
Sell: $88bn capex and fading GPU yields put target at $96.91 (−24.7%)
We initiate on Nebius with a Sell rating and $96.91 target (−24.7% from $128.64). Nebius builds liquid-cooled AI GPU clusters across US and European sites. Even after halving in three months, $128.64 prices $17.05m/MW long-run yields on depreciating GPUs.
- The price needs $17m/MW on every megawatt forever; we see $15m once supply catches up. Each $0.75m/MW shift in long-run revenue yield is worth about $30/sh as 200+ neoclouds and hyperscaler capacity normalise renewals.
- $88bn of FY26–28E capex keeps core FCF negative until FY30E. Customer prepayments save interest, not principal: deferred-revenue obligations and funding credit net to +$0.49bn across our $45.45/sh base DCF.
- Conviction is Low because 76% of value sits in a $266 bull case. Probability weighting across Bear ($0, 30%), Base ($45.45, 45%) and Bull ($265.54, 25%) yields $86.84 today, rolling at 11.6% cost of equity to $96.91.
What the market is missing. A pre-print $276.20 Street mean target leans on single-year multiples and 3Q26 sales ($980.8m, a third of which — ~$325m — was non-recurring asset-light revenue), ignoring $88.2bn of FY26–28E capex and negative core FCF to FY29E.
Catalysts. Hyperscaler capex guides (Jan 27), Feb 2027 ARR vs >$7bn goal and 4Q26 print (Feb 27), delayed 3P colocation handovers and Vera Rubin $/MW capex (May 27), and Oct 2027 bridge refinancing.
Valuation and target. FY40E DCF (10.0% WACC, 3.0% g) gives $45.45/sh base value vs $68.68/sh on 4.37x EV/CY27E EBITDA. Scenario weighting gives $86.84/sh today and a $96.91 12-month target (−24.7%, Sell).
Risks. Upside risks: >$25m/MW deal pricing holds across FY27 (+$8.56/sh), Feb ARR clears $7bn with Q2 site handovers (+$5.97/sh), or asset-light mix adds +3pt margin (+$5.07/sh).
| Dec YE, US$m | FY24A | FY25A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| Revenue | 91.5 | 529.8 | 3,131.4 | 12,631.1 | 26,695.5 |
| Growth y/y | +833.7% | +479.0% | +491.0% | +303.4% | +111.3% |
| Adj. EBITDA | (226.3) | (64.9) | 1,267.7 | 6,140.2 | 15,323.7 |
| Adj. EBITDA mgn | nm | −12.2% | 40.5% | 48.6% | 57.4% |
| Net income (GAAP) | (352.0) | 9.8 | 129.0 | (535.6) | 1,920.3 |
| Diluted EPS ($) | nr | nr | 0.40 | (1.68) | 6.01 |
| FCF incl. adv. | (1,077.4) | (3,664.1) | (21,259.4) | (17,642.2) | (14,246.6) |
| FCF excl. adv. | (1,077.4) | (4,646.6) | (30,588.4) | (26,642.2) | (17,246.6) |
| At $128.64 spot | FY24A | FY25A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| P/E | nm | nm | 318.9x | nm | 21.4x |
| EV / Adj. EBITDA | nm | nm | 37.2x | 7.7x | 3.1x |
| P/B | nr | 8.8x | 3.4x | 3.5x | 2.9x |
| FCF yield (ex-adv.) | −2.7% | −11.5% | −75.7% | −66.0% | −42.7% |
| Dividend yield | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
| ROE (avg. equity) | nr | nr | 1.5% | −4.5% | 14.6% |
| Net debt / EBITDA | nm | nm | 13.3x | 5.6x | 3.2x |
| P/E at $96.91 target | nm | nm | 240.2x | nm | 16.1x |
Source: Company filings, Penmark Research model; stock chart via Yahoo Finance. Market cap uses 313.97m shares (30 Sep shares + NVIDIA pre-funded warrants); per-share figures use 319.71m diluted shares (adding RSUs; converts count when in the money). *Principal net debt ($6,778m; carrying $6,765m) and EV ($47,167m) exclude deferred revenue. †Pre-print Street mean ($276.20, 22 analysts, S&P Global) and Oct ratings (21 analysts, 24/7 Wall St); nr = not reported, nm = not meaningful.
Base DCF is $45.45/sh ($50.72 in 12M); scenario skew lifts our 12-month target to $96.91 (−24.7%, Sell)
Source: Yahoo Finance; Penmark Research model. Price-date DCF values ($265.54 / $45.45 / $0.00; weighted $86.84) rolled 12 months at 11.6% cost of equity.
880 MW YE26, +1,300 MW/yr FY27E–29E; scarcity yield $17.5–19.5m/MW, $17.0m long run; AI Cloud margin 54% rising to 65% (1 of 4 2026 goals met on time).
720 MW YE26, +1,000 MW/yr FY27E–28E; yield $16.0m FY27E, $17.5m FY28E, fading to $15.0m FY32E+; margin 50% rising to 61% (residual weight).
GPU glut from 2028 cuts yield to $11m/MW; expansion stops once contracted builds finish in FY29E; at a 50% margin, fleet replacement at $30m/MW earns less than its cost, leaving no equity value (rents reset once in 3 yrs).
Weight sensitivity (Penmark arithmetic rolled at 11.6%): Bear/Base/Bull 25/45/30 gives ~$111.7 (Sell); 20/45/35 gives ~$126.5 (Hold).
| Income stmt, $m | FY24A | FY25A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| Revenue | 91.5 | 529.8 | 3,131.4 | 12,631.1 | 26,695.5 |
| Adjusted EBITDA | (226.3) | (64.9) | 1,267.7 | 6,140.2 | 15,323.7 |
| Share-based comp. | (54.5) | (83.2) | (256.5) | (315.8) | (587.3) |
| One-offs & other | (41.7) | (45.7) | (62.1) | 0.0 | 0.0 |
| Deprec. & amort. | (77.1) | (417.9) | (1,346.0) | (5,404.6) | (10,516.9) |
| Operating EBIT | (399.6) | (611.7) | (396.9) | 419.8 | 4,219.5 |
| Net financial & reval. | 46.6 | 625.5 | 508.6 | (955.4) | (1,925.7) |
| Income tax | 1.0 | (4.0) | 17.3 | 0.0 | (373.5) |
| Net income (cont.) | (352.0) | 9.8 | 129.0 | (535.6) | 1,920.3 |
| Cash flow, $m | FY24A | FY25A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| Net income | (352.0) | 9.8 | 129.0 | (535.6) | 1,920.3 |
| Deprec. & amort. | 77.1 | 417.9 | 1,346.0 | 5,404.6 | 10,516.9 |
| Share-based comp. | 54.5 | 83.2 | 256.5 | 315.8 | 587.3 |
| Def. rev. chg. | 0.0 | 982.5 | 9,329.0 | 9,000.0 | 3,000.0 |
| Working cap. | (49.5) | (1,091.5) | (5,014.1) | (475.0) | (703.2) |
| Operating CF | (269.9) | 401.9 | 6,046.4 | 13,709.8 | 15,321.4 |
| Cash capex | (807.5) | (4,066.0) | (27,305.8) | (31,352.0) | (29,568.0) |
| FCF incl. adv. | (1,077.4) | (3,664.1) | (21,259.4) | (17,642.2) | (14,246.6) |
| FCF excl. adv. | (1,077.4) | (4,646.6) | (30,588.4) | (26,642.2) | (17,246.6) |
| Balance sheet, $m | FY25A | 3Q26A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| Cash & equiv. | 3,678.1 | 7,425.6 | 2,000.0 | 2,000.0 | 2,000.0 |
| Receivables & oth. | 1,033.3 | 5,734.8 | 5,734.8 | 6,209.8 | 6,913.0 |
| PP&E, intang. & GW | 5,573.0 | 20,029.5 | 31,384.9 | 57,332.3 | 76,383.4 |
| Investments & ROU | 2,146.2 | 5,554.2 | 5,554.2 | 5,554.2 | 5,554.2 |
| Total assets | 12,430.6 | 38,744.1 | 44,673.9 | 71,096.3 | 90,850.6 |
| Deferred revenue | 1,577.5 | 9,409.2 | 10,909.2 | 19,909.2 | 22,909.2 |
| Gross debt | 4,127.7 | 14,190.8 | 18,825.0 | 36,467.2 | 50,713.8 |
| Total equity | 4,594.0 | 12,299.2 | 12,094.9 | 11,875.0 | 14,382.7 |
| Per share & KPIs | FY25A | 3Q26A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| Diluted shares (m) | nr | 319.7 | 319.7 | 319.7 | 319.7 |
| Diluted EPS, GAAP ($) | nr | nr | 0.40 | (1.68) | 6.01 |
| FCF ex-adv. / sh ($) | nr | (59.52) | (95.68) | (83.33) | (53.95) |
| Book value / sh ($) | nr | 38.47 | 37.83 | 37.14 | 44.99 |
| Connected MW | nr | 407.6 | 720.0 | 1,720.0 | 2,720.0 |
| Billed IT load (MW) | 104.2 | 300.0 | 529.9 | 1,265.9 | 2,001.9 |
| Yield / MW ($m) | 12.0 | 12.0 | 12.6 | 16.0 | 17.5 |
| AI Cloud EBITDA % | 12.3% | 53.2% | 42.0% | 50.0% | 58.0% |
Source: Company filings, Penmark Research model. Shaded columns are estimates (E). Exhibit 7 shows 4Q26E run-rate yield ($12.6m/MW) and AI Cloud margin (42.0%) in the FY26E column; 3Q26A FCF/sh is 9M26A. Adjusted EBITDA excludes SBC, D&A, one-offs and non-operating items.
Margins reach 57.4% by FY28E, yet ROCE reaches only 7.4% on $76bn PP&E; we sit +3.2% above FY27E revenue consensus
| Growth & margins | FY24A | FY25A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| Growth y/y | |||||
| Sales | +833.7% | +479.0% | +491.0% | +303.4% | +111.3% |
| Adj. EBITDA | nm | nm | nm | +384.4% | +149.6% |
| EBIT | nm | nm | nm | nm | +905.1% |
| Diluted EPS | nr | nr | nm | nm | nm |
| DPS | nm | nm | nm | nm | nm |
| Margins and coverage | |||||
| Gross mgn* | 52.2% | 68.6% | nr | nr | nr |
| Adj. EBITDA % | nm | −12.2% | 40.5% | 48.6% | 57.4% |
| EBIT margin | nm | −115.5% | −12.7% | 3.3% | 15.8% |
| Core FCF mgn | nm | −877.0% | −976.8% | −210.9% | −64.6% |
| EBITDA / int. | nm | nm | 3.6x | 6.4x | 8.0x |
| Capex / sales | 882.5% | 767.5% | 872.0% | 248.2% | 110.8% |
| Ratios and valuation | FY24A | FY25A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|
| Returns and leverage | |||||
| ROE (avg. equity) | nr | nr | 1.5% | −4.5% | 14.6% |
| ROA (avg. assets) | nr | nr | 0.5% | −0.9% | 2.4% |
| ROCE (avg. cap. emp.) | nr | nr | −2.0% | 1.1% | 7.4% |
| Net debt / equity | nr | 9.8% | 139.1% | 290.2% | 338.7% |
| Net debt / Adj. EBITDA | nm | nm | 13.3x | 5.6x | 3.2x |
| Valuation at $128.64 share price | |||||
| P/E | nm | nm | 318.9x | nm | 21.4x |
| EV / Adj. EBITDA | nm | nm | 37.2x | 7.7x | 3.1x |
| P/B | nr | 8.8x | 3.4x | 3.5x | 2.9x |
| FCF yield (ex-adv.) | −2.7% | −11.5% | −75.7% | −66.0% | −42.7% |
| Dividend yield | 0.0% | 0.0% | 0.0% | 0.0% | 0.0% |
| P/E at $96.91 target | nm | nm | 240.2x | nm | 16.1x |
Source: Company filings, Penmark Research model. ROE, ROA and ROCE on average capital (definitions on page 28). *Gross margin is ex-D&A; forecast model drives segment Adjusted EBITDA directly without splitting COGS from SG&A/R&D, so FY26E–28E is marked nr.
| US$m / $/sh | 3Q26A | 4Q26E | FY26E | FY27E |
|---|---|---|---|---|
| Revenue ($m) | ||||
| Penmark (3Q26A) | 980.8 | 1,169.3 | 3,131.4 | 12,631.1 |
| Consensus (pre-print) | 907.5 | 1,481.2 | 3,370.0 | 12,240.0 |
| Prior-year period | 146.1 | 227.7 | 529.8 | 3,131.4 |
| Guidance midpoint | nr | 1,137.9 | 3,100.0 | 12,500.0 |
| Diff. vs consensus | +8.1% | −21.1% | −7.1% | +3.2% |
| Growth y/y (Penmark) | +571.3% | +413.5% | +491.0% | +303.4% |
| Diluted EPS ($/sh, GAAP vs Street) | ||||
| Penmark (3Q26A) | (0.13) | (0.83) | 0.40 | (1.68) |
| Consensus (pre-print) | (1.24) | nr | (2.10) | (3.60) |
| Adj. EBITDA ($m) | 465.8 | 436.2 | 1,267.7 | 6,140.2 |
| EPS diff. vs consensus | +$1.11 | nr | +$2.50 | +$1.92 |
| EBITDA diff. vs Street* | +32.4% | nr | −6.0% | nr |
Source: LSEG/S&P Global/CNBC pre-print consensus, Penmark Research model. *FY26E Street EBITDA = $1,348m (consensus revenue × ~40% guided margin). 4Q26E guide midpoint = $3,100m FY26 low-end guide midpoint less $1,962.1m 9M26A.
Source: Company filings (A); S&P Global/CNBC pre-print consensus; Penmark Research model (E).
At 7.7x FY27E EV/EBITDA, NBIS trades at a 76% premium to CoreWeave (4.4x) despite lower margin and an unproven GW build
| CY27E, spot prices | NBIS | CRWV | Diff. |
|---|---|---|---|
| Valuation (spot EV / consensus or Penmark) | |||
| EV / CY27E Adj. EBITDA | 7.68x | 4.37x | +76.0% |
| EV / CY27E sales (Penmark) | 3.73x | 2.75x | +35.8% |
| EV / CY27E sales (Street) | 3.85x | 2.75x | +40.1% |
| LTM EV / revenue | 21.5x | 9.5x | +127.3% |
| CY27E FCF yield (ex-adv.) | −66.0% | nr | nr |
| Dividend yield | 0.0% | 0.0% | 0.0pp |
| Scale, growth and profitability | |||
| Market cap ($bn) | 40.39 | 26.87 | +50.3% |
| Enterprise value ($bn) | 47.17 | 72.90 | −35.3% |
| CY27E sales growth (Penmark) | 303.4% | 105.4% | +198.0pp |
| CY27E sales growth (Street) | 263.2% | 105.4% | +157.8pp |
| CY27E Adj. EBITDA margin | 48.6% | 63.0% | −14.4pp |
| Net debt / CY27E EBITDA (spot)* | 1.10x | 2.76x | −1.65x |
Source: Company filings, Yahoo Finance, Penmark Research model. CoreWeave ($48.73, 10 Nov) is the sole listed pure-play AI cloud with published CY27E Adj. EBITDA consensus ($16.7bn on $26.5bn sales). *Uses spot principal net debt ($6.78bn NBIS, $46.03bn CRWV) vs 5.6x projected FY27E net debt/EBITDA.
- Jan 2027Hyperscaler 2027 capex guides: Microsoft/Meta AI budgets test 92% utilisation; MSFT neocloud pullback (−$3.97/sh) vs outsource lift (+$2.98/sh).
- Feb 20274Q26 print & Feb ARR: Test >$7bn Feb ARR goal, 4Q26E revenue ($1,169m) and YE26 connected power (720 MW est.).
- May 20271Q27 & Vera Rubin $/MW: Delayed NJ/3P colocation Q2 handover check (+$5.97 to −$8.96/sh) and Rubin IT capex vs $32m/MW base.
- Oct 2027Citi $3.5bn bridge maturity: Refinancing spread on the 15–25% unfunded 2027 capex slice vs 7.4% pre-tax cost of debt (−$10.45/sh risk).
- Funding cost spike. Debt +150bp on 15–25% uncovered 2027 capex with 10Y near 5%. 40% prob.; −$10.45/sh.
- 2027 site slippage. Owned/3P builds slip −300 MW vs 1.0 GW base. 35% prob.; −$8.96/sh.
- GPU rent reset. Supply catch-up cuts FY27E yield −$1.5m/MW. 30% prob.; −$8.56/sh.
- Upside risk: >$25m/MW pricing holds. Q3 deal yields carry across FY27 bookings (+$1.5m/MW). 30% prob.; +$8.56/sh.
Source: Shareholder letters; Yahoo Finance; Penmark calculations (11 monthly/print observations; SD bands omitted on short sample).
Source: Yahoo Finance peer snapshot (12 Nov 2026); NBIS uses spot EV ($47.17bn) over LTM revenue ($2,189.8m).
Six exhibits link Nebius’s ARR-driven rally to $88.23bn of capex, fading yields and a $96.91 target
NBIS rallied alongside reported AI Cloud ARR through mid-2026, yet equity value hinges on cash returns after an $88.23bn FY26E–28E build. Even with scenario skew lifting our $45.45 base DCF to $86.84 today ($96.91 in 12M, 76% from bull), shares imply $17.05m/MW long-run yields.
Source: Shareholder letters; Yahoo Finance quarter-end closes (3Q25–3Q26).
Source: Penmark Research model (annual revenue per billed IT MW).
Source: Company filings (A); Penmark Research model (E).
Source: Company filings (A); Penmark Research model (E).
Source: Penmark Research model (three operating debates plus $31.23/sh unexplained residual).
Source: Penmark Research model (rolls at 11.6% Ke to $96.91 in 12M).
Three operating debates explain $51.96/sh of the $83.19/sh gap between our $45.45 base DCF and $128.64
We sit below market-implied expectations on post-2029 GPU pricing, next-generation IT capex per MW, and gigawatt build pace. Sections 10–12 test the three quantified operating debates ($51.96/sh combined), while Section 15 examines the $31.23/sh unexplained residual alongside WACC and prepayment mechanics.
| Debate | Market view | Penmark view | Evidence | What changes our view | Resolves |
|---|---|---|---|---|---|
| 1. Long-run returns / MW (−$30.52/sh) |
Yield holds $15.75m/MW ($17.05m spot-implied); $276 mean target. | Fades to $15.0m/MW from FY32E (3.3-year IT payback as supply catches up). | CEO sees shortage easing in 1–2 yrs; 200+ neoclouds; 2024 H100 price reset; fixed-price MSFT/Meta caps. | Token Factory & software hold renewal yields ≥$16.5m/MW as supply normalises. | FY28–29 |
| 2. IT capex intensity (−$20.55/sh) |
−$1m/MW across FY27E–40E ($31m FY27–28E, $29m FY29E+). | $32m/MW in FY27–28E (Vera Rubin), $30m/MW FY29E+, +$12m owned DC. | FY26 capex raised to ~$27bn and FY27 guided $40–45bn; Q3 1y7m payback at >$25m ACV implies ~$30m+ IT. | Disclosed Vera Rubin cluster capex lands ≤$29m per deployed IT MW. | 1H 2027 |
| 3. Gigawatt build pace (−$0.89/sh) |
+150 MW/yr above our path across FY27–31E; 5 GW online in ~2–3.5 yrs. | 720 MW YE26; +1.0 GW/yr FY27–28E, tapering to +0.5 GW by FY31E (5 GW in FY32E). | NJ & 3P colocation sites slipped to 1H27; Dec ARR moved to Feb 27; 4 owned sites only start delivering in 2027. | Feb 27 ARR clears $7bn and owned sites deliver ≥1.2 GW on budget in FY27. | Feb–Dec 27 |
| 4. WACC & prepayments ($31.23 residual*) |
Spot implies 0.93 beta (8.03% WACC) or 7.5x FY27E EBITDA ex-capex. | 1.54 beta (10.03% WACC, 7.4% pre-tax cost of debt); prepayments net +$0.49bn. | NBIS fell 50.4% in 3M; 15–25% of 2027 capex uncovered by advances/IG debt; Oct 27 Citi bridge due. | IG contract-backed facilities fund >85% of 2027 capex below 6.5% pre-tax. | Oct 2027 |
Source: Company filings and Q3 2026 shareholder letter, S&P Global pre-print consensus, Penmark Research model. *After the three quantified operating debates ($51.96/sh), the remaining $31.23/sh gap to $128.64 is an unexplained residual reflecting discount-rate, multiple or interaction effects.
Source: Penmark Research model (each operating debate shifted individually on base DCF).
How the debates map to our thesis sections
Yield and capex per MW drive 98% of the quantified operating gap. Section 10 tests Debate 1 ($30.52/sh); Section 11 tests Debate 2 ($20.55/sh) and customer-funding cash flows; Section 12 explains why Debate 3 moves base DCF by only $0.89/sh.
Why faster megawatt additions barely move DCF
At base yields, new capacity earns roughly its 10.0% cost of capital. Megawatts built in FY27E–31E earn inside our explicit period, yet +150 MW/yr adds just $0.89/sh because $30–32m/MW IT capex, 65% owned shells and 6-year GPU replacement leave minimal excess return once yields fade to $15.0m/MW.
Source: Company filings and Q3 2026 shareholder letter; Penmark Research model.
Q3 advances topped 100% of deal capex, yet Q3 operating cash flow was −$273m
Nebius builds liquid-cooled supercomputer halls across America and Europe and rents them to tech giants and AI labs. Its twist is that Q3 contracts carried upfront advances above chip costs, yet invoicing lags and cluster spending still left operating cash flow negative in Q3.
The scene
On 12 November 2026, Nebius reported Q3 revenue up 571% to $980.8m — a third of it (~$325m) from non-recurring asset-light sales — and over $12bn of contracts signed since July, yet its shares fell 16.8%. Across colocation sites in New Jersey and Missouri and its owned hub in Mäntsälä, Finland, Q3 capex hit $7.3bn while Q3 operating cash flow swung to −$273.0m as invoiced prepayments sat in receivables. Born in 2024 after divesting Yandex’s Russian assets, Amsterdam-based Nebius serves Microsoft and Meta alongside Databricks, Datadog and Snowflake.
What it actually does
Running frontier AI takes thousands of liquid-cooled NVIDIA chips wired together as one machine. Nebius buys the GPUs ($30–32m per billed IT MW), designs custom racks, adds its Token Factory inference software, and increasingly builds its own halls at $12m per connected MW.
How the money comes in
AI Cloud generated $973.8m of Q3’s $980.8m revenue (99.3%), while autonomous-driving unit Avride and edtech arm TripleTen added $7.0m. Over 90% of Q3 deal value carried upfront prepayments, lifting deferred revenue to $9,409.2m. Per $1.00 of 9M 2026 revenue ($1,962.1m), $0.23 paid power and colocation and $0.35 funded cash opex, leaving +$0.42 Adjusted EBITDA; deducting $0.10 of share-based pay, $0.03 of one-offs and $0.42 of D&A left a $0.13 operating loss ($253.5m).
Where the story turns
Q3 GPU pricing rose over 20% sequentially above $25m/MW ACV, and Token Factory bookings jumped 17-fold. Yet colocation delays in New Jersey pushed Nebius’s >$7bn ARR goal from December 2026 to February 2027 and lifted 2027 capex guidance to $40–45bn as 65% of new power shifts to owned builds.
Why it works, and why it might not
Multi-year deals and $38.7bn of backlog lock utilization near 92%. What breaks the loop is supply: CEO Arkady Volozh expects the GPU shortage to ease in one to two years, compressing post-2029 yields toward $15.0m/MW (−$30.52/sh in Debate 1).
AI Cloud generates 99.3% of Q3 revenue ($973.8m), but D&A ($826.4m) keeps 9M operating income negative
Nebius converts grid megawatts into billable GPU compute at a 0.80x IT-to-power ratio and 92% utilization. While Q3 AI Cloud Adjusted EBITDA margin reached 53.2% (lifted by ~$325m of one-off asset-light sales), heavy depreciation and upfront capex create a wide wedge between EBITDA, GAAP EBIT and cash flow.
Source: FY2025 Form 20-F; Q3 2026 earnings release and shareholder letter; Penmark Research model. Non-consolidated equity stakes in ClickHouse and Toloka (deconsolidated Q2 2025) are carried in the valuation bridge ($1,603.9m investments at 30 Sep 2026).
Source: Q3 2026 earnings release (9M 2026 continuing operations).
Segment vs group margin distinction
Group Adjusted EBITDA margin (47.5% in 3Q26; 42.4% in 9M26) trails AI Cloud segment margin (53.2% in 3Q26) by ~5–6pp. Avride, TripleTen and unallocated corporate costs lost −$52.3m in 3Q26 (−$123.9m in FY25A). Moreover, Q3 AI Cloud margin was flattered by ~$325m of non-recurring asset-light sales; our 4Q26E AI Cloud margin steps back to 42.0%.
Fixed vs variable cost shift on owned sites
Shifting 65% of 2027+ builds to owned sites swaps variable colocation rent (~$2.2m/MW illustrative industry rent) for upfront $12m/MW capex. That transition lifts base AI Cloud Adjusted EBITDA margin from 50.0% in FY27E to 58.0% in FY28E and 61.0% from FY29E, while pushing fixed D&A to $10.52bn by FY28E.
Every 100 MW of connected power yields 73.6 MW of billed IT load and $1.18bn of FY27E revenue
Nebius’s capacity-and-utilization revenue equation multiplies average billed IT megawatts by realized annual revenue per MW. Separating contracted power, connected grid power, active IT power and billed load is essential because marketing headlines quote multi-GW contracted pipelines while GAAP revenue accrues only as energized racks enter customer billing.
Source: Penmark Research model (529.92 MW YE26E + 35% × 736.0 MW FY27E additions = 787.52 MW average billed load). Highlighted nodes correspond to confirmed Section 9 drivers.
| Unit metric (per 1.0 MW billed IT load unless noted) | 3Q26 New Deals | FY27E Base | FY28E Peak | FY32E+ Steady |
|---|---|---|---|---|
| Connected grid power required (0.80x IT-to-power & 92% utilization) | 1.36 MW | 1.36 MW | 1.36 MW | 1.36 MW |
| Annualized revenue yield per billed IT MW ($m/MW) | >$25.00m | $16.00m | $17.50m | $15.00m |
| AI Cloud Adjusted EBITDA margin (%) | 53.2% (3Q seg.) | 50.0% | 58.0% | 61.0% |
| AI Cloud Adjusted EBITDA per billed IT MW ($m/MW) | ~$13.30m | $8.00m | $10.15m | $9.15m |
| IT cluster capex per billed IT MW (5-yr depreciation, 6-yr replacement) | ~$32.00m | $32.00m | $32.00m | $30.00m |
| Blended owned data-centre shell & power capex per billed IT MW (65% owned) | Colocation-heavy | $10.60m | $10.60m | $10.60m |
| Total initial capital per billed IT MW ($m/MW, pre-prepayment) | ~$32–43m | $42.60m | $42.60m | $40.60m |
| IT capex ÷ EBITDA per MW | 1y 7m (mgmt Q3) | 4.0 yrs | 3.15 yrs | 3.3 yrs |
Source: Q3 2026 shareholder letter; Penmark Research model. Owned data-centre capex of $12.0m per connected MW × 65% owned share ÷ (0.80 × 0.92) = $10.60m per billed IT MW. Management published a 1y 7m payback on Q3 deals; our model fleet payback (IT capex ÷ EBITDA per MW) reaches 4.0 years in FY27E, 3.15 years in FY28E and 3.3 years from FY32E.
Four power definitions — never mix them
Contracted power (>3 GW pipeline) is signed utility and land rights; connected power (407.6 MW at 3Q26; 720 MW YE26E) is grid-energized capacity. Active IT power applies the 0.80x IT-to-power ratio, and billed IT load (300.0 MW at 3Q26; 529.9 MW YE26E) applies 92% utilization plus a ~2-month commissioning lag from connection to billing (earning 35% of a year on uniform annual additions).
Reported ARR vs GAAP revenue
September 2026 AI Cloud ARR ($6.4bn) annualizes a single end-quarter month (and Q3 included ~$325m one-off asset-light sales), whereas 9M26 GAAP revenue was $1,962.1m. Management pushed its >$7.0bn ARR goal to February 2027 and stops reporting ARR from 2027; we therefore test February execution against 4Q26E revenue ($1,169.3m) and year-end 2026 connected power (720 MW).
Steady-state 1 MW leaves $3.62m pre-tax profit after $5.00m GPU replacement and $0.53m shell depreciation
Nebius’s 2025–26 ramp relied heavily on third-party colocation sites, where utility and developer bottlenecks in New Jersey pushed late-2026 handovers into 1H 2027. Management is responding by building 65% of 2027+ capacity on owned greenfield land across the US and Europe ($27.31bn FY26E and $31.35bn FY27E model cash capex), financed through upfront customer advances (+$9.33bn net FY26E) and contract-backed debt.
Source: Penmark Research model (FY32E+ per 1.0 MW billed IT load: $15.00m revenue less $5.85m cash costs = $9.15m EBITDA; less $5.00m GPU replacement [$30m ÷ 6 yrs] and $0.53m owned shell depreciation [$12m × 65% ÷ (0.80 × 0.92) ÷ 20 yrs] = $3.62m pre-tax profit).
Source: Q3 2026 shareholder letter and earnings call; FY2025 Form 20-F; Penmark Research model (New Jersey colocation site transitioning power toward Bloom Energy fuel cells).
Hyperscaler anchors dominate RPO
Microsoft and Meta anchor Nebius’s $38.7bn RPO backlog (including Meta’s framework of up to $15bn). While enterprise AI logos (Databricks, Datadog, Snowflake) and Token Factory broaden the tail, hyperscalers simultaneously build internal clusters and hold renewal bargaining power after 2029.
Single-vendor silicon & grid risk
Compute capex ($30–32m/MW) is concentrated in NVIDIA architectures (Hopper, Blackwell, Vera Rubin). On the infrastructure side, Q3 colocation delays in New Jersey showed third-party landlords can slip schedules, driving Nebius toward higher-capex owned sites across the US and Finland.
Why model cash differs from guide
Management guides that gross prepayments ($10bn in 2026; $17–20bn in 2027) plus IG contract-backed debt cover 75–85% of 2027 capex. Our cash-flow statement models net deferred-revenue change (+$9.33bn FY26E, +$9.00bn FY27E, +$3.00bn FY28E) as prior advances amortize into revenue.
Specialist neocloud revenue is estimated to scale from ~$48bn to ~$300bn by 2030E, while silicon captures upstream rent
Specialist GPU cloud infrastructure sits at the fastest-growing layer of enterprise computing, fuelled by frontier model training, hyperscaler overflow and enterprise inference. However, rapid demand growth does not guarantee excess equity returns for neoclouds caught between concentrated chip suppliers upstream and self-building hyperscaler customers downstream.
Source: Independent specialist neocloud category estimates (~$48bn mid-2026 annualized run-rate, ~$300bn 2030E; analyst estimates, not audited GAAP revenue); CoreWeave CY27E consensus; Q3 2026 shareholder letter; Penmark Research model.
Source: Company filings; Pantera industry framework; Penmark Research model.
Bottom-up neocloud sizing & cycle position
Bottom-up, each 1 GW of connected AI data-centre power (736 MW billed IT load at 0.80x IT-to-power ratio and 92% utilization) generates $11.0–12.9bn of annual revenue at $15.0–17.5m/MW. Independent analyst commentary places specialist neocloud annualized category revenue near ~$48bn in mid-2026 and ~$300bn by 2030E (a third-party analyst estimate, not audited GAAP revenue).
Today the market sits in physical power scarcity: multi-year utility interconnection queues lifted Q3 GPU-hour pricing over 20% sequentially above $25m/MW ACV.
Value-chain squeeze & substitution threats
Rent extraction sits at upstream silicon ($30–32m/MW IT capex) and near-term energized power rather than bare-metal GPU rental. Three forces cap post-2029 neocloud pricing:
Supply catch-up as 200+ neoclouds and REITs energize sites over the next 1–2 years; hyperscaler ASICs (AWS Trainium, Google TPU, Meta MTIA) substituting rented GPUs on internal inference; and model efficiency pushing older chip cohorts onto lower-priced tiers.
Near-term tailwind: 2026–28 transformer and utility delays keep energized halls scarce, supporting $16.0–17.5m/MW blended yields.
Medium-term reset: When 3–5 year Q3 contracts roll after 2029, hyperscaler anchors compare neocloud quotes against internal halls.
Token Factory test: Managed inference must scale from multi-hundred-million run-rate into multi-billion revenue to defend yields.
Nebius out-engineers smaller neoclouds (813 R&D staff), but trails CoreWeave on scale and hyperscalers on funding cost
Inherited hyperscale engineering DNA gives Nebius technical credibility in custom liquid-cooled racks and managed inference software. Yet CoreWeave already operates more than 3x Nebius’s current revenue base at 4.37x CY27E EV/Adj. EBITDA, while hyperscalers enjoy a 250–350bp lower cost of capital on identical NVIDIA silicon.
Source: Company filings; Penmark Research assessment.
| Cohort / Peer | Disclosed Scale & Power Basis | Backlog / RPO & Pricing Model | CY27E Valuation / Economics |
|---|---|---|---|
| Nebius (NBIS) | 1H26 AI Cloud rev. $964.6m (3Q26 $973.8m; Sep ARR $6.4bn); 407.6 MW 3Q26 connected power | $38.7bn RPO (3Q26); 1–5 yr reserved (>90% prepaid in Q3) + Token Factory API | 7.68x EV / FY27E Adj. EBITDA ($47.17bn EV; $12.63bn rev.; 48.6% margin) |
| CoreWeave (CRWV) | 1H26 GAAP rev. $4.65bn (2Q26 $2.58bn); ~1.5 GW active power | ~$104bn revenue backlog; multi-year take-or-pay hyperscaler & lab contracts | 4.37x EV / CY27E Adj. EBITDA ($72.90bn EV; $26.50bn cons. rev.; $16.70bn EBITDA) |
| Power-first / Miner Neoclouds (IREN, APLD, WULF, Crusoe) | Multi-hundred-MW to GW land/power banks; mix of powered shells, HPC leases & bare-metal GPUs | Mix of 10–15 yr colocation leases and shorter bare-metal GPU rentals | 12.4x–53.4x trailing EV/Sales; mostly negative trailing EBITDA; sparse CY27 consensus |
| Hyperscalers & OCI (MSFT, AMZN, GOOGL, ORCL) | Multi-GW global fleet; both Nebius’s largest customers and primary long-term substitutes | Bundled enterprise cloud + proprietary ASICs (TPU, Trainium) + NVIDIA clusters | ORCL 7.27x trailing EV/Sales at 48.0% EBITDA margin; IG debt ~250–350bp cheaper |
Source: Penmark Research model; CoreWeave SEC filings; Yahoo Finance (12 Nov 2026); company disclosures. Market shares cannot be calculated reliably because peers report incompatible metrics (GAAP revenue vs end-month ARR; active IT MW vs connected MW vs contracted pipeline).
Where Nebius wins today
Full-stack engineering differentiates Nebius from crypto-miner conversions. Its 813 product-development staff (up from 653 in 2024) design custom racks and orchestration software, enabling rapid cluster bring-up and driving Token Factory contract value up 17x q/q in Q3 2026.
During contract life (1–5 years), upfront prepayments and data-gravity create switching costs across $38.7bn of RPO.
Why moats narrow at contract renewal
Hardware parity and funding cost dominate multi-year renewals. Every tier-1 neocloud deploys identical NVIDIA Blackwell and Rubin systems, while hyperscalers fund builds at investment-grade coupons below the 7.4% pre-tax cost of new debt in our 10.03% WACC.
Yet NBIS trades at a 76% EV/EBITDA premium to CoreWeave (7.68x vs 4.37x CY27E), pricing sustained scarcity rents where its closest scaled peer trades on normalized multiples.
Dual-class Class B shares give CEO Volozh 51.6% voting control on a 9.8% economic stake
Founder-CEO Arkady Volozh brings a three-decade hyperscale engineering record, holding 28.78m shares (~9.8% of 292.90m 30 September 2026 shares). Dual-class 10-vote Class B shares gave him 51.6% voting power at 31 March 2026 over an $88.23bn FY26E–28E capex build where Class A holders absorb issuance and execution risk.
| Governance item | Disclosure (FY2025 20-F & 3Q26) |
|---|---|
| Share classes | Class A: 1 vote/sh; Class B: 10 votes/sh |
| CEO voting | 51.6% voting power at 31 Mar 2026 (28.66m B + 0.12m A) |
| CEO economic | 28.78m shs = ~9.8% of 292.90m 30 Sep 2026 shares |
| Nasdaq status | Controlled-company exemptions invoked |
| Board & cmtees | Maj. independent; Audit/Comp 100% ind.; Nom. incl. CEO |
| Exec. pay (FY25) | $5.9m cash pay; 6.35m options + 226.3k RSUs granted |
Source: FY2025 Form 20-F (votes at 31 Mar 2026); Q3 2026 release (292.90m shares at 30 Sep 2026).
| Metric | Prior guide | Q3 26 update & status |
|---|---|---|
| AI Cloud ARR | >$7bn Dec 26 | Pushed to Feb 2027 (3P colo delays) |
| FY26 revenue | $3.0–3.4bn | Low end ($3.0–3.2bn read; $3.13bn est.) |
| FY26 EBITDA % | ~40% margin | Reaffirmed ~40% (40.5% Penmark) |
| FY26 capex | Stepped up | ~$27bn guided ($27.31bn cash model) |
| FY27 revenue | Initial frame | $12–13bn ($12.63bn Penmark) |
| FY27 capex | Initial frame | $40–45bn ($31.35bn base model) |
Source: Q3 2026 shareholder letter; Penmark Research model.
Source: FY2025 Form 20-F (votes, 31 Mar 2026); Q3 2026 filings (shares, 30 Sep 2026).
Capital allocation, share bases & stakes
Zero payouts, 100% reinvestment: Since October 2024, Nebius has paid $0 dividends and raised $10.70bn net debt, $7.83bn customer advances and $5.92bn equity in 9M26. Market value uses 313.97m shares (292.90m 30 Sep shares + NVIDIA pre-funded warrants); per-share values use 319.71m diluted shares (adding unvested RSUs; convertibles count only when in the money).
Portfolio & non-core stakes: Toloka was deconsolidated in Q2 2025; retained equity stakes in ClickHouse and Toloka carry at $1,603.9m after $780.6m of 9M26 revaluation gains, alongside consolidated units Avride ($100m SAFE) and TripleTen.
Source: Q3 2026 earnings release (9M 2026 cash flows, $3.68bn Dec-25 to $7.43bn Sep-26 unrestricted cash); Penmark Research model.
Deal pricing, capex funding and energized MW move NBIS (+34% Q2 / −50% since Aug) — not GAAP EPS
NBIS trades on forward capacity economics and credit availability rather than reported earnings. Across eight earnings reactions, 12-month return decomposition (+36.3% vs +8.4% Nasdaq; EV/ARR down from 66.8x to 7.4x; −50.4% three-month slide), an 85-week factor regression (R² = 0.505; 49.5% idiosyncratic) and our review of 95 call questions (not company-reported), four drivers pass two or more tests.
| Confirmed driver | Tests passed | Observed market reaction | Model value vs market-implied | Debate & next watch |
|---|---|---|---|---|
| 1. GPU yield & ARR | 3 / 4 (Event, Return, Q&A) | +34.1% on Q2’26 deal surge; −16.8% on Q3’26 ARR shift; 44.2% of call Q&A | Price implies $17.05m/MW (value $128.64); the $15.75m market view is worth $75.97 vs $45.45 base (−$30.52/sh) | Debate 1 (Sec. 10) · Feb 27 ARR >$7.0bn goal; 4Q26E rev. $1,169m |
| 2. Capex & funding mix | 3 / 4 (Event, Factor, Q&A) | −16.8% on $7.3bn Q3 capex & $40–45bn FY27 guide; 23.2% of call Q&A | −$1m/MW FY27–40E capex view = $66.00 vs $45.45 base (−$20.55/sh); +150bp debt cost = −$10.45/sh | Debate 2 (Sec. 11) · FY26 cash capex ($27.31bn est. vs ~$27bn guide) |
| 3. Connected MW pace | 2 / 4 (Event, Q&A) | NJ colocation delay drove Q3 sell-off despite +571% revenue growth; 18.9% of Q&A | +150 MW/yr lifts DCF by +$0.89/sh ($46.33 vs $45.45 base); ±300 MW/yr = $43.58–$47.22 | Debate 3 (Sec. 12) · YE26 connected MW (720 MW est. vs 800–1,000 MW guide) |
| 4. Neocloud & AI beta | 2 / 4 (Return, Factor) | CoreWeave (20.6%) + Nasdaq (21.6%) + SMH (7.9%) explain 50.1% of 85-week variance | Price implies 0.93 beta (8.03% WACC) vs 1.54 beta (10.03% WACC); $31.23/sh unexplained residual | Debate 4 (Sec. 15) · CoreWeave CY27E EV/EBITDA (4.37x vs NBIS 7.68x) |
Source: Yahoo Finance, company filings and earnings call transcripts, Penmark Research model.
Source: Yahoo Finance single-day earnings reactions (Q4 2024 to Q3 2026).
Why NBIS halved (−50.4%) over the past three months
From $259.20 on 12 August to $128.64 on 12 November, NBIS fell 50.4% (−55.1% from its $286.69 June close high) in two distinct legs. Between 6 October and 6 November, shares slid 47.7% alongside CoreWeave (−49.9%) as neocloud credit stress, Microsoft’s late-October capex timing miss, and 11 November reports that Microsoft may scale back neocloud partnerships hit leveraged GPU builders.
On 12 November, shares dropped another 16.8% (from $154.69): despite beating Q3 revenue consensus by 8% ($980.8m, lifted by ~$325m one-off sales), management trimmed FY26 revenue to the low end ($3.0–3.2bn), pushed its >$7bn ARR goal to February on colocation delays, raised capex (~$27bn FY26; $40–45bn FY27) and reported into AI-tech weakness on NVIDIA CEO resignation news.
Dropped candidates & empirical test limits
GAAP EPS failed 0 of 4 tests: Q3’26 diluted EPS of −$0.13 beat −$1.24 consensus by 89.5% on ClickHouse revaluation gains yet the stock fell 16.8%, confirming investors look through non-operating mark-to-market swings.
Spot 10Y yields & non-core units failed: Weekly US 10Y yield changes failed 0/4 tests (0.0–0.4% variance share, though credit spreads govern our 7.4% pre-tax cost of debt), while Avride and TripleTen passed 1/4 (early Q&A only; 0.7% of Q3 revenue). Sample spans NBIS’s post-October 2024 relisting history (8 prints, 5 transcripts, 85 overlapping CoreWeave weeks).
Post-2029 GPU rental yields fade from >$25m/MW scarcity peaks to $15.0m/MW, cutting $30.52/sh vs market views
2 · What bulls get wrong & counter-evidence
At $128.64, reverse DCF shows the market prices either $17.05m/MW across FY32E–40E (holding FY27–31E at base) or $23.71m/MW across FY27–31E. Bulls extrapolate Q3’s +20% q/q GPU-hour price jump and >$25m/MW ACV across a 15-year asset life.
Strongest counter-evidence: Token Factory contract value jumped 17-fold q/q in Q3, and higher-margin inference software could sustain $17.0m/MW long-run yield — supporting our $265.54/sh bull case (25% weight) and explaining why conviction is Low (price-implied $17.05m/MW sits within 15% of our $15.0m/MW base).
Source: Penmark Research model.
| Evidence source | Empirical finding | Implication for yield & value |
|---|---|---|
| 1. Management call & unit payback disclosure | CEO Volozh noted on the Q3 2026 call that acute compute scarcity should ease in 1–2 years. Q3 deals pay back in 1y7m (management); our FY32E steady state pays back in 3.3 years, about twice as long, as supply catches up. | Confirms >$25m/MW Q3 signings are cyclical scarcity rents rather than perpetual renewal rates. |
| 2. CoreWeave peer multiple benchmark | CoreWeave operates ~1.5 GW active power and ~$104bn backlog on the same NVIDIA silicon, yet trades at 4.37x CY27E EV/Adj. EBITDA vs NBIS at 7.68x. | Applying 4.37x to NBIS FY27E EBITDA ($6.14bn) yields $68.68/sh (−46.6% vs $128.64). |
| 3. Competitive supply & steady-state economics | Over 200 neoclouds plus hyperscaler self-builds compete for renewals as GPU cohorts refresh every 6 years, compressing steady-state unit profit to $3.62m/MW pre-tax after $5.00m GPU replacement and $0.53m shell D&A. | 4 · What it is worth: −$30.52/sh ($45.45 base vs $75.97 at $15.75m); −$1.5m/MW wipes equity to $0.00. |
Source: Q3 2026 shareholder letter and earnings transcript, CoreWeave filings, Penmark Research model.
$88.23bn of FY26E–28E capex keeps core FCF negative until FY30E; advances save interest, not principal
2 · Why single-year multiples misread advances
Single-year EV/EBITDA views overstate equity value when EV excludes deferred revenue ($9.41bn) while forward EBITDA includes amortization of cash already collected and spent on prior capex.
Our DCF separates operating cash from financing: Deducting $9,409.2m deferred revenue alongside +$3,713.3m pre-billed receivables leaves $25.83/sh before funding credit; adding +$6,190.6m PV of interest saved (+$19.62/sh) lifts base DCF to $45.45/sh — still $20.55/sh below a −$1m/MW FY27–40E capex view ($66.00/sh).
Source: Company filings (A); Penmark Research model (E, after interest; DCF unlevered FCF ex-advances is −$26.00bn in FY27E).
| Evidence source | Empirical finding | Cash & valuation impact |
|---|---|---|
| 1. 9M 2026 cash-flow & liability filings | 9M26 GAAP OCF (+$4,231.1m) included +$7.83bn net customer advances, lifting deferred revenue to $9,409.2m; core OCF ex-advances was −$3,597.9m against $15,430.7m PP&E capex. | 9M26 core FCF burned −$19,028.6m (−$11,199.6m incl. advances). |
| 2. Q3 2026 FY27 capex escalation | Management guided FY27 capex to $40–45bn (up from ~$27bn FY26 guide) as builds shift to owned shells; even our paced 1.0 GW base models $31.35bn FY27E and $29.57bn FY28E capex. | If FY27 capex reaches $42.5bn without matching yield lift, near-term debt or dilution rises further. |
| 3. Leverage build & 6-yr GPU replacement | Gross debt rose to $14,190.8m at 3Q26, reaches $59,001.4m in FY29E (2.55x Net debt/EBITDA) and peaks at $61,761.4m in FY33E; servers depreciate over 5 yrs and replace every 6 yrs. | 4 · What it is worth: −$20.55/sh vs −$1m/MW FY27–40E capex view; +150bp debt cost cuts −$10.45/sh. |
Source: Q3 2026 earnings release and shareholder letter, Penmark Research model. By FY32E, FCF incl. advances (−$2.3bn) trails FCF ex-advances (+$0.7bn) as prior prepayments amortize.
Adding ±300 MW/yr shifts base DCF by just +$1.77 / −$1.87/sh ($43.58–$47.22) unless yields stay elevated
Source: Penmark Research model (base reaches 7.09 GW by FY40E).
Source: Penmark Research model. At −$3.0m/MW FY27–31E yield, faster builds cut value from $17.5 to $4.6/sh.
| Evidence source | Empirical finding | Model & value effect |
|---|---|---|
| 1. Q3 2026 site handover delays | Third-party colocation delays (including New Jersey) pushed >$7bn ARR from Dec 2026 to Feb 2027 and kept 3Q26 connected power at 407.6 MW vs 800–1,000 MW YE26 target. | We model 720 MW YE26E before +1,000 MW/yr in FY27E–28E (below >1 GW/yr ambition). |
| 2. Grid & power-equipment friction | New Jersey required transitioning power toward Bloom Energy fuel cells; multi-year transformer and grid-queue bottlenecks across US and EU hubs slow greenfield energization. | Explains why base FY27E capex ($31.35bn) paces below management’s $40–45bn guide. |
| 3. Corporate-finance math | In Exhibit 48, moving from 1.0 GW/yr by ±300 MW/yr at base yields shifts DCF by +$1.77 / −$1.87/sh ($47.22 / $43.58); at −$3.0m/MW yield, faster builds cut DCF from $17.5 to $4.6/sh. | 4 · What it is worth: −$0.89/sh Debate 3 gap ($45.45 vs $46.33); volume amplifies spread to WACC. |
Source: Q3 2026 shareholder letter and earnings transcript, Penmark Research model.
9M26 GAAP net income (+$393.3m) and OCF (+$4.23bn) reflect ClickHouse gains and advances, masking −$19.03bn core FCF
Nebius’s continuing-operations revenue scaled 54x from $9.8m in FY23A to $529.8m in FY25A and $1,962.1m in 9M26A (+549% y/y). Yet five accounting and cash-flow features — lengthened server depreciation, non-operating equity revaluations, excluded SBC, one-off asset-light sales and upfront customer advances — create a wide gap between headline figures and economic cash flow.
Source: Q3 2026 earnings release (9M26 operating loss −$253.5m); Penmark Research model.
Five reporting choices investors must adjust
1) Server life & replacement: From January 2026 Nebius depreciates servers over 5 years (previously 4). We depreciate over 5 years and replace GPUs after 6, in line with hyperscaler practice; a shorter replacement cycle would lower our value.
2) Non-operating ClickHouse gains: Fair-value remeasurements (+$598.9m FY25; +$780.6m 9M26 on Series D) turned GAAP EBIT losses into positive net income.
3) SBC & capitalized interest: Adjusted EBITDA excludes SBC ($83.2m FY25; $196.5m 9M26 incl. Eigen AI); FY25 capitalized $52.6m of interest. Our DCF expenses SBC.
4) One-off 3Q26 asset-light sales: ~$325m of non-recurring asset-light sales lifted 3Q26 AI Cloud EBITDA margin to 53.2% (vs 47.5% group margin).
5) Upfront customer advances: +$7,829.0m of 9M26 advances drove all +$4,231.1m GAAP OCF while core OCF ex-advances burned −$3,597.9m.
| Continuing operations ($m unless noted; Toloka deconsolidated Q2 25) | FY23A | FY24A | FY25A | 9M26A | 3Q26A |
|---|---|---|---|---|---|
| Consolidated revenue | 9.8 | 91.5 | 529.8 | 1,962.1 | 980.8 |
| Revenue growth (% y/y) | n/a | +833.7% | +479.0% | +549.5% | +571.3% |
| AI Cloud segment revenue (Avride + TripleTen less elim. = remainder) | 9.6 | 68.3 | 480.3 | n/d | 973.8 |
| GAAP operating income / (loss) | (285.7) | (399.6) | (611.7) | (253.5) | 50.4 |
| (+) Depreciation & amortization (4y server life FY23–25; 5y from Jan 26) | 29.3 | 77.1 | 417.9 | 826.4 | 354.7 |
| (+) Share-based compensation (incl. Eigen AI post-combination in 9M26) | 28.8 | 54.5 | 83.2 | 196.5 | 58.7 |
| (+) Transaction, one-off & other reconciling items | (13.1) | 41.7 | 45.7 | 62.1 | 2.0 |
| Group Adjusted EBITDA (reported) | (240.7) | (226.3) | (64.9) | 831.5 | 465.8 |
| Group Adjusted EBITDA margin (%) / AI Cloud segment margin (%) | nm | nm | −12.2% / +12.3% | 42.4% / n/d | 47.5% / 53.2% |
| ClickHouse & equity-security revaluation gains (below EBIT) | 0.0 | 0.0 | 598.9 | 780.6 | n/d |
| GAAP net income / (loss), continuing operations | (299.0) | (352.0) | 9.8 | 393.3 | (37.5) |
| GAAP operating cash flow (incl. +$982.5m FY25 & +$7,829.0m 9M26 advances) | (222.0) | (269.9) | 401.9 | 4,231.1 | (273.0) |
| (−) Purchases of PP&E and intangibles (cash capex) | (82.9) | (807.5) | (4,066.0) | (15,430.7) | (7,300.4) |
| Core free cash flow excl. customer prepayments | (304.9) | (1,077.4) | (4,646.6) | (19,028.6) | n/d |
Source: FY2025 Form 20-F, Q3 2026 earnings release, Penmark Research model (Toloka deconsolidated Q2 2025).
Revenue reaches $12.63bn in FY27E (+3.2% vs consensus) and $54.81bn in FY32E on $31.35bn FY27E capex
Because NBIS burns core FCF through FY29E and enters a 6-year GPU replacement cycle in FY31E, we model 15 explicit years to FY40E on the Section 5 driver tree (Exhibit 29). FY26E–27E revenue fits guidance, while FY27E capex paces at 1.0 GW of energized additions.
Source: Company filings (A); Penmark Research model (E, FY25A–FY32E).
Source: Q3 2026 shareholder letter; S&P Global and FactSet consensus; Penmark Research model.
FCF sensitivity to Section 9 drivers: In FY27E, ±$1.0m/MW yield shifts pre-tax cash flow by ±$788m on 787.5 MW average load, while ±$1.0m/MW IT capex shifts FY27E cash capex by ±$736m.
Volume, yield & fixed/variable margin path
Volume, yield & margins: Connected power rises from 407.6 MW (3Q26) to 720 MW (YE26E), adds +1,000 MW/yr in FY27E–28E, +900 MW (FY29E) and +700 MW (FY30E), reaching 7.09 GW in FY40E (35% year-one weight). Yield peaks at $17.5m/MW (FY28E) and settles at $15.0m/MW (FY32E+); AI Cloud EBITDA margin rises from 50.0% (FY27E) to 61.0% (FY29E+) as 65% owned shells replace colocation rent, before 6-yr GPU replacement starts in FY31E.
Explaining key gaps vs guidance & consensus
FY27E capex ($31.35bn vs $40–45bn guide, −26.2%) & advances: At ~$31.4m blended capex per connected MW, $40–45bn buys 1.3–1.4 GW (our Bull case: +1,300 MW); after 2026 colocation delays, Base builds 1,000 MW in FY27E and reaches 5 GW connected in FY32E. Management’s $10bn (2026) and $17–20bn (2027) prepayment figures are gross; we model net deferred-revenue changes (+$9.33bn FY26E, +$9.00bn FY27E).
| Line item ($m unless noted) | FY25A | FY26E | FY26 Guide / Cons | FY27E | FY27 Guide / Cons | FY28E | FY30E | FY32E |
|---|---|---|---|---|---|---|---|---|
| Connected power, YE (MW) | n/d | 720 | 0.8–1.0 GW (sup.) | 1,720 | +1,000+ MW | 2,720 | 4,320 | 5,220 |
| Billed IT load, YE (MW) | 104.2 | 529.9 | Feb 27 ARR >$7bn | 1,265.9 | n/f* | 2,001.9 | 3,179.5 | 3,841.9 |
| Revenue yield ($m / billed MW) | 12.0 | 12.6 (4Q) | ~$12m base book | 16.0 | >$25m Q3 deals | 17.5 | 16.0 | 15.0 |
| Consolidated revenue | 529.8 | 3,131.4 | $3.0–3.2bn* / $3,370 | 12,631.1 | $12–13bn / $12,240 | 26,695.5 | 45,555.2 | 54,808.0 |
| Group Adjusted EBITDA | (64.9) | 1,267.7 | ~40% mgn / $1,348 | 6,140.2 | n/g / n/a | 15,323.7 | 27,663.7 | 33,342.6 |
| Group Adj. EBITDA margin (%) | −12.2% | 40.5% | 38–42% / 40.0% | 48.6% | AI Cloud: 50.0% | 57.4% | 60.7% | 60.8% |
| GAAP EBIT (after SBC & D&A) | (611.7) | (396.9) | n/g | 419.8 | 4Q27 EBT b/even | 4,219.5 | 8,041.6 | 11,432.1 |
| GAAP diluted EPS ($/sh) | n/a | 0.40 | Cons: −$2.10 | (1.68) | Cons: −$3.60 | 6.01 | 11.44 | 19.41 |
| Cash capex (outflow) | (4,066.0) | (27,305.8) | ~$27bn ($26–28bn) | (31,352.0) | $40–45bn guide | (29,568.0) | (19,775.6) | (27,252.4) |
| Deferred revenue, YE ($m) | 1,577.5 | 10,909.2 | $10bn gross 26 | 19,909.2 | $17–20bn gross 27 | 22,909.2 | 16,909.2 | 9,909.2 |
| Core FCF ex-advances | (4,646.6) | (30,588.4) | n/g | (26,642.2) | ±$1m yield = ±$788m | (17,246.6) | 3,070.8 | 689.7 |
| Net debt excl. deferred rev. | 449.6 | 16,825.0 | $2.0bn min. cash | 34,467.2 | 5.61x ND/EBITDA | 48,713.8 | 57,930.6 | 58,782.4 |
Source: Company filings, pre-print consensus snapshot, Penmark Research model (FY24A revenue $91.5m, Adj. EBITDA −$226.3m, core FCF −$1,077.4m; see page 25). *n/f: management no longer reports ARR from 2027; FY26 revenue guide is low end ($3.0–3.2bn) of $3.0–3.4bn.
15-year unlevered DCF yields $45.45/sh base value; EV/FY27E EBITDA cross-check gives $68.68/sh
A 15-year unlevered DCF to FY40E is our primary method because negative FY27E GAAP EPS (−$1.68), FY26E capex at 8.7x sales and negative core FCF through FY29E make single-year P/E or EV/Sales unsuitable. Valued FCF excludes customer advances; at 30 Sep 2026 we deduct $9,409.2m deferred revenue, add $3,713.3m pre-billed receivables and credit +$6,190.6m interest-free funding PV (+$0.49bn net).
Source: Penmark Research model (30 Sep 2026 balance sheet; colocation lease liabilities of $1,768.8m are excluded as lease expense sits inside Adjusted EBITDA).
Source: Penmark Research model; Yahoo Finance 52-wk intraday range ($73.52–$299.86); S&P Global pre-print targets ($84–$410, mean $276.20). EV/CY27E EBITDA bar shows a ±10% illustrative valuation band ($61.89–$75.64) around the $68.68 CoreWeave 4.37x peer cross-check.
| Valuation parameter / equity bridge line | Primary: 15y DCF | Cross-check: EV/EBITDA |
|---|---|---|
| Discount rate / peer multiple (Ke 11.60% at 1.54β, pre-tax Kd 7.4%, after-tax 5.55% at 26% D) | 10.03% WACC | 4.37x CY27E |
| PV explicit unlevered FCF (4Q26E–FY40E) | (14,146.8) | n/a |
| PV terminal value (g = 3.0%; normalized FY41E NOPAT $13,025.1m) | 33,270.3 | n/a |
| Enterprise value ($m) | 19,123.5 | 26,805.1 |
| (+) Cash ($7,425.6m), investments ($1,603.9m) & pre-billed AR (+$3,713.3m) | 12,742.8 | 12,742.8 |
| (−) Deferred revenue (−$9,409.2m) + customer-funding PV (+$6,190.6m) | (3,218.6) | (3,218.6) |
| (−) Secured ABS (−$753.5m), all converts (−$13,450m), SAFE & NCI (−$101m) | (14,304.5) | (14,304.5) |
| Equity value before conversions ($m) | 14,343.2 | 22,024.9 |
| In-the-money converts / diluted shares / price-date roll | 0 conv. / 319.71m / 1.0130x | +$200m Jun-25 / 323.59m / none |
| Value per share at 12 Nov 2026 ($/sh) | $45.45 | $68.68 |
Source: Penmark Research model. DCF divides $14,343.2m by 319.71m shares ($44.86/sh at 30 Sep 2026) and rolls 0.118 yrs at 11.60% Ke (1.0130x) to $45.45/sh; EV/EBITDA adds back in-the-money Jun-25 converts ($200.0m at $51.45 strike = 3.89m shs; $22,224.9m ÷ 323.59m shs, no roll) for $68.68/sh.
Justifying $128.64 requires $17.05m/MW terminal yield (+13.7%) or an 8.03% WACC (0.93x beta)
Our terminal assumption (post-FY40E returns on new capital equal WACC) keeps base DCF within $44.42–$46.47/sh across 2.0%–4.0% terminal growth. Long-run GPU yield and replacement IT capex dominate equity sensitivity, while the +51.1% cross-check gap ($68.68 vs $45.45/sh) reflects FY27E–29E capex burn.
Source: Penmark Research model. Closed-form grid shifts WACC by ±1–2pp while holding base Ke roll and 0 convertible conversions ($127.2 at 8.03% / 3.0% g); the full beta solve (0.93x β, 8.03% WACC) updates Ke and if-converted dilution to reproduce $128.64.
Source: Penmark Research model (bar labels rounded to $0.1).
Why the EV/CY27E Adj. EBITDA cross-check ($68.68/sh) exceeds Base DCF ($45.45/sh) by +51.1%
Applying CoreWeave’s 4.37x EV/CY27E Adj. EBITDA multiple to Nebius’s $6,140.2m FY27E Adjusted EBITDA gives $26,805.1m EV and $68.68/sh — +51.1% above Base DCF ($45.45/sh), yet still 46.6% below $128.64. The +51.1% gap arises because a single-year multiple capitalizes FY27E EBITDA without deducting the $88.23bn of FY26E–28E capex and −$50.18bn of FY27E–29E core FCF burn ex-advances required to build and refresh the fleet. CoreWeave’s 4.37x multiple reflects its larger live footprint (~1.5 GW active vs Nebius 407.6 MW connected).
| Reverse-DCF / parameter test | Penmark Base | Implied / stressed input | Value ($/sh) | Analytical assessment |
|---|---|---|---|---|
| 1. Long-run GPU yield (FY32E–40E, holding WACC/capex) | $15.00m / MW | $17.05m / MW (+$2.05m) | $128.64 | +13.7% above base; achievable if Token Factory scales (hence Low conviction) |
| 2. Near-term GPU yield (FY27E–31E, terminal at $15m) | $16.00m FY27E | $23.71m / MW (+$7.71m) | $128.64 | Requires every MW across 5 yrs to earn Q3 peak signing rates |
| 3. Equity beta / WACC (full-model solve on base FCF) | 1.54x / 10.03% | 0.93x / 8.03% WACC | $128.64 | Prices a 5.6x ND/EBITDA GPU builder like a regulated utility |
| 4. Implied EV / FY27E Adj. EBITDA multiple | 4.37x peer check | 7.53x (bridge) / 7.68x (spot) | $128.64 | 72%–76% premium to CoreWeave (4.37x) despite smaller live scale |
| Stress A: High-end 2.0x beta (11.53% WACC) | 1.54x (10.03%) | 2.00x beta (11.53% WACC) | $5.09 | High operating and financial leverage wipes out 89% of base equity |
| Stress B: Terminal growth at risk-free rate (4.81%) | 3.00% terminal g | 4.81% g (terminal return = WACC) | $47.30 | Zero excess terminal return prevents terminal-growth overstatement (+$1.85/sh) |
Source: Penmark Research model (single-parameter tests on base DCF; scenario probability-weight sensitivity is shown on page 22).
Scenario weighting ($86.84/sh today) rolls at 11.60% Ke to our $96.91 target (−24.7%, Sell)
Weighting Bear ($0.00, 30%), Base ($45.45, 45%) and Bull ($265.54, 25%) produces $86.84/sh on 12 Nov 2026, rolling 12 months at 11.60% Ke to $96.91/sh (−24.7% price return; Sell). Because 76% of the $86.84 weighted value sits in the bull case, probability weights are decisive and explain our Low conviction.
| Scenario / driver assumption | Bear (30%) | Base (45%) | Bull (25%) | Prob.-weighted |
|---|---|---|---|---|
| Connected power YE26E / FY27E addition (MW) | 620 / +600 | 720 / +1,000 | 880 / +1,300 | 730 / +955 |
| Connected power FY32E / FY40E (GW) | 1.92 / 1.92 | 5.22 / 7.09 | 7.48 / 10.45 | 4.80 / 6.38 |
| GPU yield: FY27E / FY28E–29E / FY32E+ ($m / billed MW) | 14.0 / 12.5–11.5 / 11.0 | 16.0 / 17.5–16.5 / 15.0 | 17.5 / 19.5–19.0 / 17.0 | 15.8 / — / 14.3 |
| AI Cloud Adj. EBITDA margin: FY27E / steady state | 46.0% / 50.0% | 50.0% / 61.0% | 54.0% / 65.0% | 49.8% / 58.7% |
| IT capex ($m/MW, held at base) / GPU life (yrs) | $30–32m / 6.0 yrs | $30–32m / 6.0 yrs | $30–32m / 6.0 yrs | $30–32m / 6.0 yrs |
| WACC (held at base) / convertible tranches converted | 10.03% / 0 of 6 | 10.03% / 0 of 6 | 10.03% / 4 of 6 | 10.03% / n/a |
| DCF equity value at 12 Nov 2026 price date ($/sh) | $0.00 | $45.45 | $265.54 | $86.84 |
| Rolled 12m forward at 11.60% Ke ($/sh) | $0.00 | $50.72 | $296.34 | $96.91 |
| Implied return vs $128.64 (12m / spot; 0.0% yield) | −100.0% / −100.0% | −60.6% / −64.7% | +130.4% / +106.4% | −24.7% / −32.5% (Sell) |
Source: Penmark Research model (unrounded price return is −24.67%).
Defending our 30% Bear / 45% Base / 25% Bull weights & rating sensitivity
Why Bull is capped at 25% and Bear carries 30%: Our Bull case ($265.54/sh; 4 of 6 convertible tranches converted) requires simultaneous execution on >1.2 GW annual energizations AND perpetual $17.0m/MW post-2031 yields. In 2026, management met only one of four original full-year goals on schedule (EBITDA margin held ~40%, whereas FY26 revenue moved to the low end, >$7bn ARR slipped from December to February, and FY26 capex rose from $20–25bn to ~$27bn), arguing against assigning >25% odds to flawless multi-year execution. Conversely, a 30% Bear weight reflects empirical precedent: spot H100 rental rates already reset sharply once within the past three years (2024), 200+ neoclouds plus hyperscalers are adding GW-scale supply, and $14.3bn of debt and SAFE claims wipe out equity if post-2028 yields drop to $11m/MW. Assigning the 45% residual to Base ($45.45/sh) keeps Sell intact even at 25 / 45 / 30 (~$111.7 target), while 20 / 45 / 35 (~$126.5) or a ~$107.68 share price at our $96.91 target moves the call to Hold (Exhibit 14, page 4).
| Company (Ticker) | Archetype | Mkt cap ($bn) | EV ($bn) | LTM EV/Rev | Rev. YoY | EBITDA mgn | EV / CY27E EBITDA |
|---|---|---|---|---|---|---|---|
| Nebius Group (NBIS) @ $128.64 | AI Cloud / GPU | 40.39* | 47.17 | 21.5x* | +454% | 19.0% (TTM) | 7.68x (Penmark) |
| Nebius Group @ $96.91 12m target | AI Cloud / GPU | 30.43* | 37.20 | n/a | +303% (27E) | 48.6% (27E) | 6.06x (Penmark) |
| CoreWeave (CRWV) @ $48.73 | Pure-play neocloud | 26.87 | 72.90 | 9.5x | +168% | 52.6% | 4.37x ($16.7bn E) |
| IREN Ltd (IREN) | BTC / AI power | 10.89 | 10.50 | 21.5x | −5% | −4.4% | n/a |
| Applied Digital (APLD) | AI / HPC hosting | 6.84 | 8.03 | 12.4x | +322% | −14.2% | n/a |
| TeraWulf (WULF) | BTC / HPC hosting | 5.38 | 6.22 | 53.4x | +51% | −171.8% | n/a |
| Equinix (EQIX) / Digital Realty (DLR) | Colocation REITs | 89.4 / 58.6 | 105.7 / 76.8 | 12.8x / 13.0x | +10% / +21% | 46.2% / 47.6% | n/d |
| Oracle Corp. (ORCL) | Hyperscale / OCI | 441.20 | 522.40 | 7.3x | +30% | 48.0% | n/d |
Source: Yahoo Finance (12 Nov 2026), company filings, Penmark Research model. *NBIS market cap uses 313.97m market-value shares ($40.39bn at $128.64; $30.43bn at $96.91 target); LTM EV/revenue is 21.5x on $2,189.8m LTM revenue. Only CoreWeave has usable CY27E Adj. EBITDA consensus ($16,700m), so we omit a multi-peer forward median.
2027 signposts skew negative (−$3.65/sh net); software-led $17m/MW yield ($266 bull) is our main upside risk
Over the next 12–18 months, probability-weighted catalyst downside (−$12.42/sh) exceeds upside (+$8.77/sh) by −$3.65/sh on base DCF ($45.45/sh). Because our Sell call carries Low conviction, structural upside risks merit equal weight alongside governance, counterparty and debt-wall downside.
| Window | Scope | Catalyst & Section 9 driver tested | Penmark expectation vs market | Prob. | Base $/sh | In cons.? |
|---|---|---|---|---|---|---|
| Jan 2027 | Comp. | Microsoft FY27 Q2 cloud report: outsourced neocloud lease commitments (util. −4pt to 88%) | Base holds 92% util.; market weighs pullback after Nov press reports | 30% | −$3.97 | Partly |
| Jan–Feb 27 | Macro | Hyperscaler 2027 capex guides: MSFT, META, GOOGL, AMZN AI spend (util. +3pt to 95%) | Spillover stays tight in 1H27; consensus ignores 95% util. ceiling | 35% | +$2.98 | No |
| Feb 2027 | 4Q26 | 4Q26 results & final Feb ARR print: YE26 connected MW (720E), 4Q sales & >$7.0bn Feb ARR | Penmark est. $1,169.3m 4Q26E sales & 720 MW YE26E; ARR dropped in 27 | Base | Signpost | Partly |
| 1H 2027 | Macro | Post-midterm permitting moratoria: state and local AI grid approvals (−150 MW FY27E) | Grid permitting slows US greenfield connections | 30% | −$4.48 | Partly |
| 1H 2027 | Comp. | Prepayment share gains: >100% capex-prepaid model vs debt-heavy rivals (+$0.5m/MW) | Zero-coupon advances cushion NBIS vs credit-constrained peers | 35% | +$2.85 | No |
| May 2027 | Comp. | 1Q27 report & 2Q27 colocation catch-up: Feb ARR >$7bn & delayed 3P online (+200 MW) | Base models +1.0 GW FY27E; faster 3P handover lifts FY27E to 1.2 GW | 40% | +$5.97 | Partly |
| 1Q–2Q 27 | Comp. | Owned & 3P site slippage: four greenfield sites plus colocation slip again (−300 MW) | 12 Nov delay precedent; repeat slip cuts FY27E additions to 700 MW | 35% | −$8.96 | Partly |
| 4Q26–3Q27 | Debt | 2027 funding gap & Oct 2027 Citi bridge: 15–25% uncovered capex (+150bp Kd) | Base uses 7.4% pre-tax cost of debt; $40–45bn guide raises spread risk | 40% | −$10.45 | Partly |
| FY27 | Comp. | FY27 bookings & software mix: >$25m/MW pricing (+$1.5m/MW) & Token Factory (+3pt mgn) | Base models $16.0m/MW & 50.0% AI Cloud mgn; scarcity adds upside | 30% / 35% | +$8.56 / +$5.07 | No / Ptly |
| FY27–28 | Comp. | GPU rental reset on renewals: 200+ neoclouds plus hyperscalers (−$1.5m/MW FY27E yield) | CEO sees compute shortage evening out in 1–2 yrs; 2024 H100 precedent | 30% | −$8.56 | No |
Source: Company filings, Q3 2026 shareholder letter; Penmark Research model (single-event shifts on $45.45/sh base DCF).
Next numbers to watch by Section 9 driver
Drivers 1 & 2 (yield & capex): Watch 4Q26E revenue ($1,169.3m), management’s final >$7.0bn Feb 2027 ARR goal, 1H27 Vera Rubin ACV/MW, and 4Q26E capex ($11.9bn; FY26E $27.31bn) alongside net deferred-revenue inflows (+$1.50bn 4Q26E).
Drivers 3 & 4 (MW & debt): Watch YE26 connected power (720 MW E vs 407.6 MW at 3Q26) and refinancing spreads ahead of the $3.5bn Citi bridge due Oct 2027.
Catalyst pass-through & what breaks our Sell thesis
Pass-through to target (~0.50x): Because Base carries 45% weight and rolls at 1.116x, our three main company upside events (+$8.56 FY27 pricing, +$5.97 2Q27 handover, +$5.07 margin mix) add ~$9.8/sh to target (~$106.8) and do not reach Hold alone.
Thesis-breaking upside: Durable software-led pricing ($16.5m/MW = $106.41 base; $17.0m/MW = $265.54 bull) or −$2m/MW IT capex deflation (+$40.93/sh) flips our call.
| Risk factor (direction) | Mechanism & filing evidence | Valuation & rating impact |
|---|---|---|
| Software moat & prepayments (Upside vs Sell) | Token Factory (17x QoQ Q3 contract value) or full 100%+ interest-free capex prepayments insulate long-run yields above $16m/MW without debt-spread dilution. | $16.5m/MW lifts base to $106.41/sh; Bull case ($17m/MW) = $265.54/sh. |
| Controlled company & key-person (Volozh) | CEO Arkady Volozh holds 28.78m shares (~9.8% of 292.90m 30 Sep shares) and held 51.6% voting control at 31 Mar 2026 via 10-vote Class B shares under Controlled Company exemptions. | Class A holders cannot block dilutive issuance or $40–45bn capex plans. |
| Hyperscaler, NVIDIA & grid concentration | Microsoft & Meta anchor fixed-price backlog while clusters rely 100% on NVIDIA GPUs (21.07m warrants); US greenfield sites face transformer queues and local grid moratoria. | Partner pullback (−4pt util.) cuts $3.97/sh; −300 MW slip cuts $8.96/sh. |
| Debt wall & GPU glut ($0.00/sh Bear case) | GPU glut from 2028 cuts yield to $11m/MW; expansion stops once contracted builds finish in FY29E; at a 50% margin, fleet replacement at $30m/MW earns less than its cost, leaving no equity value. | $14.3bn debt & SAFE claims absorb all EV in our 30%-weighted Bear case. |
Source: FY2025 Form 20-F, Q3 2026 earnings release and call; Penmark Research model (rating triggers summarized in Exhibit 14, page 4).
~51% 13F institutional ownership, 50.9m shares short and a $276.20 pre-print Street mean
NBIS trades $4.33bn per day (21.78m shares 3M average), combining high liquidity with crowded pre-print bullish positioning ($276.20 mean across 22 analysts; 14 Buy, 5 Hold, 2 Sell in Oct 2026) and 50.94m shares short. Delta-hedging across $13.45bn of convertible notes amplifies both short interest and turnover.
Source: FINRA bi-monthly short-interest settlements through 30 Oct 2026 (2.22 days to cover; +6.92% vs 15 Oct).
Source: S&P Global pre-print mean ($276.20, 22 analysts; dashed line = $128.64 price); Oct 2026 ratings split (14 Buy / 5 Hold / 2 Sell = 21); StockAnalysis 12 Nov cuts; Yahoo ($84–$410).
Post-print target revisions: On 12 Nov, Baird cut its target from $340 to $200 and Piper Sandler cut from $185 to $148 after Citi’s pre-print cut from $324 to $265; aggregator means ($276.20 S&P Global, $278.05 Yahoo) still reflect pre-print snapshots.
| Holder (Q2 2026 13F / filings) | Shares (m) | % base |
|---|---|---|
| Total 13F institutional ownership | 158.75 | 50.56% |
| BlackRock, Inc. | 11.53 | 3.67% |
| Goldman Sachs Group, Inc. | 9.80 | 3.12% |
| Invesco Ltd. | 6.94 | 2.21% |
| Orbis Allan Gray Ltd. | 6.49 | 2.07% |
| Bank of America Corp. | 6.14 | 1.96% |
| Strategic: NVIDIA pre-funded warrants ($0.0001) | 21.07 | 6.71% |
| Insider: Arkady Volozh (Class A + Class B) | 28.78 | 9.83%* |
Source: SEC Form 13F filings (Q2 2026 cycle) divided by 313.97m market-value shares; *Volozh stake is ~9.8% of 292.90m 30 Sep 2026 shares (9.17% of 313.97m; 51.6% voting control at 31 Mar 2026).
| Liquidity / convertible tranche | Amount / shs | Price / strike |
|---|---|---|
| 3M average daily turnover (ADTV) | 21.78m shs | $4,331.4m / day |
| FINRA short interest (30 Oct 2026) | 50.94m shs | 2.22 days cover |
| Unvested RSUs in diluted count | 5.74m shs | 319.71m diluted |
| Jun-25 converts due 2029/31 (ITM @ $129) | $200.0m | $51.45 (3.89m) |
| Sep-25 converts due 2030/32 | $3,162.5m | $138.75 (22.79m) |
| Mar-26 converts due 2031 & 2033 | $4,337.5m | $180.31–183.22 |
| Aug-26 converts due 2030 & 2034 | $5,750.0m | $313.46–324.65 |
| Total convertible principal (6 tranches) | $13,450.0m | 68.60m max shs |
Source: Yahoo Finance, FINRA, company filings and Penmark Research model (foreign private issuer; routine Rule 10b5-1 Form 144 notices omitted).
Annual model: revenue scales from $529.8m (FY25A) to $54.81bn (FY32E) and $76.94bn (FY40E)
Our Capacity-archetype model links group revenue to average billed IT megawatts and blended GPU yield per MW (FY23A revenue was $9.8m; Adj. EBITDA −$240.7m; GAAP loss −$299.0m). GAAP EBIT turns positive in FY27E ($419.8m); diluted EPS is $0.40 in FY26E (on ClickHouse gains), dips to −$1.68 in FY27E on debt interest and 5-year server D&A, then reaches $6.01 in FY28E.
| Capacity & revenue build | FY24A | FY25A | FY26E | FY27E | FY28E | FY29E | FY30E | FY32E | FY40E |
|---|---|---|---|---|---|---|---|---|---|
| Connected power, YE (MW) | n/d | n/d | 720 | 1,720 | 2,720 | 3,620 | 4,320 | 5,220 | 7,090 |
| Connected power added (MW) | n/d | n/d | 312* | 1,000 | 1,000 | 900 | 700 | 400 | 200 |
| Billed IT capacity, YE (MW) | n/d | 104.2 | 529.9 | 1,265.9 | 2,001.9 | 2,664.3 | 3,179.5 | 3,841.9 | 5,218.2 |
| Average billed IT load (MW) | n/d | n/d | 369.0* | 787.5 | 1,523.5 | 2,233.8 | 2,844.6 | 3,650.6 | 5,122.6 |
| GPU deployed incl. refresh (MW) | n/d | 104.2 | 425.8 | 736.0 | 736.0 | 662.4 | 515.2 | 720.2 | 1,089.3 |
| Revenue yield ($m / billed MW) | n/d | 12.0 | 12.6* | 16.0 | 17.5 | 16.5 | 16.0 | 15.0 | 15.0 |
| AI Cloud segment revenue ($m) | 68.3 | 480.3 | 3,103.4 | 12,600.3 | 26,661.6 | 36,857.0 | 45,514.2 | 54,758.4 | 76,838.4 |
| Avride, TripleTen & other ($m) | 23.2 | 49.5 | 28.0 | 30.8 | 33.9 | 37.3 | 41.0 | 49.6 | 106.3 |
| IT capex ÷ EBITDA per MW (yrs) | n/a | n/a | 5.7* | 4.0 | 3.15 | 3.0 | 3.1 | 3.3 | 3.3 |
Source: Company filings and Penmark Research model (*4Q26E stub figures for connected additions, average billed MW, yield and payback).
| Income statement ($m) | FY24A | FY25A | FY26E | FY27E | FY28E | FY29E | FY30E | FY32E | FY40E |
|---|---|---|---|---|---|---|---|---|---|
| Consolidated revenue | 91.5 | 529.8 | 3,131.4 | 12,631.1 | 26,695.5 | 36,894.3 | 45,555.2 | 54,808.0 | 76,944.7 |
| Revenue growth (% YoY) | +833.7% | +479.0% | +491.0% | +303.4% | +111.3% | +38.2% | +23.5% | +6.8% | +3.0% |
| AI Cloud Adjusted EBITDA | (128.5) | 59.0 | n/d* | 6,300.2 | 15,463.7 | 22,482.8 | 27,763.7 | 33,402.6 | 46,871.4 |
| Other & corp. Adj. EBITDA | (97.8) | (123.9) | n/d* | (160.0) | (140.0) | (120.0) | (100.0) | (60.0) | 0.0 |
| Group Adjusted EBITDA | (226.3) | (64.9) | 1,267.7 | 6,140.2 | 15,323.7 | 22,362.8 | 27,663.7 | 33,342.6 | 46,871.4 |
| Group Adj. EBITDA margin (%) | nm | −12.2% | 40.5% | 48.6% | 57.4% | 60.6% | 60.7% | 60.8% | 60.9% |
| Share-based compensation | (54.5) | (83.2) | (256.5) | (315.8) | (587.3) | (737.9) | (820.0) | (822.1) | (1,154.2) |
| One-off items & D&A | (118.8) | (463.6) | (1,408.1) | (5,404.6) | (10,516.9) | (15,197.3) | (18,802.1) | (21,088.4) | (28,253.9) |
| Operating income / (loss) | (399.6) | (611.7) | (396.9) | 419.8 | 4,219.5 | 6,427.6 | 8,041.6 | 11,432.1 | 17,463.3 |
| Net interest income / (exp.) | 63.6 | (29.7) | (347.7) | (955.4) | (1,925.7) | (2,709.3) | (3,165.1) | (3,157.8) | (1,216.9) |
| Investment gains & other, net | (17.0) | 655.2 | 856.3 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 |
| Profit / (loss) before tax | (353.0) | 13.8 | 111.7 | (535.6) | 2,293.8 | 3,718.3 | 4,876.5 | 8,274.3 | 16,246.4 |
| Income tax benefit / (exp.) | 1.0 | (4.0) | 17.3 | 0.0 | (373.5) | (929.6) | (1,219.1) | (2,068.6) | (4,061.6) |
| Net income (cont. ops) | (352.0) | 9.8 | 129.0 | (535.6) | 1,920.3 | 2,788.7 | 3,657.4 | 6,205.7 | 12,184.8 |
| Diluted GAAP EPS ($/sh) | n/a | n/a | $0.40 | ($1.68) | $6.01 | $8.72 | $11.44 | $19.41 | $38.11 |
Source: FY2025 Form 20-F, Q3 2026 release; Penmark Research model (*9M26 segment EBITDA split not disclosed; 4Q26E AI Cloud = $488.2m, Other = −$52.0m; 319.71m diluted shs). Markers throughout: n/a not applicable; n/d not disclosed; n/f not forecast; n/g not guided; nm not meaningful; nr not reported.
Core FCF ex-advances troughs at −$30.59bn (FY26E) and turns positive at +$3.07bn in FY30E
Customer advances (+$9,329m FY26E, +$9,000m FY27E) lift GAAP operating cash flow, then unwind (−$13,000m FY29E–32E) as prepaid megawatts enter service. Gross debt reaches $61.76bn in FY33E to maintain a $2.0bn cash floor and is held flat thereafter; post-FY34E FCF builds cash, cutting net debt to about $0.4bn by FY40E.
| Cash flow ($m) | FY24A | FY25A | FY26E | FY27E | FY28E | FY29E | FY30E | FY32E | FY40E |
|---|---|---|---|---|---|---|---|---|---|
| Net income (continuing ops) | (352.0) | 9.8 | 129.0 | (535.6) | 1,920.3 | 2,788.7 | 3,657.4 | 6,205.7 | 12,184.8 |
| D&A + share-based comp. | 131.6 | 501.1 | 1,602.5 | 5,720.4 | 11,104.2 | 15,935.2 | 19,622.1 | 21,910.5 | 29,408.1 |
| Deferred rev. change (advances) | 0.0 | 982.5 | 9,329.0 | 9,000.0 | 3,000.0 | (2,000.0) | (4,000.0) | (3,000.0) | 0.0 |
| Working capital & other op. | (49.5) | (1,091.5) | (5,014.1) | (475.0) | (703.2) | (509.9) | (433.0) | (174.2) | (110.9) |
| Operating cash flow (GAAP) | (269.9) | 401.9 | 6,046.4 | 13,709.8 | 15,321.4 | 16,214.0 | 18,846.4 | 24,942.0 | 41,482.0 |
| IT capex paid (outflow) | (646.0) | (3,252.8) | (18,663.0) | (23,552.0) | (22,080.0) | (18,105.6) | (14,939.6) | (24,444.4) | (33,070.5) |
| Owned DC capex paid (outflow) | (161.5) | (813.2) | (8,642.8) | (7,800.0) | (7,488.0) | (6,396.0) | (4,836.0) | (2,808.0) | (1,578.7) |
| Total cash capex | (807.5) | (4,066.0) | (27,305.8) | (31,352.0) | (29,568.0) | (24,501.6) | (19,775.6) | (27,252.4) | (34,649.3) |
| FCF incl. customer advances | (1,077.4) | (3,664.1) | (21,259.4) | (17,642.2) | (14,246.6) | (8,287.6) | (929.2) | (2,310.3) | 6,832.8 |
| FCF excl. customer advances | (1,077.4) | (4,646.6) | (30,588.4) | (26,642.2) | (17,246.6) | (6,287.6) | 3,070.8 | 689.7 | 6,832.8 |
| Memo: DCF unlev. FCF ex-adv. | n/a | n/a | (11,498.9)* | (26,002.6) | (15,534.8) | (3,532.0) | 6,178.3 | 3,240.5 | 7,667.7 |
Source: Company filings and Penmark Research model (*4Q26E stub unlevered FCF after 30 Sep 2026 valuation date).
| Balance sheet ($m) | FY25A | 3Q26A | FY26E | FY27E | FY28E | FY29E | FY30E | FY32E | FY40E |
|---|---|---|---|---|---|---|---|---|---|
| Cash & cash equivalents | 3,678.1 | 7,425.6 | 2,000.0 | 2,000.0 | 2,000.0 | 2,000.0 | 2,000.0 | 2,000.0 | 61,332.1 |
| Receivables & other current | 1,033.3 | 5,734.8 | 5,734.8 | 6,209.8 | 6,913.0 | 7,422.9 | 7,856.0 | 8,318.6 | 9,425.5 |
| PP&E, intangibles & goodwill | 5,573.0 | 20,029.5 | 31,384.9 | 57,332.3 | 76,383.4 | 85,687.6 | 86,661.1 | 92,372.5 | 116,913.3 |
| Investments, ROU & other NC | 2,146.2 | 5,554.2 | 5,554.2 | 5,554.2 | 5,554.2 | 5,554.2 | 5,554.2 | 5,554.2 | 5,554.2 |
| Total assets | 12,430.6 | 38,744.1 | 44,673.9 | 71,096.3 | 90,850.6 | 100,664.8 | 102,071.3 | 108,245.3 | 193,225.0 |
| Payables, accruals & lease liab. | 2,131.4 | 2,844.9 | 2,844.9 | 2,844.9 | 2,844.9 | 2,844.9 | 2,844.9 | 2,844.9 | 2,844.9 |
| Deferred revenue (prepayments) | 1,577.5 | 9,409.2 | 10,909.2 | 19,909.2 | 22,909.2 | 20,909.2 | 16,909.2 | 9,909.2 | 9,909.2 |
| Gross debt (converts, ABS & draw) | 4,127.7 | 14,190.8 | 18,825.0 | 36,467.2 | 50,713.8 | 59,001.4 | 59,930.6 | 60,782.4 | 61,761.4 |
| Shareholders’ equity (incl. NCI) | 4,594.0 | 12,299.2 | 12,094.9 | 11,875.0 | 14,382.7 | 17,909.3 | 22,386.7 | 34,708.8 | 118,709.6 |
| Net debt excl. deferred revenue* | 449.6 | 6,765.2* | 16,825.0 | 34,467.2 | 48,713.8 | 57,001.4 | 57,930.6 | 58,782.4 | 429.3 |
Source: FY2025 Form 20-F, Q3 2026 release; Penmark Research model (*3Q26A carrying net debt is $6,765.2m on $14,190.8m carrying debt, vs $6,777.9m principal net debt on Page 1; $1,215.5m restricted cash sits inside other current assets).
4Q26E reaches $1,169.3m sales (+413% YoY); net debt/EBITDA falls from 13.3x (FY26E) to 5.6x (FY27E)
Reported quarters plus our 4Q26E stub capture the inflection from −$63.9m Adjusted EBITDA in 4Q24A to +$436.2m in 4Q26E. Because management guides 2027 annually ($12–13bn revenue, 4Q27 pre-tax breakeven) and drops ARR from 2027, we model FY27E+ on an annual capacity-deployment basis.
| Quarterly & annual lines ($m) | 1Q25A | 2Q25A | 3Q25A | 4Q25A | 1Q26A | 2Q26A | 3Q26A | 4Q26E | FY27E |
|---|---|---|---|---|---|---|---|---|---|
| Consolidated revenue | 50.9 | 105.1 | 146.1 | 227.7 | 399.0 | 582.3 | 980.8 | 1,169.3 | 12,631.1 |
| AI Cloud ARR, period end | n/d | n/d | 551 | 1,250 | 1,900 | 3,000 | 6,400 | n/f* | n/f* |
| Group Adjusted EBITDA | (53.7) | (21.0) | (5.2) | 15.0 | 129.5 | 236.2 | 465.8 | 436.2 | 6,140.2 |
| Group Adj. EBITDA margin (%) | nm | −20.0% | −3.6% | 6.6% | 32.5% | 40.6% | 47.5% | 37.3% | 48.6% |
| AI Cloud Adj. EBITDA margin (%) | n/d | n/d | n/d | 24.0% | 45.0% | 49.7% | 53.2% | 42.0% | 50.0% |
| Depreciation & amortization | (49.1) | (75.2) | (99.0) | (180.7) | (212.0) | (259.7) | (354.7) | (519.6) | (5,404.6) |
| Share-based compensation | (17.5) | (14.7) | (26.1) | (24.9) | (35.3) | (102.5) | (58.7) | (60.0) | (315.8) |
| Net income (cont. ops) | (104.3) | 502.5 | (119.6) | (249.6) | 621.2 | (190.4) | (37.5) | (264.3) | (535.6) |
| Operating cash flow (GAAP) | (184.1) | (167.9) | (80.6) | 834.3 | 2,258.0 | 2,246.1 | (273.0) | 1,815.3 | 13,709.8 |
| Cash capex (outflow) | (543.9) | (510.6) | (955.5) | (2,056.0) | (2,472.9) | (5,657.4) | (7,300.4) | (11,875.1) | (31,352.0) |
Source: Company filings and Penmark Research model (4Q24A revenue $35.2m, Adj. EBITDA −$63.9m, capex $417.5m). Quarters as released; FY25 totals per the 20-F, which restates them. *n/f = not forecast: management no longer reports ARR from 2027 (final >$7.0bn ARR goal set for February 2027).
| Ratio / per-share metric | FY25A | FY26E | FY27E | FY28E | FY29E | FY30E | FY31E | FY32E | FY40E |
|---|---|---|---|---|---|---|---|---|---|
| GAAP EBIT margin (%) | −115.5% | −12.7% | 3.3% | 15.8% | 17.4% | 17.7% | 18.0% | 20.9% | 22.7% |
| Core FCF margin ex-adv. (%) | −877.0% | −976.8% | −210.9% | −64.6% | −17.0% | 6.7% | 10.6% | 1.3% | 8.9% |
| Return on average equity (ROE) | 0.2% | 1.5% | −4.5% | 14.6% | 17.3% | 18.2% | 18.1% | 19.9% | 10.9% |
| Return on average assets (ROA) | 0.1% | 0.5% | −0.9% | 2.4% | 2.9% | 3.6% | 4.4% | 5.9% | 6.5% |
| Pre-tax ROCE (EBIT / avg IC) | nr | −2.0% | 1.1% | 7.4% | 9.1% | 10.1% | 10.9% | 12.5% | 10.0% |
| Net debt (ex-def. rev.) / equity | 10% | 139% | 290% | 339% | 318% | 259% | 204% | 169% | 0% |
| Net debt / Group Adj. EBITDA | nm | 13.27x | 5.61x | 3.18x | 2.55x | 2.09x | 1.81x | 1.76x | 0.01x |
| Adj. EBITDA / net interest | nm | 3.65x | 6.43x | 7.96x | 8.25x | 8.74x | 9.70x | 10.56x | 38.52x |
| EV / Adj. EBITDA @ $128.64* | nm | 37.21x | 7.68x | 3.08x | 2.11x | 1.71x | 1.51x | 1.41x | 1.01x |
| EV / Adj. EBITDA @ $96.91 target* | nm | 29.35x | 6.06x | 2.43x | 1.66x | 1.35x | 1.19x | 1.12x | 0.79x |
| P/E @ $128.64 | n/a | 318.9x | nm | 21.4x | 14.7x | 11.2x | 9.1x | 6.6x | 3.4x |
| P/E @ $96.91 target | n/a | 240.2x | nm | 16.1x | 11.1x | 8.5x | 6.8x | 5.0x | 2.5x |
| Book value / share ($) | nr | $37.83 | $37.14 | $44.99 | $56.02 | $70.02 | $86.58 | $108.57 | $371.31 |
| P/B @ $128.64 | 8.8x | 3.4x | 3.5x | 2.9x | 2.3x | 1.8x | 1.5x | 1.2x | 0.3x |
Source: Company filings and Penmark Research model (*EV excludes deferred revenue: $47,166.7m at $128.64 and $37,204.6m at $96.91 target on 313.97m market-value shares; 0.0% dividend yield in all periods).
Reference: $826.4m 9M26 D&A and $196.5m SBC turn $831.5m Adj. EBITDA into a $253.5m EBIT loss
| Reconciliation line ($m) | FY24A | FY25A | 9M26A | FY26E | FY27E | FY28E |
|---|---|---|---|---|---|---|
| Group Adjusted EBITDA (reported / Penmark) | (226.3) | (64.9) | 831.5 | 1,267.7 | 6,140.2 | 15,323.7 |
| Less: D&A (5-year depreciation, 6-year replacement; 20-yr DC) | (77.1) | (417.9) | (826.4) | (1,346.0) | (5,404.6) | (10,516.9) |
| Less: Share-based compensation (expensed as cash cost in DCF) | (54.5) | (83.2) | (196.5) | (256.5) | (315.8) | (587.3) |
| Less: One-off & transaction items (non-recurring & M&A) | (41.7) | (45.7) | (62.1) | (62.1) | 0.0 | 0.0 |
| GAAP Operating income / (loss) | (399.6) | (611.7) | (253.5) | (396.9) | 419.8 | 4,219.5 |
| Net interest + ClickHouse/equity revaluations & taxes | 47.6 | 621.5 | 646.8 | 525.9 | (955.4) | (2,299.2) |
| GAAP Net income / (loss), continuing operations | (352.0) | 9.8 | 393.3 | 129.0 | (535.6) | 1,920.3 |
Source: FY2025 Form 20-F, Q3 2026 earnings release; Penmark Research model (Toloka deconsolidated in 2Q25 discontinued operations).
Glossary & blank-marker legend
- Connected MW: Grid power energized at data-center halls (720 MW YE26E).
- Billed IT MW: Connected MW × 0.80 IT-to-power ratio (1/PUE) × 92% billed utilisation = 73.6% conversion.
- Revenue yield ($m/MW): Annual AI Cloud revenue ÷ average billed IT MW.
- AI Cloud ARR: Final month AI Cloud revenue × 12; dropped from 2027.
- Customer prepayments: Upfront contract cash in deferred revenue ($9,409.2m at 3Q26).
- Table markers: n/a not applicable; n/d not disclosed; n/f not forecast; n/g not guided; nm not meaningful; nr not reported / boundary balance.
Issuer basis & valuation methodology
Capacity archetype (foreign private issuer, controlled, levered ramp): Nebius Group N.V. (Netherlands) reports in US$ millions (FY = 31 Dec); Nasdaq Class A shares trade directly 1:1 (not an ADR; no FX translation on US$ reporting), while Class B carries 10 votes. Our 15-year DCF to FY40E captures negative core FCF to FY29E and 6-year GPU replacement, assuming returns on new capital equal the cost of capital beyond FY40E. EV/Adjusted EBITDA replaces P/E where FY27E GAAP EPS is negative; Sections 16–17 share Page 23.
Data cut-off: Filings and Q3 2026 results through 12 Nov 2026; share price 12 Nov close ($128.64); publication 13 Nov 2026.
| Measure | Definition in Penmark model | Scope & basis |
|---|---|---|
| Core FCF (excl. advances) | GAAP operating cash flow less cash capex less net change in deferred revenue. | Continuing ops |
| DCF Unlevered FCF | Unlevered EBIT (post-SBC) less 25% cash tax (after 60% US bonus dep. & 80% NOL cap) less capex & WC. | 30 Sep 2026 DCF |
| Enterprise value | Market value (313.97m shs × $128.64) + principal debt ($14,203.5m) less unrestricted cash ($7,425.6m) = $47,166.7m. | Excl. deferred rev. |
| Share counts (313.97m / 319.71m) | 313.97m market-value shares = 30 Sep shares (292.90m) + NVIDIA warrants (21.07m); 319.71m diluted adds 5.74m RSUs. | Converts if ITM |
| Pre-tax ROCE & P/B | GAAP EBIT / average (equity + gross debt); FY25A P/B uses $40.39bn mkt cap / $4.59bn equity, FY26E+ uses 319.71m BVPS. | Balance-sheet IC |
Source: Penmark Research model (operating lease liabilities excluded from net debt as colocation lease expense sits inside Adjusted EBITDA).
| Peer (Ticker) | Business model & comparability | Role in report |
|---|---|---|
| CoreWeave (CRWV) | Pure-play GPU neocloud (~1.5 GW active; $72.9bn EV; $16.7bn CY27E EBITDA). | 4.37x EV/EBITDA check |
| IREN, APLD, WULF | Listed AI/HPC power & hosting converters ($5–11bn mkt cap; no usable CY27E EBITDA consensus). | EV/Sales context only |
| EQIX, DLR, ORCL | Data-center colocation REITs & OCI hyperscale cloud ($59–441bn mkt cap). | Value-chain / margin ref. |
Source: Company filings, S&P Global and Yahoo Finance as of 12 November 2026; Penmark Research model.
Analyst certification, rating bands (−10% to +10% Hold) and important disclosures
Ratings
Ratings are based on the upside from the share price to our 12-month price target, excluding dividends. The price target is weighted across bull, base and bear scenarios.
| Buy | Upside to price target above +10% |
| Hold | Upside to price target from −10% to +10%, inclusive |
| Sell | Upside 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
Access to Penmark Research reports at penmarkresearch.com and any downloads or programmatic use are subject to Penmark Research’s applicable website 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.