The backlog has never been larger. The capital burn has rarely been more terrifying.
Revenue is compounding at 130% and a record $99B backlog secures years of demand — yet CoreWeave trades near its 52-week low because the market is pricing one question: do hyperscalers depend on CoreWeave's grid, or build around it? Four analyst lenses, three scenarios, four time horizons.
Gray line = CoreWeave's actual price into today (IPO Mar '25 at $40 → $153 peak Oct ’25 → $71.77 now); colored paths = synthesized scenario midpoints forward, probability-weighted (base 50% · bull 30% · bear 20%). Log-linear, mid-year marks. Wall Street 12-month consensus ≈ $138 (range $90–$160, mostly Buy-rated).
Re-weight the scenarios
Those probabilities are a judgment call based on how hyperscaler self-sufficiency plays out. Drag to set how likely the bear and bull cases are (base takes the remainder); the blended target below, the dotted line on the chart, and the prob-weighted row of the scenario cards all update live.
Four analyst lenses, four answers
The same $99B backlog and $32.5B CapEx bill support wildly different conclusions depending on which framework you trust. Each lens below is a synthesized expert perspective.
The Backlog Believer
The market is hyper-ventilating over Meta's internal cloud plans, ignoring the $99.4B backlog giving unprecedented multi-year visibility. Q1'26 revenue just grew 130% YoY. Nvidia isn't just a supplier; they own 7% of CoreWeave and allocate prime silicon to them to keep hyperscalers in check. Growth of this magnitude deserves a premium, not a panic.
The Physical Grid Thesis
CoreWeave is no longer just a GPU reseller; they are a physical infrastructure company. With over 1 GW of active power and 3.5 GW contracted, they’ve cornered scarce energy resources that take hyperscalers years to permit and build. Even if Custom Silicon (ASICs) take share, the underlying data center space and power will still be leased from CoreWeave.
The Hyperscaler Squeeze
Meta Compute is the canary in the coal mine. AWS (Trainium), Google (TPU), and Meta are moving to own their compute stack. CoreWeave is an expensive middleman riding a temporary GPU scarcity wave. When supply normalizes, their pricing power collapses—leaving them stranded with $17.5B in debt and $32.5B in GPUs depreciating at warp speed.
The CapEx Skeptic
$32.5B in estimated 2026 CapEx against a ~$40B market cap is terrifying. The company is burning cash aggressively to fund data center builds, evidenced by the new $8.5B delayed-draw term loan. While EBITDA looks positive ($1.04B), massive interest expenses and depreciation drag the bottom line to a -36% margin. The debt load leaves zero room for error.
Wall Street 12-month price targets
What the sell-side expects over the next year. Bars are sorted low to high; the dashed line is today's $71.77. The sell-side remains overwhelmingly bullish despite the recent "neocloud" selloff.
A selection of recent analyst ratings from the ~37 firms covering CoreWeave. The consensus maintains a $138 price target, indicating massive theoretical upside. The gap between the $71.77 live price (dashed line) and the lowest Street target ($90) illustrates the severity of the retail and institutional panic relative to sell-side models.
Where the grid leads
Synthesized scenario midpoints (mid-year). Returns shown vs. today's $71.77. These are illustrative frameworks, not predictions with certainty — five-year outcomes hinge almost entirely on hyperscaler behavior.
1 Year
Mid-20272 Years
Mid-20283 Years
Mid-20295 Years
Mid-2031▸ Bull case — show the assumptions & math
▸ Base case — show the assumptions & math
▸ Bear case — show the assumptions & math
Revenue, capex, free cash flow & debt ($B)
The reality of AI infrastructure. Revenue is scaling exponentially, but the CapEx required to buy GPUs and secure power is astronomical, driving deep negative free cash flows (FCF).
The AI infrastructure arms race visualised: CoreWeave's revenue (blue) is surging, but CapEx (clay) to buy Nvidia chips and secure data center space is accelerating faster (est. $32.5B in 2026). This creates an immense cash burn (FCF, olive, going negative), funded by an exploding debt pile (slate). The bull thesis requires CapEx to taper in 2027+ so FCF can turn positive; the bear thesis is that they run out of borrowing capacity first. Figures illustrative based on Q1'26 guidance.
EPS path underpinning the targets ($)
CoreWeave is currently unprofitable on a GAAP basis due to massive depreciation and interest expense. Target prices assume they crest the CapEx curve and begin generating positive EPS by 2027.
Adjusted (non-GAAP) EPS. Gray = reported, olive = estimates. The widening loss in 2026 reflects peak CapEx/depreciation; the base case thesis assumes massive infrastructure investments hit their payback window starting in late 2027, scaling to ~$9.50 EPS by 2031 to justify the $165 Base scenario target.
The business isn't slowing down
Q1 FY26, year-over-year — read these against a stock sitting 53% below its highs.
Every line is aggressively green — capacity is expanding +150%, revenue +130%, and the forward-looking backlog (clay) has tripled. Yet the stock trades at its 52-week low. That gap is the bull's entire case in one chart: the physical business execution is flawless; the multiple compressed strictly on hyperscaler fear.
Bull vs. Bear
The entire valuation argument compresses into one disagreement: is CoreWeave the foundational utility grid of the AI era, or a temporary hardware middleman?
▲ THE BULL CASE
- The $99.4B Backlog. Unprecedented multi-year revenue visibility, driven by massive, long-term contracts (e.g., the $21B Meta deal extending to 2032).
- Nvidia's Proxy. Nvidia holds a ~7% stake and grants CoreWeave preferential allocation of H200s/Blackwell chips to prevent hyperscalers (AWS/Azure) from gaining too much structural power.
- Real Physical Moats. GPUs are easy to buy; power is hard to find. CoreWeave’s 1 GW of active power and 3.5 GW contracted capacity gives them a massive head start over new entrants.
- Hyper-Growth is Intact. Q1'26 revenue of $2.08B (+130% YoY), easily beating estimates.
- Financing is Secured. An $8.5B delayed-draw term loan validates their business model to conservative institutional debt markets, ensuring they can fund the massive CapEx requirements.
▼ THE BEAR CASE
- The Hyperscaler Pivot. Reports of "Meta Compute" building internal cloud infrastructure terrified the market. If big tech creates their own data centers and custom chips, CoreWeave’s TAM collapses.
- Crushing Capital Intensity. $30B-$35B in 2026 CapEx means the company bleeds tens of billions in free cash flow, leaving it utterly dependent on debt markets.
- Customer Concentration. A massive portion of that backlog comes from a handful of players (Meta, Microsoft, Anthropic). Losing one breaks the model.
- Deep Negative Margins. A -36% net income margin driven by crippling depreciation and interest expenses; operating leverage is elusive when the assets (GPUs) have a 2-3 year shelf life.
- Commoditization Risk. Once the frantic GPU shortage normalizes, compute becomes a commodity. Premium pricing evaporates, squeezing CoreWeave's ability to service its $17.5B debt.
Risk map — likelihood × impact
Where each risk sits, not just how big it is. The hot upper-right cell — hyperscalers bypassing CoreWeave with internal compute — is the single variable driving the current selloff.
- Interest expense inflation
- Power grid constraints
- Hyperscaler internal compute shift
- Debt covenant pressure
- GPU utilization squeeze
- Meta contract renegotiation
- Nvidia allocation shifts
- AI model efficiency breakthrough (less compute needed)
Hyperscaler compute shift
Meta, Google, and Amazon aggressively deploy their own custom silicon (ASICs) in proprietary data centers, bypassing third-party GPU clouds.
GPU utilization squeeze
As the GPU supply shortage normalizes, pricing power plummets, leaving CoreWeave unable to generate expected returns on its $30B CapEx investments.
Meta contract renegotiation
Meta uses its leverage (a massive chunk of the $99B backlog) to force price concessions, shattering CoreWeave's margin assumptions.
Power grid constraints
Local utility delays and transmission constraints delay new data center activations, pushing back revenue recognition.
Debt covenant pressure
With $17.5B in debt, a slight miss in EBITDA targets could violate lending terms, forcing highly dilutive equity raises.
Interest expense inflation
As debt balances swell to fund CapEx, rising variable rates or higher risk premiums chip away at already negative net margins.
Nvidia allocation shifts
Nvidia decides it needs the hyperscalers more than it needs pure-plays, shifting prime hardware allocations away from CoreWeave.
Model efficiency breakthrough
New AI architectures dramatically reduce the compute power required for training, shrinking the overall TAM for GPU clouds overnight.
The jargon, decoded
Hover the dotted terms in the metrics, or scan the desk's working definitions here.
- Revenue Backlog
- Contracted, non-cancelable future revenue. CoreWeave has an astronomical $99.4B locked in, heavily weighted toward long-term hyperscaler commitments.
- Hyperscaler
- Massive cloud providers like AWS, Microsoft Azure, and Google Cloud, which operate at enormous scale and frequently design their own silicon.
- Custom Silicon (ASICs)
- Application-Specific Integrated Circuits. Chips designed purely for one task (like Google's TPUs). The primary long-term threat to general-purpose GPUs.
- CapEx (Capital Expenditures)
- Money spent on physical assets. For CoreWeave, this means buying thousands of Nvidia H200s and leasing/building data centers ($32.5B in 2026).
- Active Power Capacity
- The amount of electrical grid power (measured in Gigawatts) fully permitted, wired, and ready to feed servers. The true physical bottleneck in AI.
- Delayed-Draw Term Loan
- A debt facility where the borrower can take portions of the money as needed rather than all at once—crucial for funding phased data center construction.
- Depreciation
- The accounting process of spreading the cost of an asset over its useful life. Since GPUs obsolesce quickly (2-4 years), this is a massive hit to net income.
- Disintermediation
- The middleman getting cut out — hyperscalers building their own AI data centers instead of renting from CoreWeave.