every operator buys the same silicon from the same vendor, only two things can be defended: access to power and the capital to turn it into live clusters, and the willingness of large buyers to sign for years. Scale is the product of combining them. CoreWeave has genuine standing in both and owns neither outright. The single metric that tests the moat is the spread between revenue per megawatt and the fully loaded cost of a megawatt, meaning depreciation plus interest; gross margin is the closest public proxy for it.

Supply-side advantage: power and capital, both for sale

Supply-side advantage: Moderate. CoreWeave has no proprietary process. It buys the same accelerators, pays market rates for power, and leases or builds shells that competitors replicate; the know-how is scheduling, deployment speed and operating fleets at scale, all of which are learnable. What it had from 2023 through 2025 was a head start in signing power and bringing clusters live while demand outran supply, and that head start showed up as a 74.3% gross margin in FY2024. The edge is capital intensity itself, which is the easiest kind of edge to compete away, because the largest buyers hold cheaper capital than the seller does.

The arithmetic of the spread is unforgiving. Capital spending came to an estimated $23.56 billion in the first nine months of 2026 against $8.32 billion of revenue, roughly $2.83 of investment per revenue dollar, and the resulting fleet carries about $2.5 billion a year of depreciation at the FY2025 revenue base. Interest adds 24 cents per revenue dollar. Customers stay because capacity is available now and switching means waiting in an interconnection queue, which is a reason to sign this quarter and a weaker reason to renew in three years. The adverse outcome to follow is the gross line: it has already slipped 7.2 percentage points year over year, and if it keeps falling the spread over depreciation and interest closes. At that point the advantage is gone regardless of how many megawatts are under contract.

Demand-side advantage: contracts buy time, not loyalty

Demand-side advantage: Weak. The $275 billion of commitments booked across four neoclouds proves buyers will sign multi-year take-or-pay deals. Those deals protect the revenue line while capacity is scarce. They do not create captivity. The workload is portable, the hardware is standard, and the buyer is a frontier lab that benchmarks everything it runs; there is no network effect, because one customer's cluster does not make another customer's cluster more valuable. A contract fixes a quantity and a term. At renewal the buyer weighs CoreWeave against its own data centers and against three hyperscalers with lower costs of capital.

Pricing power would show as stable or rising gross margin on renewing capacity. So far the reverse has happened as capacity grew, which is direct evidence against captivity. The one place switching costs could be built is software: the $1.1 billion of goodwill on the September 30, 2026 balance sheet suggests acquisitions intended to put a layer above raw compute, and such a layer could eventually raise the cost of leaving. It is early, and it has not yet changed the margin trend. The contract also cuts both ways. CoreWeave owes the capacity it promised, so a customer whose own buildout arrives early can slow new orders while the asset stays on CoreWeave's books. What would break this source, and what to follow, is the price per megawatt on the first large renewals; a renewal at a lower rate per unit would confirm the buyer, not the seller, holds the leverage.

Economies of scale: secured megawatts are an option that costs money

Economies of scale: Moderate. Scale here means relative share of secured power in a handful of constrained grids, not national share of a measured market. Interconnection queues run years, substations and shells are lumpy, and a new entrant must assemble capital, an Nvidia allocation and a grid position before earning a dollar. That is a real barrier to subscale entry and the reason the seller list is short. William Blair counts more than 18.5 GW contracted or secured across the leading neoclouds.

CoreWeave's annualized revenue near $11 billion, at the industry yield William Blair puts near $12.5 million per active megawatt per year, implies something under 1 GW actually online (estimated). Active and secured capacity are different measures, so read that as an order of magnitude: most of the promised capacity still has to be financed. Total liabilities of $87.46 billion against $91.48 billion of assets leaves a thin equity cushion for that build. Scale without captivity does not keep a rival out, because a hyperscaler with a lower cost of capital can buy the same megawatts in the same grids. The adverse outcome to watch is secured capacity that sits unleased, or debt refinanced at a rate above the spread the contracts earn.

Durability: the shortage that pays for this is being fixed

The most important substitute is the customer's own capacity. If a frontier lab trains on accelerators it owns, or a hyperscaler absorbs the overflow demand that seeded this segment, CoreWeave's addressable market shrinks even while existing contracts run. In-house chips at the large cloud providers are the concrete route for that substitution, and a five-year contract does not stop a buyer from placing new workloads elsewhere when it expires. William Blair expects subscale neoclouds to be consolidated or squeezed, and sets the 18.5 GW already contracted or secured against an assumption of about 20 GW powered industry-wide by 2031, a forecast that implies the land-grab phase is closer to its end than its beginning.

Three things are commonly mistaken for protection here: brand, first-mover speed, and access to capital. All three are operating edges that get copied once capital is available, and all three describe CoreWeave's 2023-2025 run better than customer captivity does. The strongest moats combine captivity with scale. CoreWeave has scale and contracts, and the captivity is contractual and time-limited, which is why the rating sits in the middle rather than higher.

Three questions decide whether that changes:

  • Does gross margin hold in the mid-60s through 2027 as the capacity contracted in 2025 and 2026 comes online, or keep sliding toward the cost of capital?
  • At the first large renewals, does the price per megawatt hold, and does the customer renew at the same size?
  • Does the software layer grow large enough to change the switching calculation, and does that show up in gross margin?

Generated by DeepVal AI. Data and models may contain errors or be outdated. For research only; not personalized investment advice.

Continue the comparison

Continue reading the full report

  1. Comparison and summaryHow CoreWeave, Inc. Class A Common Stock compares with its closest rivals, and our moat rating

Free to sign up · 20 research points · unlocking this report uses 10

20 points cover 2 full reports. No card needed.

Research reflects its generation date. Preview the first two modules.

More on CoreWeave, Inc. Class A Common Stock

CoreWeave, Inc. Class A Common Stock

Business, moat and key questions

Open company page