Initiated 2026-10-01 · Price $84.87 (as of 2026-10-01 10:20 EDT, stockanalysis) · Mkt cap ~$46.8bn · Information technology / AI GPU cloud · Model: verified

Rating: Sell — Conviction: High

CoreWeave is growing very fast on real contracts, but renting GPUs is a capital-hungry business: the hardware wears out in about six years, so capex never falls much below depreciation, and the debt that funds the build costs ~9.5%. Once the build-out's cash needs are counted, the Base case and both valuation methods sit below today's price. Only the Bull case, with ~7 GW running at today's pricing, clears it.

Conviction tests (3a-v-c): T1 pass (Base midpoint −68% at +1pp, −59% at −1pp; Sell both) · T2 pass (Gordon $0.00, exit multiple $62.32; both below price) · T3 pass (FY0 facts from EDGAR, 10-K, 10-Q and 8-K; one est. input, the FY2023 NWC placeholder, does not feed value) · T4 pass (no pending deal, ruling or single-contract event outside the cases)

High means the sign survives the tests, not that outcomes are narrow: Bull is +75%, and the tests don't stress the revenue/capex path.

Model value range vs price
Bear $0.00Base $31.16Bull $148.49Price $84.87

Business overview

CoreWeave rents NVIDIA GPU clusters, plus storage and networking, to AI labs, hyperscalers and enterprises, mostly on multi-year take-or-pay contracts that are often partly prepaid. Revenue went from $229mm (FY2023) to $5.13bn (FY2025); FY2026 guidance is $12.4–13.2bn. Revenue backlog was ~$104bn at 6/30/26, with $25bn+ more signed early in Q3. Active power was 1.5 GW and contracted power 3.7 GW (4.2 GW after the quarter). The company leases most data-center shells and owns the GPUs. Earnings turn on three things: how fast power comes online (target >1.85 GW at YE2026, ≥8 GW by 2030), the price per GPU-hour when each contract is signed and when it rolls, and the cost of the debt behind ~$35–39bn of 2026 capex.

Competition

The main earnings driver is price per GPU-hour, and Nebius is where that pressure shows first. Its H100 on-demand list price is roughly 38–60% below CoreWeave's in 2026 comparisons (Spheron, 2026). Nebius's trailing revenue is $1.36bn against CoreWeave's $7.59bn (stockanalysis, 2026-10-01), but its market cap ($62.9bn) is already above CoreWeave's. Oracle (OCI) and the hyperscalers compete for the same large AI-lab contracts, and Microsoft is both CoreWeave's largest customer (45% of Q1 2026 revenue, down from 72% in Q1 2025, per press) and a builder of its own capacity. Miner conversions (IREN) add low-end capacity. Oracle, the only comparable with mature earnings, trades at 15.9x lease-inclusive EV/EBITDA.

Bull case

  1. Backlog converts on schedule and renews — $129bn+ of signed commitments become revenue as power comes online, and customers renew at scale. Plays out if YE2026 active power clears 1.85 GW and contracted power keeps rising toward the 8 GW target. Model: rev_growth
  2. Pricing power on new GPU generations — Blackwell and Vera Rubin pricing is "at new highs", and SKU prices rose ~25% in July. Plays out if supply of top-bin GPU capacity stays tight through 2027. Model: ebitda_margin
  3. Contract paper earns infrastructure-style valuation — take-or-pay contracts with investment-grade counterparties justify a lower discount rate and a higher terminal multiple. Plays out if spreads on GPU-backed debt keep falling (company: −300bp in a year). Model: wacc, exit_ev_ebitda

Bear case

  1. Capex never stops — GPUs depreciate over ~6 years, so a steady-state fleet needs replacement capex near D&A (~50% of revenue), and EBITDA overstates owner cash flow. Plays out if each GPU generation needs a full rebuy. Model: capex_pct_rev, exit_ev_ebitda
  2. Leverage and refinancing — ~$39bn of debt, $7.5bn of it current, at ~9.5%. Coverage on FY2025 EBITDA is 0.74x. Plays out if credit markets tighten before contracted cash flows mature. Model: wacc; credit.debt (FY0)
  3. Concentration and repricing at roll-off — Microsoft at 45% of revenue, and early Hopper contracts re-price toward commodity rates. Plays out if renewals come in smaller or cheaper. Model: rev_growth, ebitda_margin
  4. Competitive price compression — Nebius, Oracle and hyperscaler in-house capacity cap GPU-hour pricing. Plays out if GPU supply loosens in 2027–28. Model: ebitda_margin, exit_ev_ebitda

Valuation & balance sheet

Metric (definition) Current Own history (range or 5y avg) Peers Source, as-of
EV/EBITDA (model basis: mkt cap − cash + funded debt + timing adj.; GAAP EBITDA; leases excluded) 34.6x FY2025 · ~22x LTM · ~12.5x Base FY2026 n/a (IPO March 2025) n/a on this basis Model; 10-Q 2026-06-30
EV/EBITDA (aggregator, lease-inclusive, trailing) 24.6x n/a ORCL 15.9x · NBIS 258.9x · IREN 470.8x (NBIS, IREN distorted by early-ramp EBITDA) stockanalysis 2026-10-01
EV/Sales (aggregator, trailing) 12.3x n/a ORCL 7.6x · NBIS 49.3x · IREN 25.5x stockanalysis 2026-10-01
FCF (CFO − cash capex, H1 2026) −$10.5bn ($3.66bn − $14.12bn) FY2025 cash capex $11.2bn n/a (unverified) Q2 2026 release

Model-implied value range (from model-summary.json; generic module, Gordon-growth DCF and exit-EV/EBITDA DCF, midpoints): Bear $0.00 · Base $31.16 · Bull $148.49 per share, i.e. implied returns of −100.0% / −63.3% / +75.0% vs $84.87. These ranges show how the bull and bear drivers translate into value; they are not price targets. The price sits about 46% of the way from Base to Bull, so the market is paying for a good part of the Bull path (≈7 GW, pricing held, capex discipline). The exit method ($62.32) embeds 8x Y5 EBITDA, a ~36% de-rating from the ~12.5x paid today on Base FY2026 EBITDA. Gordon ($0.00) embeds far less, because Y5 unlevered FCF is thin once replacement capex runs at ~44% of revenue. The two methods disagree by more than 10%, so the exit method's 8x is itself a re-rating above Gordon's implied multiple. Gordon floors at $0.00 in Base and both methods floor in Bear: Bear is a plateau (~$25bn revenue, 40% margin) where net debt exceeds a positive EV, so a tail sensitivity is moot.

Balance sheet: net leverage 12.6x FY2025 EBITDA (~8.1x LTM). EBITDA/interest 0.74x on FY2025; the Base path is 2.3–2.9x. Liquidity: $5.52bn cash at 6/30 plus ~$3.14bn net from the September convert; restricted cash $1.38bn excluded. Nearest material maturity: $6.24bn recourse plus $1.28bn non-recourse debt due within 12 months of 6/30/26. Then $25.2bn non-current recourse debt (laddered at 2030 in the model, an approximation) and the $3.7bn 2033 convert (conversion $97.85). Ratings: n/a (unverified).

Model note: verified (LibreOffice matched all 3,207 formula cells). Unverified inputs: historicals.nwc (FY2023 is a placeholder; does not feed projections). Assumptions without basis: none. No scenario CHECKs. Cash and debt are at 6/30/26 plus the September convert, so Y1 capex is H2 only (~$22bn est.). H1 GAAP EBITDA ($2,347mm) and the H1 NWC inflow ($3,825mm) are deducted in other_claims to avoid counting them twice. NWC includes all deferred revenue (customer prepayments) and excludes accrued liabilities ($6.4bn, judged mostly capex and interest accruals; split unverified). Debt coupons are blended estimates; guidance comes from a call summary (MarketBeat). No EPS guidance exists; Base EPS stays negative through FY2030. Beta 2.0 is a judgment (no 5-year history).

Scenario stress test

Reasoned from the bull/bear drivers above. The model column comes from the scenario overlays (Base case + shock).

Scenario Effect Mechanism Magnitude Model Δ value vs Base ($/sh)
S1 Fast equity crash − Volatile AI name, ~17% short interest: the multiple de-risks; contracts unaffected over weeks High −$23.04
S2 Slow bear / recession − Take-or-pay revenue holds, but new contracts and renewals stall, spot pricing falls, spreads widen High −$31.16
S3 Rapid rate shock − ~$39bn of debt and a funding-dependent build; back-end-loaded value Med −$18.27
S4 Slow rate grind − Same channel, grinding Med −$9.86
S5 Soft-landing cuts + Cheaper GPU-backed debt and a lower discount rate; growth intact Med +$13.46
S6 Recession-driven cuts − AI-spend pullback and customer stress outweigh lower rates High −$30.46
S7a Credit liquidity shock − $35–39bn 2026 capex depends on open debt markets; a spread gap delays the build High −$21.24
S7b Slow default cycle − AI-lab counterparty risk plus refinancing of high-yield GPU-backed debt High −$29.25
S8 Stagflation − Power, lease and labor inflation against fixed-price contracts, plus higher rates Med −$13.67
S9a Dollar spike 0 No material effect, not modeled (USD revenue, mostly US assets) Low $0.00
S9b Dollar slide 0 No material effect, not modeled Low $0.00
S10 Melt-up + Momentum AI-infrastructure name: melt-up and short covering re-rate it High +$27.64
S11 Energy supply shock − Power partly passed through in colocation leases; mostly a risk-off multiple effect Low −$4.70
S12 Mega-cap/AI derating − The core risk: an AI-capex unwind stalls contracts, collapses GPU-hour pricing at roll-off and de-rates the group High −$31.16

S2 and S12 equal the full Base − Bear gap; downside rows cluster near the $0.00 floor. Currently active/on watch per the playbook: S3 partially active (the 2026-09-30 macro log shows the pace legs crossed too); S8, S10 and S11 on watch. S3 is live here through interest cost and the discount rate. S10 is the main way the price holds above Base.

Model value change vs Base, by scenario
S2 Slow bear / recession−$31.16S12 Mega-cap/AI derating−$31.16S6 Recession-driven cuts−$30.46S7b Slow default cycle−$29.25S1 Fast equity crash−$23.04S7a Credit liquidity shock−$21.24S3 Rapid rate shock−$18.27S8 Stagflation−$13.67S4 Slow rate grind−$9.86S11 Energy supply shock−$4.70S9a Dollar spike$0.00S9b Dollar slide$0.00S5 Soft-landing cuts+$13.46S10 Melt-up+$27.64

What would change the call

Upgrades if: active power reaches ≥1.85 GW by YE2026 and contracted power keeps rising toward 8 GW, and the company shows positive CFO-minus-capex on the mature (pre-2025) fleet, i.e. replacement capex well below D&A. Or debt cost falls below ~7%. Downgrades if: n/a — already Sell. The call would firm if FY2026 revenue misses the $12.4bn guidance low end, Microsoft's share rises again, or GPU-hour pricing on renewals falls.

Watch items

Sources