CoreWeave has solved the hardest problem in AI infrastructure — demand — and turned it into a $99.4B contracted backlog with take-or-pay terms and roughly five-year weighted contract life. What remains is the second-hardest problem: converting 3.5 GW of contracted power into energized megawatts on $31–35B of 2026 capex, funded almost entirely with debt, while the credit market reprices the risk in real time. Five analyst lenses, three scenarios, four horizons.
Gray line = CoreWeave’s actual closing price from the 28 Mar 2025 IPO at $40 ($33.52 low that April, $183.58 all-time-high close on 20 Jun 2025, $153.20 intraday 52-week high on 10 Oct 2025, $60.55 intraday 52-week low on 29 Jul 2026) into Friday’s $71.77 close. Colored paths are synthesized scenario midpoints, not forecasts, probability-weighted at base 40% · bull 25% · bear 35%. Mid-year marks on one continuous axis; the stock has only ~16 months of trading history, so the left side is short by construction. Wall Street’s 12-month consensus is $138.03 (37 analysts, range $36–$303, “Buy”) as of 31 Jul 2026.
Those probabilities are a judgment call — so make them yours. This desk starts bear-tilted (35/40/25) because the downside here is a capital-structure outcome, not a multiple outcome. 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.
CoreWeave measures itself in megawatts, so this report does too. Revenue is only recognised when a data hall is energized and handed to the customer — but lease, power and depreciation costs start the moment a powered shell arrives. Every dollar of the thesis lives in the gap between the olive bar and the blue one.
Each lens below was reasoned independently before the others were read, so they genuinely disagree — the spread runs from $34 to $150 on the same set of facts. These are synthesized frameworks, not real analysts or real firm ratings.
Buy the megawatt schedule, not the income statement. Revenue rose 112% to $2.08B in Q1’26 while backlog jumped ~50% sequentially to $99.4B on the biggest bookings quarter in company history. Management has raised exit-2026 ARR to $18–19B and reiterated exit-2027 ARR above $30B with more than 75% already contracted. At 2.5× forward sales for a business compounding triple digits with take-or-pay cover, the multiple is the cheapest it has been since the IPO.
The moat is not GPUs — anyone can buy those. It is the combination of secured power, NVIDIA reference status, and a software layer (Mission Control, Weights & Biases, Serverless RL) that hyperscalers rent rather than replicate under deadline. All four leading labs are now customers, and top-two revenue concentration fell from 72% to 65% year on year while non-investment-grade exposure dropped below 30%. Contract length extended from four years to five. That is a franchise forming, not a rental business.
This is a credit story wearing an equity ticker. Interest expense hit $536M in a single quarter, up 103% year on year, against $21M of adjusted operating income. Total debt is $35.1B against $2.3B of cash and $4.8B of book equity — 7.4× debt/equity. In July the company had to sweeten a $2.6B JPMorgan-led loan to as much as 5.5% over benchmark at 97 cents on the dollar, and default-protection costs rose about 55% in the month. The equity is the thin residual tranche of a leveraged infrastructure vehicle.
The whole model rests on a six-year useful-life assumption for GPUs whose resale market does not yet exist beyond the first contract term. Take-or-pay protects the initial five years; nothing protects year six onward. Meanwhile the customers are becoming the competitors — reports that Meta may expand into external cloud knocked 14% off the stock in a single session on 1 Jul 2026. GAAP losses widened to $740M, FCF is −$10.6B trailing, and the share count is up 122% in a year. When financing tightens, dilution is the only remaining lever.
Price is 53% below the October 2025 high, below both the 50-day ($93.75) and 200-day ($94.83) averages, with RSI at 41.6 — oversold but not washed out. Short interest is 11.8% of shares and 20.5% of a 314M float at just 2.2 days to cover, which is exactly what produced the 21.5% single-session rally on 30 Jul. Realised daily ranges above 10% are now routine. Position sizing, not direction, is the dominant risk here; the stock has no beta history and no earnings floor.
What the sell-side expects over the next year. Bars are sorted low to high and colored by rating; the dashed line is Friday’s $71.77 close. Note that the single lowest bar and the single highest bar are separated by a factor of two and a half — this is not a stock the Street agrees on.
Ten named desks, dated between 11 May and 30 Jul 2026 — a selection of the 37 firms covering CoreWeave, averaging $126 versus the full consensus of $138.03. The rating split across all 37 as of July: 21 strong buy, 5 buy, 9 hold, 1 sell, 1 strong sell. Every bar except Bernstein’s sits above the current price, and the widest disagreement on the Street — a $36 low against a $303 high — is itself the finding: sell-side models diverge because they disagree about the discount rate and the terminal value of a GPU, not about the revenue.
Synthesized scenario midpoints at mid-year, returns shown against Friday’s $71.77 close. These are illustrative frameworks, not predictions. The spread is unusually violent because a leveraged balance sheet converts a modest change in EBITDA or exit multiple into a very large change in residual equity value — that asymmetry is the CoreWeave story.
This is the single most important chart in the report. Read it left to right and the business looks like a rocket. Read it top to bottom in the 2026 column and you can see exactly what the bear is worried about: the clay bar is more than twice the blue one, and the gap is filled with borrowed money.
Revenue (sky) compounds from $229M in 2023 to a guided $12–13B in 2026 — the fastest cloud in history to $5B of annual revenue. But capex (clay) has exceeded revenue in every single year, and free cash flow (olive, below the line) gets more negative as the business gets bigger: −$1.1B, −$6.0B, −$7.3B, and a consensus −$26.3B for 2026 against company capex guidance of $31–35B. Total debt (slate) went $2B → $10.6B → $29.8B at successive year-ends. Two caveats on the 2026 column: revenue and FCF are analyst consensus, capex is the midpoint of company guidance, and the debt bar is the last reported balance ($35.1B at 31 Mar 2026), not a year-end estimate — with most of the year’s capex still to fund, the actual year-end figure will be materially higher. FY capex here is the cash-flow-statement figure; management’s 2025 capex number including finance leases was $14.9B.
CoreWeave has no positive EPS to build a P/E on and consensus does not expect one before 2028, so every price target on this page — ours and the Street’s — is really an EBITDA estimate times an exit multiple, minus a large and growing net debt balance. Here is that ladder.
Gray = company-reported adjusted EBITDA ($1.219B in FY2024, $3.093B in FY2025 at a 60% margin). Olive = this desk’s base-case estimates, not consensus: 2026E applies a ~55% margin to the $12.6B revenue consensus, consistent with Q1’26’s reported 56% and management’s guidance that margin troughs in Q1 then ramps. Beyond 2027 the ladder assumes deceleration from triple-digit growth toward the 20–30% range with margin drifting down as the mix shifts from scarce training capacity to competitive inference. The base-case 5-year target is this ladder’s $29B of 2031 EBITDA at a 9× exit multiple, less roughly $105B of net debt. Two things this chart hides, deliberately: adjusted EBITDA excludes stock compensation ($153M in Q1’26 alone) and sits entirely above interest expense, which ran $1.229B for FY2025 and $536M in Q1’26 by itself.
Most growth scorecards separate steady core lines from faster frontier ones. CoreWeave’s honest version separates what the business earned from what the business owes — because both are growing triple digits, and only one of them is in the bull deck.
Periods differ by line and are stated here rather than smoothed over. Q1 FY2026 vs Q1 FY2025: revenue $2.078B vs $982M (+112%), adjusted EBITDA $1.157B vs $606M (+91%), interest expense $536M vs $264M (+103%), revenue backlog $99.4B vs roughly $25B (“close to 4×” per management). Shares outstanding +122% over the trailing twelve months to 545.6M as of 31 Jul 2026. Total debt is FY2025 vs FY2024 year-end ($29.8B vs $10.6B, +181%); the 31 Mar 2026 balance was $35.1B. The bull reads the top and bottom bars — demand is real and contracted. The bear reads the middle: the equity holder’s claim on that demand is being diluted and subordinated at least as fast as the demand is arriving.
The entire argument compresses into one disagreement: is a $99.4B contracted backlog an asset that de-risks $35B of debt, or is it the reason that debt exists?
Placed over a three-to-five-year horizon. Note the shape: unlike most growth names, CoreWeave’s hot corner is already occurring — financing costs are rising now, not in some hypothetical future. The genuine tail risk sits at the bottom right.
What breaks: each 100bp of spread on a $35B-and-growing stack consumes roughly a third of a billion dollars of annual EBITDA before a single new megawatt is built.
What breaks: Microsoft, Meta or OpenAI declines to renew at term, and the backlog stops being a growth asset and starts being a runoff schedule against fixed lease obligations.
What breaks: the six-year useful life proves optimistic, non-cash charges jump, covenants tighten, and the asset backing the collateralized debt is worth less than the debt.
What breaks: an AI capex retrenchment closes the private-credit window entirely mid-buildout. With $31–35B of committed 2026 capex and no maturities until 2029, this is survivable for quarters, not years.
What breaks: the 1.7 GW end-2026 active-power target slips a couple of quarters, ARR guidance resets, and the market stops trusting the conversion schedule that the whole valuation rests on.
What breaks: lease, power and depreciation costs start when a powered shell arrives but revenue starts when it is handed over — adjusted operating margin already fell to 1% in Q1’26 from 17%.
What breaks: preferential GPU allocation, Exemplar Cloud status and a $2B equity stake are a relationship, not a contract. If NVIDIA spreads its favour, the technical edge narrows fast.
What breaks: capex per megawatt rises. Management has already raised the low end of capex guidance on component costs and is exploring derivatives to hedge memory prices.
What breaks: a pending shareholder class action over post-IPO disclosures is a distraction and a modest cash cost, but is unlikely to be thesis-determining on its own.
Hover the dotted terms in the metrics strip above, or scan the desk’s working definitions here.