
Contracted AI scale meets high capital costs; initiate at Hold with a $48 target
We initiate coverage of CoreWeave at Hold with a $48 twelve-month target (−0.1% total return from the $47.83 close on 18 November). CoreWeave runs a purpose-built artificial intelligence (AI) cloud across 52 data centers and 1.7 gigawatts (GW) of active power. A $133bn backlog anchors rapid growth, yet high leverage and 2027–2028 pricing risk balance risk and reward.
- Backlog doubles revenue, but renewals decide value. Committed contracts (98% of third-quarter sales) lift FY27E revenue to $27bn. Each $0.5m shift in fleet revenue per average active megawatt (MW), defined as annual sales per megawatt of live power, moves discounted cash flow (DCF) value by $37 per share.
- Secured debt covers 2027, leaving discount rate decisive. Delayed draw term loans (DDTLs) and $9bn of near-term prepayments fund FY27E spending without unsecured bonds. Still, a 1-point move either way in weighted average cost of capital (WACC) shifts our target from Buy ($77) to Sell ($25).
- Lower capex lifts FY28E profit, while refresh caps the multiple. Our FY27E–FY28E capital spending sits $27bn below consensus, lifting FY28E adjusted operating income 9% above Street. Six-year graphics processing unit (GPU) refresh then caps base DCF at $41.
What the market is missing: Consensus models $99bn of FY27E–FY28E capital spending. Our committed-backlog path ($71bn) trims depreciation and lifts FY28E operating income, while six-year refresh limits terminal cash conversion.
Catalysts: Planned $4bn+ investment-grade DDTL at 7%–7.5% in December 2026, FY27 guidance and renewals on over 130,000 paid-off GPUs in late February 2027, and rival funding delays in 1H27.
Valuation and target: Probability-weighted DCF value is $42 (25% bear nil, 55% base $41, 20% bull $98), rolled 12 months at 13.3% cost of equity to our $48 target. Cross-check 14.0x FY28E enterprise value to earnings before interest and taxes (EV/EBIT) yields $54.
Risks: Beta at 3.71 (−$35/sh), a 2027–2028 supply wave (−$37/sh), 54% top-two customer share, and a Texas permitting pause.
| FY Dec, $m | FY24A | FY25A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| Revenue | 1,915 | 5,131 | 13,214 | 26,862 | 42,625 |
| Growth y/y | 736.2% | 167.9% | 157.5% | 103.3% | 58.7% |
| Adj. EBITDA | 1,219 | 3,093 | 8,108 | 16,940 | 28,294 |
| Margin | 63.7% | 60.3% | 61.4% | 63.1% | 66.4% |
| Net income (GAAP) | (863) | (1,167) | (2,585) | (2,784) | 1,426 |
| EPS, GAAP ($) | (4.30) | (2.81) | (4.72) | (4.87) | 2.42 |
| EPS growth y/y | n.d. | NM | NM | NM | NM |
| FCF ex-prepay | n.d. | (15,961) | (34,245) | (28,630) | (11,830) |
| At $47.83 | FY24A | FY25A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| P/E (GAAP) | NM | NM | NM | NM | 19.7x |
| EV / EBITDA | 67.9x | 26.8x | 10.2x | 4.9x | 2.9x |
| P/B (dil. cap) | NM | 8.8x | 7.9x | 8.3x | 4.7x |
| FCF yield ex-p. | n.d. | −54.6% | −117.2% | −98.0% | −40.5% |
| Dividend yield | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
| ROE (GAAP) | NM | −79.9% | −73.7% | −77.5% | 29.3% |
| Net debt/EBITDA | 4.8x | 5.6x | 5.1x | 3.6x | 2.5x |
| P/E at target | NM | NM | NM | NM | 19.7x |
Source: CoreWeave filings (FY24A–FY25A); Penmark Research model (FY26E–FY28E). EV of $82.8bn uses 611.1m diluted shares and includes $11.9bn customer prepayments. Abbreviations: EBITDA = earnings before interest, taxes, depreciation & amortisation; EPS = earnings per share; FCF = free cash flow; GAAP = U.S. generally accepted accounting principles; P/E = price/earnings; P/B = price/book; ROE = return on equity.
Base DCF of $41 and a $48 target leave little upside; bull $98, bear nil
Source: Signals daily closes (to 18 Nov 2026); Penmark Research model. Scenario values ($98.08 / $41.02 / $0.00) rolled 12 months at 13.3% cost of equity.
Renewals lift fleet pricing to $15.0m/MW by FY30E then fade to $13.7m; active power hits 6.5GW; capex $35m/MW; 12m target $111.11 (+132%).
Fleet pricing reaches $14.3m/MW by FY30E then fades to $13.5m; 1.08GW added in FY27E at $36m/MW; 12m target $46.47 (−3%).
Supply wave caps pricing near $11–12m/MW; fleet runs off from FY29E without refresh; run-off EV ($31.8bn) sits below $53.6bn claims.
| Income stmt, $m | FY24A | FY25A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| Revenue | 1,915 | 5,131 | 13,214 | 26,862 | 42,625 |
| Cash op. costs | (696) | (2,038) | (5,106) | (9,922) | (14,331) |
| Adjusted EBITDA | 1,219 | 3,093 | 8,108 | 16,940 | 28,294 |
| D&A | (863) | (2,454) | (6,779) | (13,186) | (19,126) |
| Adj. EBIT | 356 | 666 | 1,375 | 3,754 | 9,169 |
| Interest, net | (361) | (1,229) | (3,122) | (5,269) | (6,464) |
| Net income | (863) | (1,167) | (2,585) | (2,784) | 1,426 |
| Memo: SBC | (31) | (630) | (643) | (940) | (1,279) |
Source: CoreWeave filings; Penmark Research model. SBC = stock-based compensation (excluded from Adj. EBIT).
| Cash flow, $m | FY24A | FY25A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| Net income | (863) | (1,167) | (2,585) | (2,784) | 1,426 |
| D&A + SBC | 894 | 3,084 | 7,422 | 14,126 | 20,404 |
| Prepay, net | 2,049 | 4,174 | 8,892 | 7,675 | 1,756 |
| WC & other | 669 | (3,033) | (1,082) | (1,092) | (1,261) |
| Operating CF | 2,749 | 3,058 | 12,647 | 17,924 | 22,326 |
| Capex (co.) | (8,702) | (14,845) | (38,000) | (38,880) | (32,400) |
| FCF incl. pp. | n.d. | (11,787) | (25,353) | (20,956) | (10,074) |
| FCF ex-pp. | n.d. | (15,961) | (34,245) | (28,630) | (11,830) |
Source: CoreWeave filings; Penmark Research model.
| Bal. sheet, $m | FY24A | FY25A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| Cash & sec. | 2,064 | 4,164 | 4,000 | 4,000 | 4,000 |
| PP&E, net | 11,915 | 30,557 | 61,778 | 87,472 | 100,747 |
| Leases & oth. | 3,854 | 14,581 | 19,949 | 21,041 | 22,302 |
| Total assets | 17,833 | 49,302 | 85,727 | 112,514 | 127,049 |
| Total debt* | 7,926 | 21,373 | 45,380 | 64,662 | 74,737 |
| Prepayments | 4,064 | 8,185 | 17,302 | 24,977 | 26,732 |
| Oth. liab. | 6,257 | 16,409 | 19,365 | 19,365 | 19,365 |
| Equity | (414) | 3,335 | 3,680 | 3,510 | 6,215 |
Source: CoreWeave filings; Penmark model. *FY24A–25A carrying; FY26E–28E principal; FY24A oth. liab. includes $1,722m preferred stock.
| KPIs & sh. | FY24A | FY25A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| Active YE (GW) | n.d. | 0.85 | 1.88 | 2.96 | 3.86 |
| Avg. GW | n.d. | n.d. | 1.37 | 2.42 | 3.41 |
| Rev/MW ($m) | n.d. | n.d. | 9.68 | 11.10 | 12.50 |
| Opex/MW ($m) | n.d. | n.d. | 3.74 | 4.10 | 4.20 |
| Capex/MW ($m) | n.d. | n.d. | 36.89 | 36.00 | 36.00 |
| Net debt ($m) | 5,862 | 17,209 | 41,380 | 60,662 | 70,737 |
| EPS, GAAP ($) | (4.30) | (2.81) | (4.72) | (4.87) | 2.42 |
| Basic sh. (m) | 218 | 436 | 548 | 572 | 589 |
Source: CoreWeave filings; Penmark model. FY27E includes 35m planned at-the-market (ATM) shares.
Adjusted EBIT margins double from FY26E to FY28E; our FY28E EBIT sits 9% above consensus on lower capex
| Growth, margins | FY24A | FY25A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| Growth y/y | |||||
| Revenue | 736.2% | 167.9% | 157.5% | 103.3% | 58.7% |
| Adj. EBITDA | n.d. | 153.7% | 162.1% | 108.9% | 67.0% |
| Adj. EBIT | n.d. | 87.1% | 106.5% | 173.0% | 144.2% |
| GAAP EPS | n.d. | NM | NM | NM | NM |
| DPS | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
| Margins & coverage | |||||
| EBITDA margin | 63.7% | 60.3% | 61.4% | 63.1% | 66.4% |
| Adj. EBIT margin | 18.6% | 13.0% | 10.4% | 14.0% | 21.5% |
| D&A / revenue | 45.1% | 47.8% | 51.3% | 49.1% | 44.9% |
| FCF ex-p. margin | n.d. | −311% | −259% | −107% | −28% |
| EBITDA / int. (x) | 3.4x | 2.5x | 2.6x | 3.2x | 4.4x |
| Capex / revenue | 454% | 289% | 288% | 145% | 76% |
Source: CoreWeave filings; Penmark Research model.
| Ratios, at $47.83 | FY24A | FY25A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| Returns & leverage | |||||
| ROE (GAAP) | NM | −79.9% | −73.7% | −77.5% | 29.3% |
| ROA (GAAP) | −4.8% | −3.5% | −3.8% | −2.8% | 1.2% |
| ROIC (NOPAT/IC) | n.d. | n.d. | n.d. | 3.1% | 6.7% |
| Net debt / equity | NM | 5.2x | 11.2x | 17.3x | 11.4x |
| Net debt / EBITDA | 4.8x | 5.6x | 5.1x | 3.6x | 2.5x |
| Valuation at $47.83 | |||||
| P/E (GAAP) | NM | NM | NM | NM | 19.7x |
| EV / EBITDA | 67.9x | 26.8x | 10.2x | 4.9x | 2.9x |
| EV / Adj. EBIT | 232.6x | 124.4x | 60.2x | 22.1x | 9.0x |
| FCF yield ex-p. | n.d. | −54.6% | −117% | −98.0% | −40.5% |
| Dividend yield | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
| P/E at target | NM | NM | NM | NM | 19.7x |
Source: CoreWeave filings; Penmark Research model. EV ($82.8bn) includes $11.9bn customer prepayments.
| $m, except EPS ($) | 4Q26E | FY26E | FY27E | FY28E |
|---|---|---|---|---|
| Revenue | ||||
| Penmark estimate | 4,900 | 13,214 | 26,862 | 42,625 |
| Consensus | 4,586 | 12,900 | 26,500 | 42,800 |
| Prior-year period | 1,572 | 5,131 | 13,214 | 26,862 |
| % diff vs consensus | +6.8% | +2.4% | +1.4% | −0.4% |
| Penmark y/y growth | +212% | +158% | +103% | +58.7% |
| Adjusted operating income (EBIT) | ||||
| Penmark estimate | 847 | 1,375 | 3,754 | 9,169 |
| Consensus (pre-3Q) | 538 | 1,066 | 4,070 | 8,429 |
| Prior-year period | 88 | 666 | 1,375 | 3,754 |
| % diff vs consensus | +57.4% | +29.0% | −7.8% | +8.8% |
| Penmark y/y growth | +863% | +107% | +173% | +144% |
| Capital spending (company def.) | ||||
| Penmark estimate | 7,350 | 38,000 | 38,880 | 32,400 |
| Consensus (pre-3Q) | 5,260 | 35,910 | 47,029 | 51,536 |
| % diff vs consensus | +39.7% | +5.8% | −17.3% | −37.1% |
| GAAP EPS ($) | ||||
| Penmark estimate | (0.93) | (4.72) | (4.87) | 2.42 |
| Consensus | n.d. | (5.13) | (4.70) | n.d. |
Source: S&P Global consensus via Yahoo Finance / MarketScreener (revenue and EPS 11–12 Nov 2026; EBIT and capex pre-3Q26; 42 analysts); Penmark Research model. 4Q26E GAAP EPS uses FY26E average basic shares (548m).
Source: CoreWeave filings (A); S&P Global consensus (pre-3Q26, Oct 2026; amber ticks); Penmark Research model (E).
At 8.5x run-rate EBITDA, CoreWeave trades at a 45% discount to cloud peers, reflecting debt and asset lives
| Current price, 18 Nov 26 | CRWV | Cloud med. | Diff. |
|---|---|---|---|
| Valuation (run-rate* & forward) | |||
| EV / run-rate sales* | 4.8x | 5.6x | −13% |
| EV / run-rate EBITDA* | 8.5x | 15.3x | −45% |
| EV / run-rate EBIT* | 85.6x | 20.2x | +323% |
| EV / FY28E Adj. EBIT** | 9.0x | n.d. | n.d. |
| Dividend yield | 0.0% | 0.7% | −0.7pp |
| Growth, margins & capacity | |||
| 3Q26 revenue growth y/y | 168.3% | 18.8% | +149.5pp |
| 3Q26 GAAP EBITDA margin | 57.3% | 52.1% | +5.1pp |
| 3Q26 GAAP EBIT margin | 5.7% | 34.8% | −29.1pp |
| Net debt / run-rate EBITDA | 5.0x | 0.4x | +4.6x |
| Core asset useful life | 6 yrs | Longer | Shorter |
Source: Signals 18 Nov 2026 closes; Company filings; Penmark Research model. Cloud median = Oracle, Microsoft, Amazon. Run-rate = 3Q26 annualised. *Uses ex-prepayment EV ($70.9bn); **uses EV including $11.9bn customer prepayments ($82.8bn).
- Dec 2026Investment-grade DDTL pricing: Planned $4bn+ loan at 7%–7.5% with >90% advance rate lowers debt cost (+$6/sh).
- Late Feb 20274Q26 results & FY27 guidance: First numeric FY27 guide (Street $27bn) and pricing on >130k paid-off GPU renewals (+$22 to +$37/sh).
- Jan–May 2027Texas legislative session: Resolution of governor-elect’s proposed data-center permitting pause (−$3/sh risk if delayed).
- 1H 2027Rival funding & prepayments: Peer financing gaps steer demand to scaled operators (+$20/sh).
Source: CoreWeave 3Q26 call (10 Nov 2026); Bloomberg (Nov 2026); Penmark Research model.
- High discount rate (beta 3.71). Persistent 5%+ Treasury yields and 3.71 equity beta raise WACC by 1.6pp. Likelihood 35%; −$35/sh.
- 2027–2028 supply wave. Hyperscaler and lab capacity cuts fleet pricing by $0.5m/MW. Likelihood 40%; −$37/sh.
- Customer concentration & prepayments. Top two customers are 54% of 3Q26 sales; $1.9bn of prepaid revenue is refundable under cancellable contracts. Customer and prepayment downside cuts base value by $18/sh. Likelihood 20%.
- Upside risk (against Hold): Strong GPU renewals. Paid-off Hopper/Ampere renewals sustaining +$0.5m/MW lift base value to $78. Likelihood 55%; +$37/sh.
Source: Penmark Research model (Sensitivities tab, impacts vs $41.02 base value).
Standard deviation (SD) bands show how far the current multiple sits from its post-IPO average (10.0x): ±1 SD (6.4x–13.5x) covers typical swings; today’s 4.8x reading sits 1.5 SD below mean.
Source: Signals daily closes (to 18 Nov 2026); CoreWeave quarterly filings; Penmark Research model.
Data-center real-estate investment trusts (REITs: Equinix, Digital Realty) earn 26x–27x EBITDA on long-lived buildings; CoreWeave trades below Oracle (10.7x) due to 6-year GPU lives and leverage.
Source: Signals 18 Nov 2026 closes; Company filings (positive-EBITDA peers).
Why we rate CoreWeave Hold: surging backlog meets a $37-per-share pricing tug-of-war and rate risk
Four charts frame our investment call. CoreWeave’s contracted revenue backlog more than doubled over the past year while its share price fell 36%. That disconnect reflects two forces that dominate equity value: renewal pricing after the 2027–2028 supply wave and the cost of funding $71bn of committed FY27E–FY28E capital spending.
Backlog alone does not move the stock. Revenue backlog climbed from $56bn in 3Q25 to $133bn in 3Q26, yet the shares dropped from $137 to $48 by 18 November as investors repriced credit spreads and capital intensity.
Source: CoreWeave quarterly releases and 10-Q; Signals daily closes (to 18 Nov 2026).
Committed spending lowers depreciation drag. Street estimates assume $99bn of combined FY27E–FY28E capex. Aligning builds with committed backlog ($71bn) lifts our FY28E adjusted operating income 9% above consensus.
Source: S&P Global consensus (pre-3Q26, Oct 2026; amber ticks); Penmark Research model.
Two variables can push the rating to Buy or Sell. Moving fleet revenue per average active megawatt by ±$0.5m swings base value by ±$37 per share; raising equity beta from 2.67 to 3.71 cuts base value to $6.
Source: Penmark Research model (base DCF value $41.02 at 18 Nov 2026).
Why our DCF sits below peer multiples. Valuing FY28E adjusted operating income at 14.0x yields $54, but omits the 6-year refresh of the 2026 GPU cohort and return fade, which subtract $13 per share in our DCF.
Source: Penmark Research model.
Four named debates decide whether CoreWeave compounds to $98 or erodes toward zero
Our Hold rating reflects an unresolved clash between near-term funding relief and post-2027 pricing power. The table below sets out the four debates the stock trades on, where our model diverges from consensus or market-implied expectations, and the dated events that will settle each debate.
| Debate | Market view | Penmark view | Evidence | What changes our view | Resolves |
|---|---|---|---|---|---|
| Post-2027 Renewal Pricing | At $48, shares sit slightly above our base, implying about $14.4m/MW by FY30E (26% bull weight); bears fear a supply wave near $11–12m/MW. | Base lifts revenue/MW to $14.3m by FY30E then fades to $13.5m; our bear case ($11–12m/MW) captures supply-wave risk. | Ampere and Hopper renewals rose 30%–50% q/q in 3Q26; new 4–5y deals signed at $20–25m/MW; ~$4bn 2027 renewal annual recurring revenue (ARR). | Paid-off 2027 renewals hold +$0.5m/MW (+$37/sh); or supply cuts fleet below $12m/MW (−$37/sh). | Late Feb 2027 (4Q26 call & 2027 renewals) |
| Cost of Capital & Funding Access | At $48, shares imply a 9.9% WACC (below our 10.2%); unsecured senior notes were valued at 87.8% of par at 30 September. | Secured DDTLs (70%–100% advance) and $9bn near-term prepayments cover FY27E without unsecured bonds. | 7.9% avg. debt cost at 3Q26 (−250bp y/y); >$4bn paid down; planned $4bn+ IG DDTL at 7%–7.5%. | IG DDTL closes at ≤7.5% (lifts base to $47) or WACC drops 1pt ($77, Buy); or WACC rises 1pt ($25, Sell). | Dec 2026 (IG DDTL) & Federal Open Market Committee (FOMC) 9 Dec 2026 |
| Committed Capex Discipline | Consensus models $47bn FY27E and $52bn FY28E capex, inflating debt and depreciation. | We model spending aligned to committed backlog ($39bn FY27E, $32bn FY28E), beating FY28E EBIT by 9%. | CFO noted that if CoreWeave focuses only on committed backlog, 2027 capex would match 2026 ($37–39bn), with $0–$5bn parent capital. | FY27 capex guide exceeds $45bn without matching prepayments; or refresh costs exceed 85% of original capex. | Late Feb 2027 (FY27 capex guide) |
| Customer Concentration & Power Permitting | Microsoft capex slowdown, $1.9bn of refundable customer liabilities under cancellable contracts and a Texas pause threaten 2027 delivery. | Top-two concentration (54%) is real, but permitted sites are active and backlog spans 52 data centers. | Customer A fell from 67% of 3Q25 sales to 24% in 3Q26; Customer B is 30% (top two 54%); 98% of sales committed; 4.6GW contracted. | Texas pause cuts FY27 additions from 1.08GW to 0.85GW (−$3/sh); or customer and prepayment downside (prepayments −$6bn, −$0.25m/MW) cuts $18/sh. | Jan–May 2027 (Texas legislature) |
Source: CoreWeave 3Q26 10-Q and call (10 Nov 2026); S&P Global consensus (revenue and EPS 11–12 Nov 2026; EBIT and capex pre-3Q26); Penmark Research model.
Source: CoreWeave 3Q26 earnings call; Bloomberg (Nov 2026); Penmark Research estimates.
From a 2017 crypto miner to a $133bn AI backlog, funded by $48bn of debt
CoreWeave began as a New Jersey crypto-mining shop and became the largest listed pure-play artificial intelligence cloud. It now serves the biggest software and AI labs on earth, yet its stock turns on how it finances six-year silicon.
On 10 November 2026, CoreWeave posted $3.7bn of quarterly revenue, adding over $1bn in ninety days, and projected $9bn of customer prepayments in the next three months. Across 52 data centers running 1.7 gigawatts of live electricity, hundreds of thousands of liquid-cooled NVIDIA graphics chips trained frontier models and answered live inference queries for major cloud and AI labs. Three years earlier, FY23 annual revenue was $229m.
What CoreWeave does is simple in outline and unforgiving in practice. General-purpose clouds such as Amazon Web Services and Microsoft Azure were designed for millions of small web servers and databases. Modern AI requires tens of thousands of identical GPUs linked by high-speed InfiniBand networking running at peak utilization without silent failures. CoreWeave leases powered data-center shells, installs bare-metal GPU clusters, and wraps them in proprietary orchestration software (Mission Control and Slurm on Kubernetes) plus Weights & Biases developer tools.
Cash arrives before servers switch on. Ninety-eight percent of third-quarter sales came from committed multi-year contracts, typically four to five years long, with upfront prepayments on 70% of 2026 deals. CoreWeave pairs those contracts with secured DDTLs covering 70%–100% of server hardware. Each FY26E revenue dollar earns 61 cents of adjusted EBITDA, but 51 cents goes to six-year depreciation and 24 cents to interest, keeping GAAP profit negative until FY28E.
The company turned three times in seven years. Natural-gas traders Michael Intrator, Brian Venturo and Brannin McBee founded Atlantic Crypto in September 2017 and pivoted into cloud rendering in 2019 after crypto crashed. In 2022, CoreWeave moved early into NVIDIA data-center GPUs as generative AI took off, then began raising GPU-backed debt in 2023. Customer A reached 67% of third-quarter 2025 sales before falling to 24% in third-quarter 2026 as Customer B grew to 30%, while a March 2025 Nasdaq listing at $40 and major AI-lab deals lifted backlog to $133bn.
The third turn came in October 2026, when credit markets balked at $48bn of debt principal and $38bn of annual capital spending. CoreWeave’s senior notes were valued at 87.8% of par at 30 September, and shares sank 38% in October as credit spreads widened. Management paused unsecured borrowing, leaned on prepayments, and prepared an investment-grade secured loan backed by blue-chip contracts.
CoreWeave’s loop works because speed to market, high uptime and early NVIDIA access make it an essential capacity partner. Once a four-year contract pays off its loan, the remaining two years of GPU life throw off high-margin cash. What could break the loop is equally clear: if the 2027–2028 supply wave caps renewal pricing, six-year refreshes earn below the 10.2% cost of capital, framing our Post-2027 Renewal Pricing and Cost of Capital & Funding Access debates.
Hyperscaler or AI lab commits 4–5 years of GPU capacity at $20–25m/MW on new deals.
70% of 2026 deals include prepayments ($9bn in 90 days); DDTLs fund 70%–100% of capex.
CoreWeave brings shells live ($36m/MW capex), earning 61%–69% EBITDA margin across 52 sites.
After loan payoff (>130k GPUs in 2027), re-contracted servers throw off cash before 6-year refresh.
Why the story matters for the stock: CoreWeave has proved it can deploy gigawatts faster than any standalone rival. Re-rating now hinges on proof that paid-off GPUs renew at premiums and secured funding lowers equity beta.
High EBITDA margins mask depreciation equal to 51% of revenue and capital intensity
CoreWeave operates a capacity-led infrastructure model where multi-year committed contracts pay for six-year GPU clusters. Unlike general-purpose cloud platforms that sell diversified software services, CoreWeave monetizes powered megawatts dedicated to AI training and inference, earning high cash operating margins before heavy depreciation and interest.
Source: CoreWeave Form 10-K (FY25), Form 10-Q (3Q26) and Form S-1 filings. Highlighted nodes mark key strategic turning points.
Source: Penmark Research model (FY26E; committed share applies 98% 3Q26 10-Q split; SBC includes $2m other operating items).
Source: CoreWeave 3Q26 Form 10-Q and earnings release; Penmark Research model.
What CoreWeave sells and how customers pay: CoreWeave leases dedicated bare-metal GPU compute clusters, high-speed networking, AI storage and orchestration software (Mission Control, Slurm on Kubernetes and Weights & Biases). Committed multi-year contracts generated 98% of 3Q26 revenue, typically spanning four to five years with 70% of 2026 deals carrying upfront customer prepayments.
Fixed versus variable cost structure: Power, colocation leases and facility operations sit in cash operating costs ($3.7m per average active MW in FY26E, rising to $4.2m by FY28E). The dominant economic cost is fixed hardware depreciation ($6.8bn or 51% of FY26E revenue) on GPUs amortized over six years, creating steep operating leverage as clusters fill.
Revenue equals active power times pricing per MW; six-year GPU refresh caps cash flow
CoreWeave’s economics boil down to four unit numbers: average active megawatts, revenue per megawatt ($9.7m in FY26E rising to $14.3m by FY30E), cash cost per megawatt ($3.7m–$4.4m), and capital cost per new megawatt ($36m). Because GPUs require replacement every six years at an estimated 85% of initial cost, capital intensity resurfaces sharply in FY32E.
Source: Penmark Research model (FY27E base case). Highlighted boxes mark confirmed thesis drivers.
Source: Penmark Research model (base case: $36m/MW new build, 85% end-of-life refresh at 6 years).
Source: CoreWeave 3Q26 Form 10-Q.
Customer concentration is easing but remains high. Customer A fell from 67% of 3Q25 sales to 24% in 3Q26, while Customer B reached 30%, leaving the top two at 54% of 3Q26 sales. Separately, $1.9bn of prepaid revenue sits in refundable customer liabilities under cancellable contracts.
Vendor and geographic dependence: NVIDIA supplies nearly all compute silicon and owns 8.5% of basic shares (47.2m Class A shares). Operations are heavily U.S.-weighted (1,967 of 2,189 employees at year-end 2025), exposing delivery pace to regional grid queues such as Texas.
Power scarcity supports near-term pricing, while a 2027–2028 supply wave looms
Specialized AI compute is expanding rapidly, yet profit pools remain concentrated upstream in silicon and downstream at scale hyperscalers. Near-term demand outstrips grid-connected power, allowing CoreWeave to sign four-to-five-year contracts at $20–$25m/MW. The key market question is whether industry capacity additions in 2027–2028 compress pricing once supply catches up.
Source: Company filings; Penmark Research analysis. REIT EBITDA margins from 3Q26 Equinix (44.5%) and Digital Realty (44.2%) filings.
Source: CoreWeave releases (FY25A); Penmark Research model (FY26E–FY30E). Contracted power reached 4.6GW in 3Q26.
| Market scope | Publisher | Start | Forecast |
|---|---|---|---|
| GPU as a Service | MarketsandMarkets (Nov 2026) | $37.4bn ’26 | $224.1bn ’32 |
| AI infrastructure | Grand View Research (2025) | $33.5bn ’25 | $120.7bn ’33 |
| Cloud computing (all deployment models) | MarketsandMarkets (2025) | $1.29tn ’25 | $2.28tn ’30 |
Source: MarketsandMarkets (Nov 2026, GPUaaS; 2025, Cloud Computing); Grand View Research (2025, AI Infrastructure). Definitions differ in scope, so the estimates are not comparable and market-share ratios are not meaningful.
Bottom-up demand and cycle position: Training frontier models and running agentic inference require megawatt-scale liquid-cooled clusters. Grid interconnection delays keep 2026 supply tight, lifting CoreWeave’s 3Q26 Ampere and Hopper renewals by 30%–50% quarter over quarter. However, hyperscalers and frontier labs are also leasing power shells directly, creating a 2027–2028 supply wave that could pressure renewal economics.
Value-chain bargaining power: NVIDIA captures the richest margin pool in AI hardware and controls next-generation Blackwell and Rubin allocations. Upstream power landlords (Equinix, Digital Realty) earn ~44% EBITDA margins on long-lived buildings. Specialized AI clouds carry heavy capital risk between both layers.
Regulatory and substitution threats: State-level power moratoria (including Texas governor-elect Hinojosa’s pledge to pause new data-center approvals) and custom hyperscaler silicon (Google TPUs, Amazon Trainium) threaten both delivery timing and long-run NVIDIA cluster share.
CoreWeave leads neoclouds on scale and software, but trails hyperscalers on funding cost
CoreWeave wins business on deployment speed, bare-metal GPU uptime and early NVIDIA access, not balance-sheet cost. Among pure-play AI clouds, CoreWeave has pulled sharply ahead on live power (1.7GW) and contracted backlog ($133bn). Yet its largest competitors (Microsoft, Amazon and Oracle) enjoy investment-grade balance sheets and far lower funding costs.
Source: Penmark Research assessment (Rivals = Nebius, IREN, Applied Digital).
Source: Signals (18 Nov 2026); 3Q26 filings (bubble size reflects EV; APLD +432%/−6% off-axis).
| Peer group | Key players | Scale & 3Q26 profitability | Core advantage vs CoreWeave | Vulnerability |
|---|---|---|---|---|
| Hyperscalers | Microsoft, Amazon | Company-level $369bn–$808bn ann. rev; 24%–63% EBITDA; low debt | AA/AAA funding costs, massive enterprise base, custom silicon (Trainium) | Multi-tenant overhead; slower single-cluster GPU rollouts; rely on CRWV overflow |
| Oracle Cloud (OCI) | Oracle | $77.4bn ann. rev (+30% y/y); 52% EBITDA; 16.0x EV/EBIT | Investment-grade balance sheet, large remaining performance obligations (RPO), bare-metal GPUs | Also raising debt for AI capex ($88.3bn net debt); narrower AI developer tools |
| Data-center REITs | Equinix, Digital Realty | $7.6bn–$10.3bn ann. rev; 44% EBITDA; 26x–27x EV/EBITDA | Long-lived buildings, diversified tenants, no GPU obsolescence risk | No GPU fleet or AI orchestration software; supply shells rather than compute |
| Smaller neoclouds | IREN, Applied Digital, Nebius | IREN ($0.9bn) & APLD ($1.4bn) posted negative 3Q26 EBITDA | Lower legacy debt at select sites; opportunistic power contracting | Funding gaps (IREN ~$6bn, APLD 800MW unfunded) and build delays |
Source: Signals (18 Nov 2026); Company 3Q26 filings; Penmark Research analysis. Hyperscaler figures are consolidated company totals.
Durability of CoreWeave’s edge: Mission Control automation, Slurm on Kubernetes and Weights & Biases reduce GPU downtime across massive clusters, giving frontier labs a real operational edge. Early access to NVIDIA hardware also gives CoreWeave a deployment advantage over smaller neoclouds.
Where the moat is thin: Switching costs are low once a multi-year training contract expires, because workloads are portable across standard NVIDIA clusters. Long-term pricing power therefore depends less on lock-in and more on whether rivals’ funding bottlenecks keep industry supply scarce.
Insiders hold 72% of voting power via 10:1 Class B stock; most capital funds buildout
| Who | Tenure & prior | FY25 pay & mix (DEF 14A) | Ownership at $47.83 & Form 4 |
|---|---|---|---|
Michael Intrator CEO, President & Chair (Co-founder) |
9.2 yrs in role (Sep 2017); 9.2 yrs at firm. Ex-CEO Hudson Ridge AM; Natsource AM. | $3.45m SCT total ($20.0m IPO RSU grant*)Discretionary cash + 4-yr time/IPO RSUs. | $2,942m (61.5m sh, 11.1% sh, 38.7% votes**); >3,000x sal. Compliant with 5x CEO guideline. |
Nitin Agrawal CFO & Treasurer |
2.7 yrs in role (Mar 2024); ex-VP Finance Google Cloud, CFO Mapbox, AWS Finance Dir. | $1.01m SCT total ($7.4m IPO RSU grant*)Salary, bonus + 4-yr RSUs; no formal EPS weight. | $24.2m (0.51m Class A, 0.1% sh); >40x sal. Sold 22.6k sh ($1.8m) via 10b5-1 in Oct 2026. |
Brian Venturo Chief Strategy Officer (Co-founder) |
2.7 yrs CSO (Mar 2024; CTO 2017–24); 9.1 yrs at firm. Ex-Hudson Ridge; Natsource. | $2.80m SCT total ($11.2m IPO RSU grant*)Cash bonus + 4-yr time/IPO RSU retention grants. | $1,461m (30.5m sh, 5.5% sh, 20.3% votes**); >2,000x sal. Sold 65.6k sh ($5.8m) in Oct 2026. |
Brannin McBee Chief Dev. Officer (Co-founder) |
2.7 yrs CDO (Mar 2024; CSO 2017–24); 9.2 yrs at firm. Leads power & site buildout. | $1.98m SCT total ($7.6m IPO RSU grant*)$591k sal, $600k bonus, $786k other + RSUs. | $1,029m (21.5m sh, 3.9% sh, 14.6% votes**); >1,700x sal. Sold 56.5k sh ($5.0m) in Oct 2026. |
Jonathan Jones Chief Revenue Officer (New Oct 25) |
1.1 yrs in role (hired Oct 2025). Enterprise sales veteran. | $18.58m SCT total ($18.1m new-hire RSUs)$250k sign-on cash + $18.1m 4-yr RSU grant. | n.d. in Apr 26 proxy (unvested RSUs). Sold 37.4k sh ($2.9m) in Oct 2026 (incl. 10b5-1). |
Source: CoreWeave DEF 14A (record date 15 Apr 2026; filed 22 Apr 2026); Form 4 via Signals. *Summary Compensation Table (SCT) omits 2025 IPO liquidity-conditioned restricted stock units (RSUs) under ASC 718. **Individual vote shares sum above the 72.3% group total because SEC rules add each person’s 60-day exercisable Class B options only to their own denominator.
Source: CoreWeave DEF 14A (record date 15 Apr 2026; filed 22 Apr 2026; all 12 executives & directors).
Governance & pay alignment: Class B shares (10 votes vs 1 for Class A) give all insiders 72% voting control, though all board committees are 100% independent. FY25 bonuses were discretionary rather than tied to return on invested capital (ROIC) or FCF/sh, while 15 Form 4 sales ($17m total; zero buys) occurred between 30 Sep and 10 Nov 2026.
Capital allocation record: Most post-IPO capital funds GPU clusters ($15bn FY25A; $31bn 9M26A), alongside targeted software acquisitions (Weights & Biases, OpenPipe, Marimo) and a modeled 35m-share ATM ($1.7bn in FY27E). Large founder stakes ($5.4bn combined) align against dilutive equity raises and support contract-backed spending.
Source: Penmark Research model (FY27E–FY28E cumulative cash sources and uses; full-width flow proportional to dollars).
CoreWeave trades on funding access and pricing power, not near-term GAAP earnings
Empirical testing confirms three company-specific drivers explain CoreWeave’s extreme volatility. Across four tests (event-day stock reactions, trailing 12-month return decomposition, 86-week factor regressions and 35 analyst questions across four calls), funding access, renewal pricing and power delivery each pass at least two tests, while headline GAAP EPS does not.
| Confirmed driver | What moves the stock | Tests confirmed | Market move vs model value | Debate link | Next number to watch |
|---|---|---|---|---|---|
| Financing Access & Cost of Capital | Secured DDTL spreads, bond prices, lender headlines and beta/WACC shifts. | Confirmed by 3 of 4 tests (events, return decomp., factor variance). | −38% Oct credit shock vs −$35/sh (beta 3.71) & +$6/sh (IG DDTL). | Cost of Capital & Funding Access | Dec 2026: ≥$4bn IG DDTL priced at 7.0%–7.5% (>90% advance). |
| Pricing Power & Renewal Economics | Fleet revenue per MW, paid-off GPU renewal rates and hyperscaler capex signals. | Confirmed by 3 of 4 tests (events, Q&A share 26%, return decomp.). | −9.7% on MSFT capex miss vs ±$37/sh per ±$0.5m/MW shift. | Post-2027 Renewal Pricing | Late Feb 2027: 2027 renewal pricing on >130k paid-off GPUs. |
| Power Delivery & Capex Pace | Active GW additions, FY27 capex discipline and state permitting headlines. | Confirmed by 2 of 4 tests (events, Q&A share 31%, top topic). | −9.9% on 4 Nov Texas news vs −$3/sh model hit for 0.23GW delay. | Committed Capex & Permitting | Late Feb 2027: FY27 capex outlook (Penmark $39bn vs $47bn Street). |
Source: Signals daily closes and transcripts (to 18 Nov 2026); Yahoo Finance (18 Nov 2026); Penmark Research model.
Test 1 (Event-day reactions): Six of seven post-IPO earnings days produced double-digit swings (only 1Q25 moved less, at −2.5%), culminating in a +24.7% jump on 11 November 2026 when management shelved unsecured debt and detailed $9bn of prepayments. Non-earnings drops in October–November (−10.1% lender selloff, −9.7% Microsoft capex, −9.9% on 4 November Texas news before a −9.8% sector AI drop on 5 November) confirm credit and permitting sensitivity.
Source: Signals; Yahoo Finance daily closes (May 2025–Nov 2026).
Test 2 (Return decomposition): Over the twelve months to 18 November 2026, CoreWeave fell 36% from $75 to $48 despite growing annualized quarterly sales 168% (worth +$126/sh at a constant multiple). Valuation compression from ~13x to 4.8x EV/run-rate sales (−$108/sh) and $27bn of incremental net debt and share dilution (−$45/sh) drove the entire loss.
Source: Signals (18 Nov 2026); CoreWeave filings; Penmark Research decomposition.
73% of weekly moves are company-specific; funding and pricing dominate value
Source: Yahoo Finance weekly returns (n=86 weeks, Apr 2025–Nov 2026; OLS R²=26.6%).
Source: Signals transcripts (35 analyst questions across 4Q25–3Q26 calls through 10 Nov 2026).
What the market does: Credit headlines moved shares −38% in October 2026 and +24.7% on 11 November when management ruled out unsecured bonds. Our base model uses a 2.67 regression equity beta (with a 3.71 high-beta stress test), making discount-rate shifts dwarf operating beats.
Model value vs market reaction: Securing a $4bn+ investment-grade DDTL at 7%–7.5% lifts secured debt share to 80%, adding +$6/sh (+13%). Raising equity beta from 2.67 to 3.71 cuts base value by −$35/sh (−74% of share price). Next number: December 2026 DDTL pricing (Penmark: ≤7.5%, >90% advance; likely move ±10%–15%).
What the market does: Pricing and renewals drew 26% of analyst questions across the last four calls (with management confident on 3Q26 Ampere/Hopper uplifts of 30%–50% q/q), while Microsoft’s 28 October capex miss triggered a −9.7% selloff on oversupply fears.
Model value vs market reaction: Each ±$0.5m/MW shift in fleet revenue moves equity value by ±$37/sh (±77% of share price), nearly 8x the single-day reaction to Microsoft. Next number: Late Feb 2027 renewal rates on >130,000 paid-off GPUs (Penmark base: FY27E $11.1m/MW; likely move ±15%–25%).
What the market does: Power delivery and capex ranked first in call Q&A (31% of questions). News of Texas governor-elect Hinojosa’s proposed data-center pause drove a 9.9% drop on 4 November, followed by a 9.8% sector-wide AI selloff on 5 November after the midterms.
Model value vs market reaction: Slowing FY27E net power additions from 1.08GW to 0.85GW trims our base DCF by only −$3/sh (−6%), because permitted Texas sites remain active. Next number: Late Feb 2027 FY27 capex outlook (Penmark $39bn vs Street $47bn).
| Dropped candidate | Why commonly cited | Test result & reason dropped |
|---|---|---|
| Weekly 10-year Treasury moves | High leverage implies direct rate sensitivity | Added <0.1% incremental R² in 86-week regression; credit spreads, not weekly yields, drive CRWV. |
| Software & M&A mix (W&B) | Platform moat narrative | Only 5.7% of call Q&A; software is bundled with compute contracts and moves zero near-term EPS. |
| Headline quarterly GAAP EPS | Standard equity driver | Negative through FY27E; 3Q26 net loss of $709m still triggered a +24.7% rally on funding clarity. |
Source: Yahoo Finance; Signals transcripts (n=4 calls, 86 weeks since Mar 2025 IPO). Limit: short 20-month trading history constrains long-cycle factor tests.
Contracted backlog doubles revenue, but post-2027 renewal pricing decides equity value
The claim: CoreWeave’s $133bn backlog anchors rapid revenue scaling through FY27E, but equity value rests on whether paid-off GPU renewals hold pricing power through the 2027–2028 supply wave. This chapter rests on our Pricing Power & Renewal Economics stock driver.
At $48, shares sit slightly above our base DCF. Reverse DCF shows the price implies about $14.4m/MW by FY30E or a 26% bull weight, while consensus models $27bn of FY27E sales and $43bn in FY28E. Shares trade at 4.8x run-rate EV/sales (on $70.9bn ex-prepayment EV) and 9.0x FY28E adjusted EBIT (on $82.8bn EV including prepayments); our bear case ($11–$12m/MW) captures supply-wave risk.
Source: Penmark Research model (FY26E–FY38E revenue per average active MW).
Source: CoreWeave quarterly releases and 3Q26 Form 10-Q.
1. Contracted backlog and revenue visibility (3Q26 Form 10-Q): Revenue backlog grew 138% year over year to $133bn at 30 September 2026 (including remaining performance obligations and committed contracts). Committed multi-year contracts generated 98% of 3Q26 sales across 52 active data centers.
2. Paid-off GPU renewal uplifts (3Q26 earnings call): Older Ampere and Hopper clusters re-contracted at 30%–50% quarter-over-quarter premiums in 3Q26, generating over $3bn of prepayments within the $9bn near-term inflow. Over 130,000 paid-off GPUs roll off initial contracts in 2027, offering about $4bn of annualised revenue with zero new debt.
3. Counter-evidence from supply-wave and customer filings: Microsoft’s 28 October capex miss, direct shell leasing by frontier AI labs, and $1.9bn of prepaid revenue sitting in refundable customer liabilities under cancellable contracts show renewal pricing is not guaranteed once 2027–2028 rival gigawatts arrive.
Every $0.5m/MW shift moves equity value by $37 per share. Our base case lifts fleet revenue/MW from $9.7m in FY26E to $14.3m by FY30E then fades it to $13.5m ($41 base DCF). Peaking at $15.0m/MW before fading to $13.7m in our bull case lifts value to $98 (+105%); stalling near $11.9m/MW in our bear case makes 6-year refreshes value-destructive, driving equity value to nil.
| Date | Signpost | Confirms upside | Signals downside |
|---|---|---|---|
| Late Feb 2027 | 4Q26 call & first 2027 GPU renewal disclosures | >130k GPU renewals hold premiums tracking +$0.3m/MW (+$22/sh); FY27 rev >$26.9bn | Renewal discounts appear or FY27 guide trails $26.5bn consensus |
| 1H 2027 | Hyperscaler capex & neocloud delivery pace | Rival funding delays keep AI compute scarce through FY28 | Hyperscalers cut third-party GPU leasing; cancellable liabilities grow |
Source: CoreWeave 3Q26 call; Penmark Research estimates.
Secured DDTLs and prepayments cover 2027 capex, leaving WACC as the rating swing
The claim: CoreWeave can fund $39bn of FY27E capital spending without costly unsecured bonds, easing near-term liquidity panic, yet its 10.2% WACC and high equity beta (2.67 base; 3.71 stress) leave valuation fragile to discount-rate shifts. This chapter rests on our Financing Access & Cost of Capital stock driver.
Bond prices reflect real balance-sheet stress. CoreWeave’s unsecured senior notes were valued at 87.8% of par at 30 September, implying roughly 45%–50% risk-neutral five-year default odds (about 25% real-world, anchoring our 25% bear weight). Investors worry that $48bn of existing debt principal and $38bn of annual capex could tighten parent-level funding access.
Source: CoreWeave 3Q26 10-Q Note 10; Penmark Research model (excl. $2.4bn 4Q26E).
Source: Penmark Research model (probability-weighted 12-month price targets).
1. Self-amortising secured debt (3Q26 10-Q & call): Weighted debt cost fell 250bp y/y to 7.9% at 3Q26, with over $4bn repaid from contract cash flows. Management stated on the 3Q26 call that over 75% of committed capex through 2027 backs investment-grade and enterprise customers, supporting 70%–100% DDTL advances at Secured Overnight Financing Rate (SOFR) + 225–550bp.
2. Prepayments and planned IG facility (3Q26 call & Bloomberg, 11 Nov): CoreWeave expects about $9bn of prepayments within ninety days and plans a $4bn+ investment-grade DDTL at 7.0%–7.5% in December 2026. Paired with $18bn of FY27E operating cash flow and $1.7bn of modeled ATM equity, these cover FY27E capex and $11bn of repayments without unsecured bonds.
3. Counter-evidence from rates and bond pricing: With 10-year U.S. Treasury (UST) yields at 5.10% (2026 peak 5.31%), a hot October consumer price index (CPI) print (released 11 November) and a 3.71 high-beta stress test, wider credit spreads push cost of capital sharply higher.
Discount-rate moves override operating beats. Closing the investment-grade DDTL (80% secured debt share) trims WACC by 0.2pp and adds +$6/sh. A 1-point WACC move either way shifts our target from $77 (Buy) to $25 (Sell), while beta at 3.71 cuts it to $15 (Sell). Issuing $5bn of parent equity at $40 cuts base value by under $1/sh ($0.42), showing capital access and debt cost drive value far more than dilution.
| Date | Signpost | Confirms upside | Signals downside |
|---|---|---|---|
| 9 Dec 2026 | Federal Reserve FOMC rate decision | 10y UST falls below 4.75%; AI credit spreads tighten | Hawkish rate path lifts 10y UST >5.3% and senior notes <85% par |
| Dec 2026 | Investment-grade DDTL syndicate closing | ≥$4bn facility closes at 7.0%–7.5% with >90% advance rate | Facility undersized, delayed to 2027, or priced >8.5% |
Source: CoreWeave 3Q26 call (10 Nov 2026); Bloomberg (Nov 2026); Penmark Research model.
Committed capex lifts FY28E EBIT 9% above Street, but six-year GPU refresh caps value
The claim: Aligning capital spending with committed backlog saves $27bn versus consensus across FY27E–FY28E and beats FY28E adjusted operating income by 9%, yet mandatory six-year hardware refreshes prevent CoreWeave from earning a landlord multiple. This chapter rests on our Power Delivery & Capex Pace stock driver.
Consensus models far higher two-year capex while Street targets sit well above our DCF. S&P Global consensus assumes $47bn of FY27E capex and $52bn in FY28E, holding FY28E adjusted EBIT at $8.4bn. Street price targets average $123 (ranging from $45 to $278), far above our DCF, which charges for the FY32E–FY33E refresh.
Source: Penmark Research model (base case FCF excl. customer prepayments).
| $m (base vs Street) | FY26E | FY27E | FY28E |
|---|---|---|---|
| Penmark capex | 38,000 | 38,880 | 32,400 |
| Consensus capex | 35,910 | 47,029 | 51,536 |
| Capex diff. vs Street | +2,090 | −8,149 | −19,136 |
| Penmark Adj. EBIT | 1,375 | 3,754 | 9,169 |
| Consensus Adj. EBIT | 1,066 | 4,070 | 8,429 |
| EBIT gap vs Street | +29.0% | −7.8% | +8.8% |
Source: S&P Global consensus (pre-3Q26, Oct 2026); Penmark Research model.
Why our FY27E EBIT sits 8% below consensus before beating in FY28E: Front-loaded D&A ($13bn) on FY26–FY27 builds hits FY27E earnings before new megawatts reach full utilization; by FY28E, lower cumulative capex yields lower D&A and a 21.5% adjusted operating margin.
1. Conditional capex commentary (3Q26 call): CFO Nitin Agrawal noted that if CoreWeave focuses only on committed backlog, 2027 capex would stay in line with 2026 ($37–$39bn). That remark is conditional rather than a formal guide, and management gave no FY28 figure.
2. Unit build math (Model capacity build): Adding 1.08GW of net active power in FY27E and 0.90GW in FY28E at $36m/MW implies $39bn and $32bn of capex, $27bn below consensus across two years.
3. Six-year depreciation policy (FY25 Form 10-K): CoreWeave depreciates GPUs over six years (versus long-lived buildings at Equinix and Digital Realty). Refreshing FY26–FY27 cohorts at 85% of cost adds $32bn in FY32E and $33bn in FY33E, briefly dipping FCF ex-prepayments to −$1.2bn in FY32E.
Disciplined capex helps, but refresh explains our DCF discount. Our FY28E adjusted EBIT sits 9% above Street ($9.2bn vs $8.4bn), yet charging for the 2026-cohort refresh and ROIC fade to 12.5% cuts our DCF-implied multiple to 13.0x FY28E EBIT ($41) versus 14.0x ($54) in our single-multiple cross-check, a −$13/sh reality check.
| Date | Signpost | Confirms upside | Signals downside |
|---|---|---|---|
| Late Feb 2027 | FY27 numeric capex & GW outlook | FY27 capex lands near $37–$39bn with ≥1.08GW active additions | Capex tops $45bn without matching prepayments, or <0.85GW added |
| FY27–FY28 | Useful-life & GPU refresh disclosures | Paid-off Ampere/Hopper GPUs run >6 years profitably | NVIDIA Rubin cadence forces earlier 4–5 year write-downs |
Source: CoreWeave filings; Penmark Research model.
Reported operating cash flow flatters cash generation via about $12bn of customer prepayments
CoreWeave’s short public track record shows rapid revenue scaling paired with wide gaps between non-GAAP profit, GAAP earnings and true free cash flow. Revenue grew 168% in FY25A and 168% year over year in 3Q26, driven primarily by active megawatt additions alongside rising fleet pricing rather than acquisitions. Investors must still adjust for stock-based compensation, six-year depreciation and upfront customer prepayments.
Source: CoreWeave filings (FY24A–FY25A); Penmark Research model (FY26E–FY27E).
| $m | FY24A | FY25A | FY26E | FY27E |
|---|---|---|---|---|
| Revenue | 1,915 | 5,131 | 13,214 | 26,862 |
| YoY growth | 736.2% | 167.9% | 157.5% | 103.3% |
| GAAP operating income | 324 | (46) | 684 | 2,814 |
| (+) SBC, intangibles & other | 32 | 712 | 691 | 940 |
| Adjusted operating income | 356 | 666 | 1,375 | 3,754 |
| Adj. minus GAAP gap | 32 | 712 | 691 | 940 |
| GAAP net income | (863) | (1,167) | (2,585) | (2,784) |
| Operating cash flow (CFO) | 2,749 | 3,058 | 12,647 | 17,924 |
| Less: net prepayments | (2,049) | (4,174) | (8,892) | (7,675) |
| FCF ex-prepayments | n.d. | (15,961) | (34,245) | (28,630) |
Source: CoreWeave Form 10-K, 3Q26 10-Q; Penmark Research model.
| Accounting area | Reported treatment | Analytical reality & Penmark adjustment |
|---|---|---|
| Customer prepayments | Booked in operating cash flow ($4.2bn FY25A; $8.9bn FY26E) as cash arrives, ahead of revenue delivery. | Prepayments are customer financing, not recurring free cash flow. We strip net prepayments out of FCF and deduct the $11.9bn balance ($10.0bn deferred + $1.9bn refundable customer liabilities) in our equity bridge. |
| 6-year GPU useful life | Technology equipment is depreciated straight-line over 6 years (51.3% of FY26E revenue). | Matches hyperscaler policy, but if NVIDIA’s annual chip cadence shortens economic life to 4–5 years, D&A rises sharply and equity value compresses. We model an explicit 85% refresh in Year 6. |
| Stock-based comp. | Excluded from Adjusted EBITDA and Adjusted EBIT ($630m FY25A; $643m FY26E; $940m FY27E). | Real economic dilution for shareholders. We deduct SBC in GAAP EBIT and include 57m diluted instruments (611m diluted shares before FY27E ATM) in per-share equity value. |
| Company capex vs cash | Company capex ($14.5bn in 3Q26) is PP&E additions including finance leases, less change in construction in progress; 3Q26 cash capex ($9.4bn) was 35% below company capex. | Our model uses company-definition capex ($38bn FY26E; $39bn FY27E), treats OEM-financed equipment as capex and OEM loans as debt, and holds construction in progress ($9.3bn) constant. |
Source: CoreWeave Form 10-K (FY25), Form 10-Q (3Q26); Penmark Research analysis.
Revenue reaches $43bn by FY28E as active power hits about 3.9GW and EBIT margins double
Our five-year operating forecast models capacity-driven growth through FY30E, anchored inside every FY26 guidance range. Revenue rises from $5.1bn in FY25A to $13bn in FY26E, $27bn in FY27E (+103%) and $43bn in FY28E (+59%), before moderating as active power additions step down from 1.08GW/year to 0.65GW/year by FY30E.
| Capacity & unit driver | FY25A | FY26E | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|---|
| Active power, year-end (GW) | 0.85 | 1.88 | 2.96 | 3.86 | 4.61 | 5.26 |
| Net active additions (GW) | n.d. | 1.03 | 1.08 | 0.90 | 0.75 | 0.65 |
| Average active power (GW) | n.d. | 1.37 | 2.42 | 3.41 | 4.24 | 4.94 |
| Revenue / avg active MW ($m) | n.d. | 9.68 | 11.10 | 12.50 | 13.80 | 14.30 |
| Cash op. cost / avg MW ($m) | n.d. | 3.74 | 4.10 | 4.20 | 4.31 | 4.42 |
| Capex / new MW ($m) | n.d. | 36.89 | 36.00 | 36.00 | 36.00 | 36.00 |
| Refresh capex ($m) | n.a. | n.a. | 0 | 0 | 2,502 | 7,397 |
Source: CoreWeave releases (FY25A); Penmark Research model (Capacity tab, FY26E–FY30E base case).
Source: CoreWeave filings (FY25A); Penmark Research model (FY26E–FY32E).
| Line item ($m) | FY26 Guide | FY26E | FY27E |
|---|---|---|---|
| Revenue (Penmark) | 13,000–13,300 | 13,214 | 26,862 |
| vs Consensus (%) | n.a. | +2.4% | +1.4% |
| Adj. EBITDA (Penmark) | n.g. | 8,108 | 16,940 |
| vs Consensus (%) | n.a. | +7.1% | +1.8% |
| Adj. EBIT (Penmark) | 1,275–1,375 | 1,375 | 3,754 |
| vs Consensus (%) | n.a. | +29.0% | −7.8% |
| Capex (Penmark) | 37,000–39,000 | 38,000 | 38,880 |
| vs Consensus (%) | n.a. | +5.8% | −17.3% |
| Year-end power (GW) | >1.85 | 1.88 | 2.96 |
| Rev / avg MW ($m) | n.g. | 9.68 | 11.10 |
| GAAP EPS ($) | n.g. | (4.72) | (4.87) |
Source: CoreWeave 3Q26 release (10 Nov 2026); S&P Global consensus (revenue and EPS 11–12 Nov 2026; EBIT and capex pre-3Q26); Penmark model (FY28E in Exhibit 10).
Revenue equation reasoning: Active power reaches 1.88GW at year-end 2026 (inside the >1.85GW guide), 2.96GW in FY27E (+1.08GW net) and 3.86GW in FY28E (+0.90GW), staying within the 4.6GW contracted footprint. Fleet revenue/MW rises from $9.7m in FY26E to $12.5m in FY28E and $14.3m by FY30E as newer clusters replace lower-priced legacy racks.
Fixed vs variable cost leverage: Cash operating costs grow slowly from $3.7m/MW in FY26E to $4.2m/MW in FY28E, lifting adjusted EBITDA margins from 61% to 66%. As committed capex decelerates in FY28E, depreciation falls from 51% of sales in FY26E to 45% in FY28E, lifting adjusted operating margin to 21.5% (and 29.3% in FY30E, inside management’s 25%–30% goal).
Net debt peaks at about $71bn in FY28E before dropping to 1.3x EBITDA by FY30E
CoreWeave’s balance sheet crosses its funding peak over the next twenty-four months. Combining $39bn of FY27E committed capex with $11bn of scheduled debt repayments creates a $50bn cash need, fully covered in our base case by $18bn of operating cash flow (including $12bn gross prepayments), $30bn of new secured DDTL draws ($1.4bn covers in-period interest) and $1.7bn of ATM equity.
Source: Penmark Research model (FY27E base case; cash held at $4.0bn minimum).
Source: Penmark Research model (FY26E–FY32E).
| Driver sensitivity (all else equal) | Confirmed stock driver | Base DCF / sh | vs Base ($41.02) | % vs Price ($47.83) |
|---|---|---|---|---|
| Fleet revenue/MW +$0.5m across curve | Pricing Power & Renewals | $77.73 | +$36.71 | +62.5% |
| Fleet revenue/MW −$0.5m across curve | Pricing Power & Renewals | $4.30 | −$36.71 | −91.0% |
| Fleet revenue/MW +$0.3m from FY29E | Pricing Power & Renewals | $61.21 | +$20.19 | +28.0% |
| WACC −1.0pp (to 9.18%) | Financing & Cost of Capital | $72.77 | +$31.75 | +52.1% |
| WACC +1.0pp (to 11.18%) | Financing & Cost of Capital | $17.12 | −$23.89 | −64.2% |
| Equity beta at 3.71 (vs 2.67 base) | Financing & Cost of Capital | $5.56 | −$35.46 | −88.4% |
| Secured debt share 80% (IG DDTL closes) | Financing & Cost of Capital | $47.30 | +$6.29 | −1.1% |
| FY27E net power additions cut to 0.85GW | Power Delivery & Capex Pace | $38.18 | −$2.83 | −20.2% |
| Customer & prepayment downside (−$6bn, −$0.25m/MW) | Pricing & Financing | $22.66 | −$18.36 | −52.6% |
| Parent ATM equity issued at $40 (vs $47.83) | Financing & Cost of Capital | $40.60 | −$0.42 | −15.1% |
Source: Penmark Research model (Sensitivities tab; base DCF $41.02 before 12m roll-forward). FY27E prepayments +$6bn has $0.00/sh DCF impact because prepayment inflows are excluded from valued FCFF.
Interest coverage inflects rapidly in FY27E: Adjusted EBITDA less cash interest more than doubles in FY27E (2.34x FY26E, confirming management’s “more than double” guidance), while adjusted net income flips positive to $2.7bn in FY28E.
Free cash flow trajectory: Excluding customer prepayments, free cash flow improves from −$34bn in FY26E to −$29bn in FY27E and −$12bn in FY28E before turning positive at +$1.6bn in FY29E and +$7.6bn in FY30E.
DCF values CoreWeave at $41 vs $54 at 14x FY28E EBIT; 6-year refresh explains the gap
We value CoreWeave primarily via a DCF of free cash flow to the firm excluding customer prepayments, cross-checked against 14.0x FY28E adjusted operating income. A multi-year DCF is essential for a capacity-led infrastructure build because near-term earnings are negative while six-year GPU replacement waves alter long-run cash conversion.
Source: Penmark Research model (18 Nov 2026 close $47.83; target $47.78). *Street bar capped at $150 for scale; full range $45–$278.
| DCF component (base case) | Value | Notes |
|---|---|---|
| PV of 4Q26E–FY38E FCFF | 22,915 | Excl. prepayments |
| PV of terminal value | 55,062 | 70.6% of EV |
| Enterprise value (30 Sep 26) | 77,976 | WACC 10.18% |
| Less: debt principal & leases | (48,154) | $47.5bn + $0.65bn |
| Less: deferred rev & cust. liab. | (11,902) | $10.0bn + $1.9bn |
| Plus: cash, restr. & mkt. sec. | 6,466 | 30 Sep 2026 |
| Net claims ahead of equity | (53,590) | Deducted at par |
| Equity value (30 Sep 26) | 24,386 | Rolled to 18 Nov |
| Base DCF value / share ($) | $41.02 | 646.1m dil. sh* |
| EV/EBIT cross-check (14x FY28E) | $53.75 | +$12.73 vs DCF |
Source: Penmark Research model (*incl. 35m FY27E ATM shares; proceeds in DCF).
WACC and terminal assumptions: Our WACC (the blended return required by equity holders and lenders) is 10.18%: 5.10% risk-free rate, 4.14% equity risk premium, and 1.98 relevered beta (from a 2.67 regression beta) give a 13.29% cost of equity, paired with a 7.07% after-tax cost of debt (8.95% pre-tax, 70% secured / 30% unsecured) at 50/50 target leverage. Terminal growth (steady annual cash-flow growth beyond FY38E) is 3.0% at a 12.0% terminal return on new invested capital (RONIC).
Why the two valuation methods differ by 31% ($13/sh): Our 14.0x FY28E EV/EBIT cross-check yields $54/sh versus $41/sh in the DCF (which implies 13.0x FY28E EBIT). A single-year FY28E multiple ignores two drags captured in our 12-year DCF: $65bn of FY32E–FY33E six-year GPU refresh capex and post-2030 ROIC fade toward 12.0%.
| Company (Ticker) | Type | EV ($bn) | Sales ($bn) | YoY | EBITDA | EV/Sales | EV/EBITDA | EV/EBIT |
|---|---|---|---|---|---|---|---|---|
| CoreWeave (CRWV): 3Q26 ann. | AI Cloud | 70.9* | 14.6 | 168% | 57.3% | 4.8x | 8.5x | 85.6x |
| CoreWeave: FY28E ($47.83) | Forward | 82.8 | 42.6 | 59% | 66.4% | 1.9x | 2.9x | 9.0x |
| Oracle (ORCL) | Cloud | 429.9 | 77.4 | 30% | 52.1% | 5.6x | 10.7x | 16.0x |
| Microsoft (MSFT) | Hyperscaler | 3,717.9 | 369.1 | 19% | 62.9% | 10.1x | 16.0x | 20.2x |
| Amazon (AMZN) | Hyperscaler | 3,006.7 | 807.8 | 12% | 24.3% | 3.7x | 15.3x | 26.8x |
| Equinix (EQIX) | DC REIT | 124.3 | 10.3 | 11% | 44.5% | 12.1x | 27.2x | 51.8x |
| Digital Realty (DLR) | DC REIT | 86.4 | 7.6 | 21% | 44.2% | 11.4x | 25.7x | 71.5x |
| IREN / Applied Digital | Neocloud | 12.8 / 7.7 | 0.9 / 1.4 | −5 / 432% | −17 / −6% | 14.0 / 5.6x | n.m. | n.m. |
| Cloud median (ORCL/MSFT/AMZN) | Core | n.a. | n.a. | 19% | 52.1% | 5.6x | 15.3x | 20.2x |
Source: Signals; CoreWeave & peer 3Q26 filings (18 Nov 2026). *Standard peer-basis EV excludes $11.9bn prepayments; forward EV ($82.8bn) includes prepayments and diluted shares. CRWV P/E at $47.78 target = 19.7x FY28E GAAP EPS.
At $48, the market prices slightly more than our base case; Hold is balanced
Source: Penmark Research model (base case centered at 10.18% WACC, 3.0% g = $41.02).
Source: Penmark Research model (base centered at $14.3m/MW FY30E, $36m/MW capex).
| Scenario (10.2% WACC*) | Prob. | FY30E Rev/MW | YE30E GW | Capex/MW | Value today | Value ret. | 12m target* | 12m ret. |
|---|---|---|---|---|---|---|---|---|
| Bull case (supply tight) | 20% | $15.0m | 6.5GW | $35.0m | $98.08 | +105.1% | $111.11 | +132.3% |
| Base case (committed capex) | 55% | $14.3m | 5.3GW | $36.0m | $41.02 | −14.2% | $46.47 | −2.9% |
| Bear case (supply wave) | 25% | $11.9m | 2.7GW | $37.0m | $0.00 | −100.0% | $0.00 | −100.0% |
| Probability-weighted | 100% | n.a. | n.a. | n.a. | $42.17 | −11.8% | $47.78 | −0.1% |
Source: Penmark Research model (Scenarios tab). *All three scenarios use our 10.18% WACC; values rolled forward 12 months at 13.29% cost of equity (no dividend). Rating = 12m total return to prob-weighted target (−0.1% = Hold).
Target reconciliation to Page 1: Weighting Bull ($98 × 20%), Base ($41 × 55%) and Bear (nil × 25%) produces a present value of $42/sh (within 3% of base). Rolling forward twelve months at our 13.3% cost of equity yields our $48 12-month target (−0.1% total return), firmly in our −10% to +10% Hold band.
Limits of probability weights & Low conviction: Our 25% bear weight is anchored by senior notes valued at 87.8% of par at 30 September (~25% real-world five-year default odds). Because shifting WACC by 1 point either way flips the rating to Buy ($77) or Sell ($25), we assign Low conviction and await December DDTL and February renewal prints.
December DDTL and late-February FY27 guidance are the decisive rating triggers
| Date | Catalyst event | Penmark expectation vs consensus / market | Linked driver |
|---|---|---|---|
| 9 Dec 2026 | Federal Reserve FOMC meeting | Rate hold or dovish tone eases 10y UST below 4.75%; hawkish tone widens AI credit spreads. | Financing & WACC |
| Dec 2026 | Investment-grade DDTL close | We expect ≥$4bn closed at 7.0%–7.5% (>90% advance), easing near-term bond stress (+$6/sh). | Financing & WACC |
| Late Feb 2027 | 4Q26 results & FY27 guidance | Penmark: FY27 rev $26.9bn (vs $26.5bn Street), capex $39bn (vs $47bn Street) & >130k GPU renewals. | Pricing & Capex |
| 1Q–2Q 2027 | Texas permitting legislative session | Permitted CoreWeave sites continue unaffected; clarity would ease concern after 4 Nov (−9.9%). | Power Delivery |
| May / Aug 2027 | 1Q27 & 2Q27 execution prints | Active power ramps toward 2.4GW mid-year; EBITDA less interest tracks >2x FY26 level. | Power & Pricing |
| 1H 2028 | First positive adjusted net income | Inflection to positive adjusted net income (model FY28E $2.7bn) validates operating leverage. | Pricing & Capex |
Source: CoreWeave 3Q26 call (10 Nov 2026); Bloomberg (Nov 2026); S&P Global consensus (revenue and EPS 11–12 Nov 2026; EBIT and capex pre-3Q26); Penmark Research estimates.
| Indicator to watch | Upgrade trigger (Move to Buy) | Downgrade trigger (Move to Sell) |
|---|---|---|
| 2027 GPU renewal pricing | >130k paid-off GPUs re-contract at premiums tracking +$0.3m/MW (+$22/sh) or +$0.5m/MW (+$37/sh). | Renewal pricing compresses toward $11.9m/MW or customers walk from expiring clusters. |
| Secured funding & WACC | ≥$4bn IG DDTL closes ≤7.5% (lifts base to $47) or WACC drops 1pt (lifts target to $77, Buy). | DDTL slips or prices >8.5%, WACC rises 1pt (cuts target to $25, Sell), or equity beta nears 3.71. |
| FY27 capex & active GW | FY27 capex outlook lands near $37–$39bn with ≥1.08GW active additions and ≥$10bn prepayments. | Uncontracted capex tops $45bn or permitting cuts FY27 net additions below 0.85GW. |
Source: Penmark Research model (Sensitivities and Scenarios tabs).
| Risk factor | Likelihood | Value impact | Transmission channel & thesis impact |
|---|---|---|---|
| Post-2027 supply wave | Medium | −$37/sh (−$0.5m/MW) | Hyperscaler buildouts and direct lab shell leasing cap fleet pricing near $11.9m/MW, driving bear value to nil. |
| Credit squeeze / high beta | Med-High | −$24 to −$35/sh | +1pt WACC or a 3.71 equity beta overwhelms operating beats via higher discount rates (equity dilution itself costs under $1/sh). |
| Customer / prepayment loss | Low-Med | −$18/sh (downside) | Top two customers are 54% of 3Q26 sales; $1.9bn of prepaid revenue is refundable under cancellable contracts. Customer and prepayment downside (−$6bn, −$0.25m/MW) cuts $18/sh. |
| Power / ESG grid delays | Medium | −$3/sh (0.23GW lag) | Texas permitting pauses, utility interconnection queues or carbon/water mandates delay active megawatts. |
| Governance & key-person | Low-Med | Multiple discount | Insiders hold 72% voting power (10:1 Class B); NVIDIA supplies ~100% of GPUs and holds 8.5% basic shares. |
| Upside risk (to Hold) | Medium | +$22 to +$57/sh | Tight 2027–2028 supply lifting fleet pricing to $15.0m/MW before fading to $13.7m, with 6.5GW power and $35m/MW capex, drives bull value of $98. |
Source: CoreWeave 3Q26 Form 10-Q; Penmark Research model (Sensitivities and Scenarios tabs).
Short interest climbed to 16% of Class A while 42 Street analysts average a $123 target
CoreWeave combines deep trading liquidity ($3.1bn three-month average daily value, or ADV) with polarized positioning. Institutional investors reported 382.4m shares (82.1% of 465.8m tracked Class A shares) at 30 September 2026, led by Magnetar and NVIDIA, while short interest jumped 35% between mid-September and end-October to 74.2m shares.
| Institutional holder | Shares (m) | QoQ (m) | Value ($m)* |
|---|---|---|---|
| Magnetar Financial LLC | 47.69 | −4.38 | 4,154.5 |
| NVIDIA Corp. | 47.21 | 0.00 | 4,113.2 |
| Vanguard Group | 38.75 | +3.20 | 3,376.0 |
| BlackRock, Inc. | 29.72 | +5.61 | 2,589.1 |
| Invesco Ltd. | 12.67 | +0.26 | 1,103.8 |
| Value Aligned Research | 11.63 | +0.75 | 1,013.4 |
| Aspex Management | 10.31 | +2.35 | 898.3 |
| Total 13F reported (998) | 382.44 | n.a. | 33,318 |
Source: Form 13F filings via Signals (as of 30 Sep 2026). *Reported at 30 Sep 2026 quarter-end price ($87.12). Missing filers in later quarters are not treated as confirmed exits.
Source: FINRA short interest via Signals (15 Apr to 30 Oct 2026; 16.0% of Class A). Days to cover = 1.07 on FINRA settlement-period volume (69.2m/day), or 1.6 days on 3-month average volume (46.9m/day).
| Trading & Street metric | Current reading | Analytical significance |
|---|---|---|
| 3-month average daily value / volume | $3,110m / 46.85m sh | Deep liquidity; 74.2m shares short equals 1.6 days to cover on 3-month volume (1.07 days on FINRA settlement-period volume). |
| 52-week trading range (to 18 Nov 26) | $40.58 – $138.25 | Stock sits 65% below 52-week peak and 18% above 52-week low after October credit stress. |
| Street consensus target (42 analysts) | Mean $123 ($45–$278) | Penmark’s $48 target sits 61% below Street mean because our DCF deducts prepayments and 6-year refresh. |
| Recent insider Form 4 activity | 15 sales ($17.4m) / 0 buys | 30 Sep–10 Nov 2026 filings (incl. COO & CFO); zero open-market purchases observed. |
| Share overhangs & dilution | 35m ATM + 57m dilutive | Sep 2026 ATM program (modeled at $1.7bn FY27E) plus options/RSUs lift diluted count to 646.1m. |
Source: Signals (18 Nov 2026); Yahoo Finance (18 Nov 2026); Street price targets (11–14 Nov 2026); CoreWeave 3Q26 Form 10-Q.
Why our target sits below the $123 Street mean: Our $48 target rests on a 12-year FCFF DCF that deducts $12bn of customer prepayments at par in the equity bridge, models $65bn of FY32E–FY33E six-year GPU refresh spending, and applies a 10.2% WACC.
Positioning creates two-way squeeze risk: With 16% of Class A shares sold short and 3Q26 results sparking a +24.7% single-day jump, closing the December investment-grade DDTL could trigger sharp short covering even if long-run renewal debates remain unresolved.
Annual financial statements (FY24A–FY30E) and quarterly progression (3Q25A–4Q26E)
| US$m (unless stated) | FY24A | FY25A | FY26E | FY27E | FY28E | FY29E | FY30E |
|---|---|---|---|---|---|---|---|
| Income statement & operating KPIs | |||||||
| Active power, year-end (GW) | n.d. | 0.85 | 1.88 | 2.96 | 3.86 | 4.61 | 5.26 |
| Revenue / avg active MW ($m) | n.d. | n.d. | 9.68 | 11.10 | 12.50 | 13.80 | 14.30 |
| Revenue | 1,915 | 5,131 | 13,214 | 26,862 | 42,625 | 58,443 | 70,571 |
| Revenue growth (%) | 736.2% | 167.9% | 157.5% | 103.3% | 58.7% | 37.1% | 20.8% |
| Cash operating costs | (696) | (2,038) | (5,106) | (9,922) | (14,331) | (18,243) | (21,789) |
| Adjusted EBITDA | 1,219 | 3,093 | 8,108 | 16,940 | 28,294 | 40,200 | 48,781 |
| Adjusted EBITDA margin (%) | 63.7% | 60.3% | 61.4% | 63.1% | 66.4% | 68.8% | 69.1% |
| Depreciation & amortisation | (863) | (2,454) | (6,779) | (13,186) | (19,126) | (24,039) | (28,093) |
| Stock-based comp. & other | (32) | (685) | (645) | (940) | (1,279) | (1,607) | (1,764) |
| Operating income (GAAP) | 324 | (46) | 684 | 2,814 | 7,890 | 14,554 | 18,924 |
| Adjusted operating income (EBIT) | 356 | 666 | 1,375 | 3,754 | 9,169 | 16,162 | 20,688 |
| Adjusted operating margin (%) | 18.6% | 13.0% | 10.4% | 14.0% | 21.5% | 27.7% | 29.3% |
| Interest expense, net | (361) | (1,229) | (3,122) | (5,269) | (6,464) | (6,806) | (6,517) |
| Income tax & other | (826) | 108 | (147) | (330) | 0 | (988) | (2,934) |
| GAAP net income (loss) | (863) | (1,167) | (2,585) | (2,784) | 1,426 | 6,760 | 9,473 |
| Adjusted net income (ex SBC) | (65) | (606) | (2,030) | (1,844) | 2,705 | 8,367 | 11,237 |
| GAAP EPS ($) | (4.30) | (2.81) | (4.72) | (4.87) | 2.42 | 11.48 | 16.08 |
| Cash flow & balance sheet | |||||||
| Operating cash flow (CFO) | 2,749 | 3,058 | 12,647 | 17,924 | 22,326 | 30,458 | 36,848 |
| of which: net prepayments | 2,049 | 4,174 | 8,892 | 7,675 | 1,756 | (683) | (1,512) |
| Capex (company definition) | (8,702) | (14,845) | (38,000) | (38,880) | (32,400) | (29,502) | (30,797) |
| FCF excl. customer prepayments | n.d. | (15,961) | (34,245) | (28,630) | (11,830) | 1,639 | 7,563 |
| Cash, restricted cash & sec. | 2,064 | 4,164 | 4,000 | 4,000 | 4,000 | 4,000 | 4,000 |
| Property & equipment, net | 11,915 | 30,557 | 61,778 | 87,472 | 100,747 | 106,209 | 108,913 |
| Total debt (principal) | 7,926* | 21,373* | 45,380 | 64,662 | 74,737 | 73,781 | 67,730 |
| Customer prepayments (BS liab.) | 4,064 | 8,185 | 17,302 | 24,977 | 26,732 | 26,049 | 24,537 |
| Stockholders’ equity | (414) | 3,335 | 3,680 | 3,510 | 6,215 | 14,582 | 25,819 |
| Net debt / Adjusted EBITDA (x) | 4.8x | 5.6x | 5.1x | 3.6x | 2.5x | 1.7x | 1.3x |
Source: CoreWeave filings (FY24A–FY25A); Penmark Research model (FY26E–FY30E). *FY24A/FY25A debt on carrying-value basis; FY26E–FY30E at principal.
| Quarterly metric | 3Q25A | 4Q25A | 1Q26A | 2Q26A | 3Q26A | 4Q26E |
|---|---|---|---|---|---|---|
| Revenue ($m) | 1,365 | 1,572 | 2,078 | 2,575 | 3,661 | 4,900 |
| Adjusted EBITDA ($m) | 838 | 898 | 1,157 | 1,510 | 2,256 | 3,185 |
| Adjusted EBITDA margin (%) | 61.4% | 57.1% | 55.7% | 58.6% | 61.6% | 65.0% |
| Adjusted operating income ($m) | 217 | 88 | 21 | 128 | 379 | 847 |
| GAAP net income (loss) ($m) | (110) | (452) | (740) | (626) | (709) | (510) |
| Capex, company definition ($bn) | 1.9 | 8.2 | 6.8 | 9.4 | 14.5 | 7.4 |
| Revenue backlog ($bn) | 55.6 | 66.8 | 99.4 | 104.2 | 132.5 | n.e. |
| Active power, period-end (GW) | n.d. | 0.85 | 1.00 | 1.50 | 1.70 | 1.88 |
Source: CoreWeave quarterly releases and 3Q26 Form 10-Q (3Q25A–3Q26A); Penmark Research model (4Q26E).
Methodology, ratio definitions, glossary and source log (as of 18 November 2026)
Operating model: Capacity and utilization (specialized AI cloud infrastructure). Lead revenue equation: Revenue = Average Active Power (MW) × Revenue per MW. Primary valuation is a 12-year FCFF DCF excluding customer prepayments (valued at 30 Sep 2026, rolled to 18 Nov 2026, then rolled 12 months at 13.29% cost of equity for target price). Cross-check: 14.0x FY28E adjusted operating income.
Key structural features: Recently listed (IPO 28 Mar 2025; 20 months of trading history), leveraged ($47.5bn debt principal), dual-class controlled (Class B 10 votes/sh; 72% insider voting control), and high customer concentration (54% from top two customers in 3Q26).
Valuation & metric conventions: (1) Because GAAP EPS is negative through FY27E, Exhibit 11 tracks adjusted EBIT vs consensus, Exhibit 15 tracks historical EV/run-rate sales, and Exhibit 16 plots EV/EBITDA vs EBITDA margin rather than P/E; (2) Customer prepayments ($11.9bn balance) are stripped from valued FCFF and deducted as claims ahead of equity to avoid double-counting financing cash.
Adjusted EBITDA: GAAP operating income plus depreciation and amortisation (D&A), stock-based compensation (SBC) and acquisition-related costs.
Adjusted operating income (Adj. EBIT): Adjusted EBITDA less D&A; historical figures add back SBC, acquired-intangible amortisation and deal costs, while forecasts include acquired-intangible amortisation within D&A.
FCF ex-prepayments: Operating cash flow less net customer prepayment inflows less company-definition capital expenditures.
Enterprise Value (model basis): Diluted equity value (611.1m shares × $47.83) plus $47.5bn debt principal, $0.65bn finance leases, $10.0bn deferred revenue and $1.9bn refundable customer liabilities less $6.47bn cash, restricted cash and marketable securities = $82.82bn.
WACC & Reverse DCF: WACC (10.18%) blends 13.29% cost of equity and 7.07% after-tax cost of debt at 50/50 target leverage. Reverse DCF solves for the pricing or WACC implied by the $47.83 share price.
Active Power (GW/MW): Utility power energized and available to run live GPU servers inside operating data centers (1.7GW at 3Q26).
Contracted Power (GW): Total utility power committed under lease or grid agreements, including sites still under construction (4.6GW at 3Q26).
DDTL (Delayed Draw Term Loan): Asset-backed project debt drawn against GPU equipment and customer contracts, self-amortising over contract life.
GPUaaS / Neocloud: GPU-as-a-Service; specialized cloud providers operating bare-metal NVIDIA clusters for AI training and inference.
Mission Control / Slurm on Kubernetes: CoreWeave’s proprietary cluster automation and workload scheduling software stack.
Revenue Backlog: Remaining performance obligations (RPO) plus committed contract values estimated to be recognized ($132.5bn at 3Q26).
Valuation & model: Penmark Research valuation model (cut-off 18 Nov 2026; official close $47.83).
Company filings: CoreWeave FY25 Form 10-K, 3Q26 Form 10-Q (filed 12 Nov 2026), DEF 14A (record date 15 Apr 2026; filed 22 Apr 2026), 8-K & 3Q26 call (10 Nov 2026) via Signals.
Market & peer data: Signals (peer 3Q26 filings, 13F as of 30 Sep 2026, Form 4 through 10 Nov 2026, FINRA short interest 30 Oct 2026); Yahoo Finance daily prices (to 18 Nov 2026).
Consensus & credit: S&P Global consensus (42 analysts; FY26E–FY27E revenue & FY27E EPS post-3Q26 [11–12 Nov 2026]; EBITDA, EBIT, capex & interest pre-3Q26, Oct 2026); Bloomberg (11 Nov 2026 IG DDTL); U.S. Treasury 10y yield 5.10% (17 Nov 2026 close).
Peer set (7 comps): Cloud/hyperscalers (ORCL, MSFT, AMZN), data-center REITs (EQIX, DLR), specialized neoclouds (IREN, APLD).
| $m | FY24A | FY25A | FY26E | FY27E |
|---|---|---|---|---|
| GAAP operating income | 324 | (46) | 684 | 2,814 |
| Stock-based compensation | 31 | 630 | 643 | 940 |
| Acq. costs & intang. amort. | 1 | 82 | 48 | 0 |
| Adjusted operating income | 356 | 666 | 1,375 | 3,754 |
Source: CoreWeave Form 10-K, 3Q26 Form 10-Q; Penmark Research model.
Analyst certification and important disclosures
Ratings
Ratings are based on the total return from the share price to our 12-month price target, including dividends. The price target is weighted across bull, base and bear scenarios.
| Buy | Total return to price target above +10% |
| Hold | Total return to price target from −10% to +10%, inclusive |
| Sell | Total return to price target below −10% |
We review ratings at each set of results and whenever the facts change materially. A share price can move outside its band before the rating changes, and targets may not be reached.
Analyst certification
The Penmark Team certifies that (1) the views in this report accurately reflect its views about the subject company and its securities, and (2) no part of its compensation was, is or will be directly or indirectly related to the specific ratings, price targets or views in this report.
Contact
Penmark Research
Questions about our research:
research@penmarkresearch.com
All our research and coverage:
penmarkresearch.com
Basis of preparation
This report uses information publicly available up to its publication date. Share prices are closing prices on the trading day before the publication date, unless stated otherwise. Forecasts, valuations and price targets are Penmark Research’s own, and forecast periods are marked E. Sources are shown beneath each exhibit.
Valuation and price targets
Price targets are based on the valuation methods set out in each report and are weighted across bull, base and bear scenarios. The main risks to each rating and price target are described in the report. Penmark Research updates its coverage after each set of results but is not obliged to update a report at any other time.
Independence and conflicts of interest
Penmark Research is independent. It does not provide investment banking, brokerage, market-making or asset management services, does not trade securities, and does not accept payment from any company for research coverage. Companies do not see or review Penmark Research reports before publication.
No member of the Penmark Team holds a long or short position, directly or through derivatives, in the securities of any company Penmark Research covers.
Regulatory status
Penmark Research publishes general, impersonal investment research on a regular basis. It is not registered as an investment adviser with the US Securities and Exchange Commission or any state securities authority, is not a registered broker-dealer and is not a member of FINRA. Penmark Research does not give advice tailored to any person’s circumstances.
Not investment advice or an offer
This report is for information only. It is not personalised investment advice and is not an offer or solicitation to buy or sell any security or other financial instrument. It does not take into account any reader’s objectives, financial situation or needs. Readers should make their own investment decisions and seek independent professional advice where appropriate.
Forward-looking statements and risk
Forecasts, valuations and price targets are estimates based on assumptions that may prove wrong, and they may not be achieved. Opinions are current as of the publication date and may change without notice. Investing in equities involves risk, including the loss of capital. Share prices can be volatile, and past performance is not a reliable indicator of future results.
Data and sources
Company financials, operating data and filings come from company filings, earnings releases and earnings calls. Market data and any consensus estimates come from third-party sources named beneath the relevant exhibit; consensus figures from aggregators are unofficial. Penmark Research does not use other brokers’ research. Third-party data belong to their providers and may not be extracted from this report.
Accuracy and liability
Information is taken from sources believed to be reliable, but Penmark Research makes no representation or warranty as to its accuracy, completeness or timeliness. To the fullest extent permitted by law, Penmark Research accepts no liability for any loss arising from use of this report. Descriptions of litigation and regulatory matters reflect allegations and public records, not findings, unless stated. If a published report is corrected, the corrected version carries an “Updated” date at penmarkresearch.com.
Distribution and access
Penmark Research reports are available to registered users of penmarkresearch.com. Downloads and programmatic access are governed by the Penmark Research subscription terms. This report is not directed at, or intended for, any person in Singapore or India, and access from those countries is blocked. It is not intended for distribution to, or use by, any person in any jurisdiction where that would be contrary to law or regulation or would subject Penmark Research to registration or licensing requirements there.
Copyright and permitted use
© Penmark Research. All rights reserved. Penmark Research is a trade name of Brightriver Labs. Readers may share a link to this report but may not copy, reproduce, redistribute, sell or use its content, including for text and data mining, without written consent. Contact: research@penmarkresearch.com.