Financial analysis, ratio analysis, sector benchmarking, three-year scenarios and sensitivity. All figures from SEC filings unless marked otherwise.
Share price
$213.93
26 Aug 2026; $223.90 on 19 Aug per 6-K
Enterprise value
$62.5B
mkt cap $62.0B + net debt $0.5B
ARR, Jun-2026
$3.0B
+598% YoY; +56% QoQ
EV / ARR
20.8x
7.8x on YE-26 guide midpoint
Q2-26 adj. EBITDA
$236.2M
40.6% margin; AI cloud 49.7%
LTM capex
$11.1B
8.2x LTM revenue
Nebius has already won the operating argument and now trades entirely on the financing one. Revenue grew 454% year over year in Q2-2026 and the AI cloud segment earns a 49.7% adjusted EBITDA margin, but the company has burned $7.3B of free cash flow in the last six quarters and carries $13.5B of debt against a business that recognised $1.36B of revenue in the trailing twelve months. The equity is a levered call option on one variable: whether customer prepayments keep covering half of capex.
Three findings drive everything below. First, the unit economics inflected in Q2-2026. New contracts price at $20-25M of annual contract value per megawatt against a 2026 base of $12M, cutting payback to 1 year 10 months from two to three years, and 70% of those deals carry prepayments covering 50-60% of the associated capex. Second, Nebius funds cheaper than any listed peer. Its July secured facility priced at SOFR + 250bp and its August convertibles at 0.50% and 4.50%, while CoreWeave repriced a term loan to SOFR + 550bp and trades on roughly 855bp of five-year CDS. Third, the market has already paid for this. At 1.56x enterprise value per dollar of contracted customer commitments, Nebius costs twice what CoreWeave costs (0.75x), and at 20.8x current ARR it needs the $7-9B year-end ARR guide to land near the top to justify today's price.
Base case values the equity at $340-355 per share by end-2029 (+59% to +66%, a 15% IRR). Bull case reaches $700-710 (+230%). Bear case falls to $28-42 (-80% to -87%). The distribution is unusually wide because the same balance sheet that produces a 43% IRR on success produces near-total loss on failure.
Nebius Group is the post-divestiture remainder of Yandex N.V. It sold its Russian operations in July 2024, renamed, and rebuilt around an AI cloud business that began generating revenue in 2023. Three reporting segments remain: Nebius AI cloud (98% of Q2-2026 revenue), Avride (autonomous vehicles) and TripleTen (edtech). It also holds minority stakes in ClickHouse and Toloka, both spun out of the group.
Two accounting facts govern every historical comparison. Russian operations sit in discontinued operations from 2024. Toloka was deconsolidated in Q2-2025 and every prior period was recast, which is why FY2024 revenue reads $91.5M here rather than the $117.5M reported at the time. FY2022 is not comparable on any basis and is excluded.
Segment detail, FY2025 and Q2-2026 ($M)
Segment
FY24 rev
FY25 rev
Q2-26 rev
Q2-26 adj. EBITDA
Margin
Nebius AI cloud
68.3
480.3
574.9
285.7
49.7%
Avride
0.3
1.3
1.0
(40.1)
n/m
TripleTen
28.8
54.1
10.0
(9.4)
(94)%
Eliminations
(5.9)
(5.9)
(3.6)
—
—
Group
91.5
529.8
582.3
236.2
40.6%
The AI cloud segment carries the entire group. Avride's loss widened from $17.3M to $40.1M year over year and is now a $160M annualised drag on group EBITDA.
Consolidated statements of operations, continuing operations ($M). LTM = FY2025 less H1-2025 plus H1-2026.
Income statement
FY2023
FY2024
FY2025
H1-2025
H1-2026
LTM Jun-26
FY23-25 CAGR
Revenue
9.8
91.5
529.8
156.0
981.3
1,355.1
635%
Nebius AI cloud
9.6
68.3
480.3
135.1
964.6
1,309.8
607%
Operating costs and expenses
Cost of revenue (ex-D&A)
19.6
43.7
166.2
54.8
237.4
348.8
191%
Product development
87.1
114.8
177.3
79.3
258.4
356.4
43%
Sales, general and administrative
159.5
255.5
380.1
129.1
317.7
568.7
54%
Depreciation and amortisation
29.3
77.1
417.9
124.3
471.7
765.3
278%
Total operating costs
295.5
491.1
1,141.5
387.5
1,285.2
2,039.2
96%
Loss from operations
(285.7)
(399.6)
(611.7)
(231.5)
(303.9)
(684.1)
—
Interest income
3.3
63.6
31.8
12.1
38.3
58.0
Interest expense
—
—
(61.5)
(4.8)
(182.8)
(239.5)
Gain on revaluation of equity securities
—
—
598.9
597.4
780.6
782.1
Equity method and other income, net
(14.6)
(17.0)
56.3
26.7
106.5
136.1
Net income / (loss), continuing ops
(299.0)
(352.0)
9.8
398.2
430.8
42.4
—
Non-GAAP and supplementary
Adjusted EBITDA
(240.7)
(226.3)
(64.9)
(74.7)
365.7
375.5
—
Adjusted net loss
n/d
n/d
n/d
(175.2)
(133.5)
n/d
—
Share-based compensation
28.8
54.5
83.2
32.1
137.8
188.9
70%
Reading note. FY2025 net income of $9.8M is not operating profit. It exists only because a $598.9M non-cash gain on the ClickHouse stake offset a $611.7M operating loss. The same applies to H1-2026: $430.8M of net income contains a $780.6M ClickHouse revaluation. Strip both and the business lost $589M in FY2025 and $350M in H1-2026 before tax. Cost of revenue and product development are stated exclusive of D&A throughout, so the 74% "gross margin" implied by revenue less cost of revenue is not an economic gross margin.
Balance sheet ($M)
Dec-24
Dec-25
Jun-26
Cash and equivalents
2,434.7
3,678.1
8,042.1
Accounts receivable
11.2
720.3
288.6
Total current assets
2,533.3
4,711.4
9,615.6
Property and equipment, net
846.7
5,553.3
13,045.2
Goodwill and intangibles
4.9
19.7
689.5
Operating lease right-of-use
44.8
918.8
1,855.1
Non-marketable equity securities
90.7
836.6
1,606.7
Total assets
3,548.6
12,430.6
27,961.5
Accounts payable and accrued
228.0
1,210.1
1,301.0
Deferred revenue, current
16.3
275.5
979.4
Deferred revenue, non-current
—
1,302.0
4,995.8
Debt, total
6.1
4,127.7
8,545.7
Operating lease liabilities
30.3
760.5
1,510.4
Total liabilities
294.9
7,836.6
17,621.0
Total shareholders' equity
3,253.7
4,594.0
10,340.5
Cash flow ($M)
FY2023
FY2024
FY2025
H1-26
Operating cash flow
(222.0)
(269.9)
401.9
4,504.1
of which increase in deferred revenue
1.9
10.0
1,565.8
4,395.0
OCF excluding deferred revenue
(223.9)
(279.9)
(1,163.9)
109.1
Capital expenditure
(82.9)
(807.5)
(4,066.0)
(8,130.3)
Free cash flow
(304.9)
(1,077.4)
(3,664.1)
(3,626.2)
Financing inflow
375.6
825.5
5,125.5
9,160.4
Convertible notes issued
—
—
4,162.5
4,337.5
Equity and warrants
—
700.0
1,150.0
4,846.7
The single most important line in the accounts
Positive operating cash flow at Nebius is a prepayment phenomenon, not a profit phenomenon. Deferred revenue supplied $1,565.8M of FY2025 operating cash flow and $4,395.0M of the $4,504.1M reported in H1-2026. Strip it out and FY2025 operating cash flow was negative $1,163.9M and H1-2026 was positive $109.1M. Deferred revenue on the balance sheet rose from $16.3M to $5,975.2M in eighteen months.
Quarterly progression, group continuing operations ($M except ARR in $B)
Q1-25
Q2-25
Q3-25
Q4-25
Q1-26
Q2-26
Revenue
50.9
105.1
146.1
227.7
399.0
582.3
QoQ growth
—
107%
39%
56%
75%
46%
Adjusted EBITDA
(53.7)
(21.0)
(5.2)
15.0
129.5
236.2
Adjusted EBITDA margin
(105)%
(20)%
(4)%
7%
32%
41%
AI cloud adj. EBITDA margin
n/d
10%
n/d
24%
45%
50%
Cost of revenue % of revenue
n/d
29%
n/d
30%
26%
23%
D&A % of revenue
n/d
72%
n/d
79%
53%
45%
Capital expenditure
543.9
510.6
~900
~2,100
2,472.9
5,657.4
Exit ARR ($B)
n/d
0.43
0.55
1.25
1.90
3.00
Contracted power target for YE-2026
—
>2.5 GW
>2.5 GW
>3 GW
>4 GW
5 GW
Q3-25 and Q4-25 capex are the company's approximations from the shareholder letters. Q2-25 ARR of $0.43B is implied from the Q2-26 disclosure of +598% year over year. The contracted-power target has been raised at every reporting date for four consecutive quarters.
Ratios computed from the filings above. LTM balance-sheet ratios use the 30 June 2026 balance sheet.
Ratio
FY2023
FY2024
FY2025
LTM Jun-26
Q2-26 alone
Direction
Growth
Revenue growth, year over year
n/m
834%
479%
n/m
454%
Decelerating from a tiny base
ARR growth, year over year
n/d
n/d
>700%
—
598%
Still accelerating in absolute dollars
Profitability
Revenue less cost of revenue, % of revenue
(100)%
52.2%
68.6%
74.3%
77.1%
Improving; excludes D&A
Adjusted EBITDA margin
n/m
(247)%
(12.2)%
27.7%
40.6%
Crossed zero in Q4-2025
GAAP operating margin
n/m
(437)%
(115)%
(50.5)%
(30.2)%
Still deeply negative
Net margin, continuing operations
n/m
(385)%
1.8%
3.1%
(32.7)%
Distorted by ClickHouse mark
Cost structure
Cost of revenue % of revenue
200%
47.8%
31.4%
25.7%
22.9%
Operating leverage is real
Product development % of revenue
889%
125%
33.5%
26.3%
32.8%
Q2 inflated by Eigen AI charge
SG&A % of revenue
1,628%
279%
71.7%
42.0%
29.9%
Scaling well
D&A % of revenue
299%
84.3%
78.9%
56.5%
44.6%
The structural margin ceiling
Share-based comp % of revenue
294%
59.6%
15.7%
13.9%
17.6%
Rising again in 2026
Capital intensity and efficiency
Capex % of revenue
846%
883%
768%
822%
971%
No sign of easing
Capex / depreciation and amortisation
2.8x
10.5x
9.7x
14.6x
21.8x
Depreciation burden is front-loaded ahead
Revenue / closing net PP&E
n/a
0.11x
0.10x
0.10x
0.18x ann.
Asset base is not yet earning
Total asset turnover
n/a
0.03x
0.04x
0.05x
0.08x ann.
Improving
Days sales outstanding
n/a
45
496
78
45
Dec-25 spike reversed in H1-26
Liquidity, leverage and coverage
Current ratio
n/a
9.6x
3.1x
4.0x
—
Comfortable
Cash / total debt
n/a
399x
0.89x
0.94x
1.02x pro forma
Roughly cash-neutral
Total debt / equity
0.00x
0.00x
0.90x
0.83x
—
Converts dominate
Debt / total capital
0%
0%
47%
45%
—
Adjusted EBITDA / interest expense
n/m
n/m
(1.1)x
1.6x
2.0x
Thin, and interest is rising fast
Deferred revenue / revenue
0.19x
0.18x
2.98x
4.41x
—
The funding model in one number
Returns
Return on closing equity, continuing ops
n/m
(10.8)%
0.2%
0.4%
—
Only positive via the ClickHouse mark
Return on invested capital, pre-tax
n/m
(12.3)%
(7.0)%
(3.6)%
—
Improving toward zero
Ratio 1: D&A at 45% of revenue
Depreciation is the ceiling on GAAP profitability. Even at Q2-2026's 41% adjusted EBITDA margin, D&A of 45% of revenue leaves a 30% operating loss. Nebius extended server and network useful life from four to five years effective Q1-2026. On the old four-year life, Q2-2026 D&A would have been roughly $325M and the operating loss roughly $241M rather than $176M.
Ratio 2: deferred revenue at 4.4x revenue
Customers have prepaid $5,975.2M against a business that recognised $1,355.1M over the last twelve months. That is a real funding advantage and a real obligation: it must be delivered as capacity, on time, at contracted prices. It also means reported operating cash flow leads revenue by several quarters and will reverse when growth slows.
Ratio 3: interest coverage at 1.6x
LTM interest expense of $239.5M against $375.5M of adjusted EBITDA. Interest expense ran at $119.1M in Q2-2026 alone, and the August 2026 issuance adds $5.0B of face at a 0.50% and 4.50% blended coupon with accretion to 110% and 125% of principal at maturity. Coverage improves only if the ARR ramp lands.
Hyperscaler capital expenditure, cash purchases of property and equipment plus finance-lease additions ($B). Microsoft and Oracle are June and May fiscal years.
Company
FY2023
FY2024
FY2025
FY2026 / TTM
Company guidance
Source
Microsoft
31.2
56.1
85.1
140.6
~$175B calendar 2026; "rise further" in FY2027
FY26 10-K, Q4 call 29-30 Jul 26
Amazon
48.1
78.5
131.8
173.0
~$220B 2026, raised from ~$200B
FY25 10-K, Q2 10-Q, Q2 call 30 Jul 26
Alphabet
32.3
52.5
91.4
132.4
$195-205B 2026, raised from $180-190B
FY25 10-K, Q2 10-Q, Q2 call 22 Jul 26
Meta
28.1
39.2
72.2
92.4
$130-145B 2026; declined to guide 2027
FY25 10-K, Q2 10-Q, Q2 call 29-30 Jul 26
Oracle
8.7
6.9
21.2
55.7
~$70B net cash capex FY2027
FY26 10-K, Q4 call 10 Jun 26
Aggregate guided 2026
148.4
233.2
401.7
594.1
$720-745B big four, ~$835B with Oracle
No hyperscaler cut capex in 2026. Every one raised or held. Amazon stated it will not have enough capacity to meet demand in 2026 or 2027. Microsoft's headline $15B reduction was an accounting reclassification of building useful life from 15 to 25 years, not a spending decision. The one soft signal is Meta explicitly declining to guide 2027 capex.
Nvidia data centre revenue ($B) and forward commitments
Quarter
DC revenue
YoY
Q3 FY26 (Oct-25)
51.2
+66%
Q4 FY26 (Jan-26)
62.3
Q1 FY27 (Apr-26)
75.2
+92%
Q2 FY27 (Jul-26)
89.0
+117%
Supply and capacity commitments
$119B (Apr-26) to $279B (Jul-26)
Nvidia's Q2 FY2027 10-Q, filed 26 August 2026, states the company is "currently experiencing certain supply constraints" and that "the availability of land, power, shell, and capital is crucial".
The sentence that frames the Nebius thesis
Nvidia's own 10-Q, filed today: AI clouds and AI model makers "have significant demand ... but currently lack the ability to secure long-term infrastructure contracts and investment-grade financing capacity." Nebius is the neocloud furthest along on both. It holds $40B of customer commitments, delivered every Microsoft tranche on schedule, and raised $775M in July at SOFR + 250bp on the back of an investment-grade counterparty contract.
Nvidia's stake. Nvidia bought $2.0B of pre-funded warrants over 21,065,936 Class A shares at a $0.0001 exercise price in March 2026, an effective $94.94 per share. Its 13F position of 1,190,476 ordinary shares has been unchanged since December 2024, so the warrants are the real exposure and they do not appear in 13F reporting.
Indicative GPU rental prices, $ per GPU-hour. Commercial aggregators, not audited data.
GPU class
Neocloud median, Aug-26
Hyperscaler median, Aug-26
Peak 2024 (hyperscaler)
Direction
H100 80GB
$4.17
$7.89
$8.98-9.39
+7.0% recent period
H200
$4.50
$10.44
n/a
+14.4% recent period
B200
n/d
$8.60
n/a
Supply-constrained
GB300 NVL72, per GPU
n/d
$7.20-7.95
n/a
Supply-constrained
Nebius's own disclosure corroborates this: Q2-2026 deals repriced above $20M of annual contract value per megawatt with more than 30% higher pricing on older-generation GPUs versus Q1, and a first capacity auction cleared the highest price the company has achieved for Blackwell chips. The bear thesis that GPU rental pricing collapses as supply arrives is not visible in 2026 data.
Neocloud peer set. Prices 26 August 2026; balance-sheet items at each company's latest period end. IREN data is 31 March 2026 and Applied Digital is 31 May 2026, so both are stale relative to the calendar-quarter peers.
Company
Price
Mkt cap $B
EV $B
TTM rev $B
EV / TTM rev
Latest Q rev ann. $B
EV / that
Latest Q adj. EBITDA ann. $B
EV / that
Backlog $B
EV / backlog
Nebius
213.93
62.0
62.5
1.36
46.1x
2.33
26.8x
0.94
66.1x
>40.0
1.56x
CoreWeave
88.01
48.5
78.3
7.59
10.3x
10.30
7.6x
6.04
13.0x
~104.0
0.75x
IREN
39.58
14.1
15.9
0.76
21.0x
0.58
27.4x
0.24
66.8x
3.1 ARR
n/a
Applied Digital
27.21
7.9
11.4
0.61
18.6x
1.03
11.0x
0.17
67.1x
~36.0
0.32x
Cipher Digital
16.02
6.6
11.4
0.19
59.5x
0.10
114.5x
(0.12)
n/m
~11.4
1.00x
TeraWulf
15.97
8.0
10.6
0.17
63.9x
0.18
58.9x
(0.07)
n/m
~19.0
0.56x
Enterprise value convention: market cap plus total borrowings and finance leases less unrestricted cash. Operating lease liabilities are excluded, which understates CoreWeave by $16.3B, and restricted cash is excluded, which overstates Cipher by $3.7B and Applied Digital by $2.4B. Nebius market cap uses 289.9M shares (271.9M outstanding at 30 June 2026, plus ATM sales, plus 15.8M shares issued in the 24 August note exchange).
SOFR + 250bp, backed by deployed GPUs and an investment-grade customer contract
Jul 2026
Nebius
Convertible notes, $5.0B
0.50% due 2030 and 4.50% due 2034; conversion at $313.46 and $324.65, 40% and 45% premiums
19 Aug 2026
CoreWeave
Delayed-draw term loan repricing
SOFR + 425-450 at 99 repriced to SOFR + 550 at 97; 10.44% all-in; 1.35x DSCR covenant added
3 Aug 2026
CoreWeave
Five-year credit default swap
~855bp, implying roughly 50% five-year default probability
28 Jul 2026
CoreWeave
Senior notes
8.50% to 9.75% coupons across five series
FY2025-26
Applied Digital
Senior secured notes
6.750% to 9.250%
FY2026
Cipher Digital
Senior secured notes
6.000% to 7.125%
FY2026
TeraWulf
Senior secured notes due 2030
7.750% coupon, 8.7% effective
FY2025
The spread between Nebius at SOFR + 250bp and CoreWeave at SOFR + 550bp with 855bp of CDS is the sharpest read on the sector. It is not GPU pricing that separates these businesses. It is who can raise capital and at what cost, which is exactly the constraint Nvidia identified in its own filing.
Each factor is stated with the disclosure that supports it and the specific way it fails.
#
Factor
Evidence and insight
How it breaks
Weight in bull case
1
Contract pricing stepped up, and payback halved
Q2-2026 deals averaged above $20M of annual contract value per MW against a $12M 2026 base, with $20-25M per MW on four deals each above $1B of total contract value. Total contract value of Q2 wins grew roughly 4x quarter over quarter and new-customer TCV grew more than 9x. Expected payback on Q2 deals fell to 1 year 10 months from a two to three year range. A first short-term capacity deal signed in Q3 points to a $40-50M per MW price band.
Pricing is set at contract signature against capacity that mostly arrives in late 2026. If Blackwell and Rubin supply normalises faster than power does, the $20M+ per MW band compresses back toward $12M on renewals, and payback returns to three years.
Highest. Sets the return on every dollar of the $47B capex programme.
2
Customers fund half the capex
70% of Q2 deals included prepayments, an all-time high, covering 50-60% of the associated capex. Deferred revenue rose from $1,577.5M at Dec-2025 to $5,975.2M at Jun-2026. The company expects more than $9B of customer prepayments in 2026. This converts a capital-markets-dependent business into a partly self-funding one.
Prepayments come from AI labs that are themselves loss-making and funded by venture and strategic capital. A funding squeeze at the lab layer stops prepayments before it stops demand, and it is the first thing to stop.
Very high. Determines whether external capital need is $4B or $20B.
3
Power, not GPUs, is the scarce asset
Contracted power target raised four consecutive quarters: >2.5 GW in Aug-2025, >3 GW in Feb-2026, >4 GW in May-2026, 5 GW now. PJM and ERCOT interconnection queues exceed 300 GW with three to four year waits in data-centre load zones; substation transformer lead times exceed 160 weeks. PJM peak demand grows 32 GW to 2030, of which 30 GW is data centres.
Contracted power is an option, not capacity. Converting 5 GW contracted into 800MW-1GW connected by year end and >1 GW per year from 2027 requires construction execution Nebius has done once, at 170MW scale.
High. Gates the volume side of every scenario.
4
Cheapest capital in the peer set
SOFR + 250bp secured in July; 0.50% and 4.50% converts in August, upsized from $4.5B to $5.0B, at 40% and 45% conversion premiums. Simultaneously exchanged $800M of 2029 and 2031 notes for 15.8M shares, retiring the most dilutive paper. Compare CoreWeave at SOFR + 550bp and 855bp CDS.
The August converts accrete to 110% and 125% of principal at maturity, so the true cost is well above the coupon. If the equity de-rates, converts become straight debt at $12.7B of face.
High. A 300bp funding advantage on $40B+ of capital is worth more than most operating improvements.
5
Anchor contracts with a floor
All Microsoft capacity tranches delivered on schedule. Meta orders total $12B of contract value with deployments in tranches from early 2027, and one order obliges Meta to purchase any capacity Nebius fails to sell to third parties from specified clusters. Nvidia holds $2.0B of pre-funded warrants.
Customer concentration is not disclosed by Nebius. Microsoft plus Meta plus four deals each above $1B implies concentration comparable to CoreWeave's disclosed 36% and 26% top-two, without the disclosure.
Moderate to high. The Meta backstop puts a floor under 2027 utilisation.
6
Software and inference mix
Token Factory production inference workloads more than tripled in Q2. Acquisitions of Tavily (2.5M developers, up from 1M in February), Clarifai and Eigen AI brought inference optimisation in house. Inference is less capital-intensive per dollar of revenue than training.
Acquisition cost is visible and the revenue is not. Q2 product development carried $115.9M of non-recurring share-based and post-combination expense from Eigen AI alone. None of these acquisitions has disclosed revenue.
Moderate. Matters for terminal margin, not for the next three years.
7
Non-core stakes
ClickHouse marked at a reported $15B January 2026 Series D, driving $598.9M of gain in FY2025 and $780.6M in H1-2026. Non-marketable equity securities carried at $1,606.7M at Jun-2026, plus a Toloka stake and Avride and TripleTen.
Marks are level 3 and follow third-party rounds. A down round at ClickHouse reverses through the income statement. Avride is a $160M annualised EBITDA drag with $1.9M of half-year revenue.
Low. Roughly $2-3B of the $62.5B enterprise value.
Bears are not arguing that demand is weak. They argue three things, and each has evidence behind it.
The prepayment chain is circular. AI labs prepay neoclouds using capital raised on the expectation of AI revenue that does not yet exist. Nvidia sells GPUs to those neoclouds, invests in them, and in Nebius's case guarantees residual capacity elsewhere. Nvidia's own 10-Q discloses $279B of supply commitments, $36B of AI cloud partnership commitments, $25B of equity investment commitments and MOUs to mobilise more than $500B of third-party capital. A stall anywhere in that chain propagates.
Depreciation lives are unresolved. Nebius extended server life from four to five years in Q1-2026. Microsoft runs 2-6 years, Alphabet and Oracle 6, Meta 5 to 5.5, and Amazon shortened a subset from six to five. Michael Burry has argued the cumulative reported-earnings overstatement across major clouds exceeds $176B for 2026-2028 if lives revert to two to three years. The sector is not converging, and a one-year change moves neocloud margins far more than hyperscaler margins because depreciation is a much larger share of the cost base.
The addressable market estimates do not reconcile. Synergy Research sizes the neocloud market at $25B in 2025 heading to roughly $400B by 2031 at a 58% CAGR. Gartner puts neoclouds at 20% of a $267B AI cloud market by 2030, roughly $53B. Those are about eight times apart on the same horizon. Any terminal-value assumption in this sector rests on which one is closer.
What the bears get wrong so far
The 2026 evidence runs against the digestion thesis. Hyperscaler capex rose at all five majors. Nvidia data centre revenue grew 117% year over year and forward supply commitments went from $119B to $279B in a single quarter. GPU rental prices rose rather than fell. No neocloud contract default has been reported. The one confirmed pullback is OpenAI and Oracle ending talks on a 600MW Abilene expansion in March 2026, attributed to interconnection-queue constraints rather than weak demand.
What the bulls get wrong so far
Every bull argument prices the contracts and none prices the delivery risk. Nebius has operated at scale for six quarters. It moved from 170MW active at the end of 2025 to a target of 800MW-1GW connected by the end of 2026 and more than 1GW per year from 2027. It also wrote off $43.6M of property and equipment lost in transit in Q4-2025, which is the kind of thing that happens when a supply chain scales faster than the processes around it.
Assumptions held across all three scenarios unless the scenario table overrides them.
Assumption
Basis
Treatment in the model
Revenue from ARR
ARR is defined by the company as the last month of the quarter multiplied by 12. Calibrated against actuals: FY2025 model output $526M versus $529.8M reported.
Annual revenue equals the sum of monthly ARR over the year, assuming constant monthly compounding between exit ARR points, multiplied by a 0.93 delivery factor.
Segment mix
AI cloud was 98% of Q2-2026 revenue.
AI cloud held at 97.5% of group revenue. Avride and TripleTen modelled as a fixed EBITDA drag of $180M-$260M per year.
Depreciation
Five-year useful life on servers and network equipment, revised from four years in Q1-2026. Capitalised interest of $28.0M at Dec-2025.
Straight line over five years on average gross property and equipment, starting from $14.5B at 30 June 2026.
Share-based compensation
13.9% of LTM revenue, 17.6% in Q2-2026 including the Eigen AI charge.
13% of revenue, excluded from adjusted EBITDA and included in GAAP operating income.
Deferred revenue mechanics
Deferred revenue $5,975.2M at Jun-2026; $9B of prepayments expected in 2026; prepayments cover 50-60% of associated capex.
Prepayments received per scenario; recognition capped at 40% of revenue. Operating cash flow equals adjusted EBITDA plus the change in deferred revenue less cash interest.
Debt and interest
$12,712.5M of convertible face plus $775M secured, after the $800M August exchange. Cash coupons 0.50% to 4.50%; accretion to 110% and 125% on the 2030 and 2034 notes.
2.1% cash interest and 5.2% GAAP interest including accretion. New capital raised 50/50 debt and equity when cash falls below $2.0B.
Share count
271,855,218 outstanding at 30 June 2026 plus ATM sales plus 15.8M from the August exchange equals 289.9M. Plus 21.07M Nvidia pre-funded warrants. Plus 66.0M shares behind $12.7B of converts.
Bull and base assume full conversion by 2029 and extinguishment of convert debt. Bear assumes only the $51.45 series converts and $12.5B stays as debt. All cases add 8% cumulative dilution from share-based compensation.
Exit multiple
Applied to FY2029. Cross-checked on EV/ARR, EV/Sales and EV/EBITDA. CoreWeave trades at 13.0x annualised adjusted EBITDA and 0.75x backlog today; Nebius at 66.1x and 1.56x.
All figures $M except ARR in $B and per-share values in $. "Cov" is customer prepayments as a percentage of that year's capex. FY2026 flows are half-weighted in the cash roll because H1-2026 is already in the 30 June balance sheet.
Scenario / year
Exit ARR $B
Revenue
Adj. EBITDA
Margin
D&A
GAAP op. income
Net income
Capex
Prepay
Cov
Free cash flow
Cash
Bull: pricing holds above $20M ACV per MW, 5GW converts on schedule, prepayments stay above 55%
FY2026
9.0
3,960
1,624
41%
4,900
(3,790)
(4,038)
20,000
9,500
48%
(5,371)
8,386
FY2027
19.5
13,041
6,529
50%
8,500
(3,666)
(4,091)
16,000
11,000
69%
(3,971)
4,415
FY2028
33.0
24,392
12,880
53%
11,500
(1,791)
(2,347)
14,000
9,000
64%
(2,160)
2,255
FY2029
47.0
37,362
20,220
54%
14,100
1,263
636
12,000
7,000
58%
(8)
2,247
Base: pricing settles near $18M ACV per MW, connected power lands within a quarter of plan, prepayments fade with growth
FY2026
8.0
3,650
1,401
38%
4,700
(3,773)
(4,020)
18,000
9,000
50%
(4,671)
9,086
FY2027
16.0
11,047
5,048
46%
7,700
(4,089)
(4,490)
12,000
7,000
58%
(4,654)
4,432
FY2028
24.5
18,884
8,976
48%
9,850
(3,329)
(3,884)
9,500
5,000
53%
(3,361)
2,000
FY2029
32.0
26,409
12,654
48%
11,600
(2,379)
(3,038)
8,000
3,500
44%
(2,702)
2,000
Bear: ARR misses the 2026 guide, pricing reverts to $12M ACV per MW, prepayments dry up and capex is cut to defend liquidity
FY2026
6.5
3,170
1,088
34%
4,450
(3,774)
(4,021)
15,500
7,500
48%
(4,232)
9,526
FY2027
9.8
7,603
2,567
34%
6,650
(5,071)
(5,458)
6,500
3,000
46%
(4,258)
5,268
FY2028
12.0
10,188
3,127
31%
7,650
(5,847)
(6,375)
3,500
1,500
43%
(3,231)
2,037
FY2029
13.0
11,658
3,284
28%
8,250
(6,482)
(7,116)
2,500
1,000
40%
(3,163)
2,000
Note on GAAP losses. All three scenarios show GAAP operating losses through 2028 and only the bull case turns positive in 2029. This is the depreciation ceiling, not a modelling artefact. A business with 48% adjusted EBITDA margins and D&A at 44% of revenue does not produce GAAP profit until revenue outruns the depreciation of the asset base that produced it.
FY2029 exit valuation. Horizon 3.35 years from 26 August 2026. Base share price $213.93.
Scenario
FY2029 ARR $B
FY2029 revenue $M
FY2029 adj. EBITDA $M
Method
Multiple
Implied EV $M
2029 net debt $M
2029 shares M
Value per share
Return
IRR
Bull
47.0
37,362
20,220
EV / ARR
6.0x
282,000
(1,472)
401
$706
+230%
+43%
EV / Sales
7.5x
280,217
$702
+228%
+43%
EV / EBITDA
14.0x
283,077
$709
+231%
+43%
Base
32.0
26,409
12,654
EV / ARR
4.5x
144,000
591
408
$351
+64%
+16%
EV / Sales
5.5x
145,250
$354
+66%
+16%
EV / EBITDA
11.0x
139,199
$340
+59%
+15%
Bear
13.0
11,658
3,284
EV / ARR
2.1x
27,300
12,875
357
$40
-81%
-39%
EV / Sales
2.4x
27,978
$42
-80%
-38%
EV / EBITDA
7.0x
22,985
$28
-87%
-45%
Why the bear case is so much worse than the bull case is good
Convertible debt is the asymmetry. In the bull and base cases the $12.7B of converts turn into 66.0M shares and the debt disappears, so leverage self-liquidates into dilution. In the bear case the stock sits below every conversion price except the $51.45 series, $12.5B stays as debt against $3.3B of adjusted EBITDA (3.8x gross leverage on a business with negative free cash flow), and the equity becomes a residual claim. A $213.93 share price is priced for the base case with meaningful bull-case optionality and almost no discount for the bear.
Weights are illustrative, not derived. At 25/50/25 the expected value is $360 and the implied IRR is 16%. The market price of $213.93 is consistent with roughly 40% bear weight, or with the base case at a 3.0x EV/ARR exit multiple instead of 4.5x.
Value per share ($), FY2029 exit ARR versus EV/ARR multiple. Base-case net debt of $591M and 408M shares held constant.
FY2029 ARR
2.0x
3.0x
4.0x
4.5x
5.5x
6.5x
$13B (bear)
62
94
126
142
174
206
$20B
97
146
195
219
268
317
$26B
126
190
253
285
349
413
$32B (base)
155
234
312
351
430
508
$40B
195
293
391
440
538
636
$47B (bull)
229
344
459
517
632
747
Shaded cells at or below $213.93 are capital losses. The breakeven contour runs from $20B ARR at 4.5x through $32B at 2.8x. Nebius currently trades at 20.8x ARR, so every cell in this table already assumes a substantial de-rating.
Same grid, total return versus $213.93
FY2029 ARR
2.0x
3.0x
4.0x
4.5x
5.5x
6.5x
$13B
-71%
-56%
-41%
-34%
-19%
-4%
$20B
-55%
-32%
-9%
+2%
+25%
+48%
$26B
-41%
-11%
+18%
+33%
+63%
+93%
$32B
-27%
+9%
+46%
+64%
+101%
+138%
$40B
-9%
+37%
+83%
+106%
+151%
+197%
$47B
+7%
+61%
+115%
+142%
+195%
+249%
Over 3.35 years, +64% is a 16% IRR and +142% is a 30% IRR. Roughly half the grid produces a loss, which is the correct read of a business at this stage.
External capital required, mid-2026 through FY2029 ($B). Rows are customer prepayments as a percentage of capex; columns are cumulative capex over the period.
Prepay coverage
$35B
$42B
$47B
$55B
$62B
25%
11.3
16.6
20.3
26.3
31.6
35%
7.8
12.4
15.6
20.8
25.4
45%
4.3
8.2
10.9
15.3
19.2
55% (guided)
0.8
4.0
6.2
9.8
13.0
65%
0.0
0.0
1.5
4.3
6.8
Assumes base-case cumulative revenue of $58.2B and adjusted EBITDA of $27.4B, opening liquidity of $13.8B pro forma for the July facility and August notes, $0.9B of cash interest, a $2.0B minimum cash balance and prepayment revenue recognition at 40% of cumulative revenue. This is the grid that matters most: at 55% coverage and $47B of capex, Nebius needs roughly $6B of new capital over three years, which is one convertible offering. At 35% coverage it needs $15.6B, which is a different company.
FY2029 group adjusted EBITDA ($B): AI cloud margin versus FY2029 revenue
FY29 revenue
28%
35%
42%
50%
56%
$11.7B
3.0
3.8
4.6
5.5
6.2
$18.0B
4.7
5.9
7.2
8.6
9.6
$22.0B
5.8
7.3
8.8
10.5
11.8
$26.4B
7.0
8.8
10.6
12.6
14.2
$32.0B
8.5
10.7
12.9
15.4
17.3
$37.4B
10.0
12.5
15.1
18.0
20.2
Q2-2026 actual AI cloud margin was 49.7%, up from 24% in Q4-2025 and 45% in Q1-2026. Assumes a $220M annual drag from Avride and TripleTen.
FY2029 GAAP operating margin (%) by assumed server useful life. Gross property and equipment of $62.0B on the base-case capex path; adjusted EBITDA margin held at 50%; share-based comp at 13% of revenue.
FY29 revenue
3 years
4 years
5 years (current)
6 years
$11.7B
-143
-99
-72
-54
$18.0B
-80
-52
-34
-23
$22.0B
-59
-36
-22
-12
$26.4B
-43
-24
-12
-4
$32.0B
-30
-13
-4
+3
$37.4B
-20
-6
+2
+8
FY2029 D&A as a percentage of revenue, same inputs
FY29 revenue
3 years
4 years
5 years
6 years
$11.7B
177
132
106
88
$18.0B
115
86
69
57
$22.0B
94
70
56
47
$26.4B
78
59
47
39
$32.0B
65
48
39
32
$37.4B
55
41
33
28
Reverting to a four-year life costs roughly 12 points of operating margin at base-case revenue. Nebius made the opposite move in Q1-2026, extending from four to five years, which flattered 2026 reported earnings by roughly $170M on the H1 asset base. Adjusted EBITDA is unaffected, which is precisely why the sector reports on adjusted EBITDA.
Each row moves one variable and holds the rest at base case.
Variable
Downside setting
Downside value
Base
Upside setting
Upside value
Swing
FY2029 exit ARR
$24.5B
$269
$351
$40B
$440
$171
EV / ARR exit multiple
3.5x
$273
$351
5.5x
$430
$157
Share count in 2029
+15% dilution
$306
$351
-5%
$370
$64
FY2029 net debt
+$8B
$332
$351
-$3B
$359
$27
Volume and multiple dominate. Dilution and leverage matter roughly one third as much, which is the counter-intuitive result: with $13.8B of pro forma liquidity and cheap convertible funding, the balance sheet is not the near-term risk. The risk is that ARR growth and the multiple compress together, which is exactly what they do in a funding squeeze.
Q4-2026 exit ARR below $6.5B. The $7-9B guide has been reaffirmed twice. A miss would be the first, and it would reset the entire ARR ladder that every scenario is built on.
Prepayment coverage falling below 45% of capex on new deals. Disclosed quarterly in the shareholder letter. This is the cleanest single indicator, and it leads revenue by two to four quarters.
ACV per MW on new deals falling back toward $12-15M. Would restore the two to three year payback and cut the return on the $47B capex programme by roughly a third.
Connected power materially short of 800MW at year end. Would signal that contracted power is not converting, and 5GW becomes an option that expires.
A funding round at ClickHouse below $15B, or any writedown in the $1,606.7M of non-marketable equity securities.
A second secured facility at or inside SOFR + 250bp. Nebius flagged more asset-backed financing against its $40B of commitments. Terms at or better than July's would confirm the funding advantage is structural.
Disclosure of customer concentration below 30% for the top customer. Currently not disclosed. Would remove the largest unquantified risk in the file.
Positive free cash flow before 2029. No scenario here produces it in the base case. Achieving it would mean prepayments or margins beat by a wide margin.
Short-term capacity deals clearing at $40-50M per MW at scale. One signed in Q3. A second and third would re-rate the pricing assumption in every year of the model.
Sustained GPU rental price increases into 2027. Prices rose 7% to 14% in the recent period; continuation would break the commoditisation thesis outright.
Primary SEC filings, all retrieved via sec-api.io (CIK 1513845). Form 20-F for FY2025, filed 30 April 2026, accession 0001104659-26-052948, and Form 20-F/A filed 22 May 2026, accession 0001104659-26-065681. Form 20-F for FY2024, filed 30 April 2025, accession 0001558370-25-005991. Form 6-K Q2-2026 results, filed 12 August 2026, accessions 0001104659-26-094568 (press release and shareholder letter) and 0001104659-26-094844 (operating and financial review, condensed financial statements). Form 6-K Q1-2026, filed 20 May 2026, accession 0001104659-26-064092. Form 6-K Q4 and FY2025 results, filed 12 February 2026, accession 0001104659-26-013946. Form 6-K convertible note offering announcement and pricing, filed 19 August 2026, accessions 0001104659-26-098590 and 0001104659-26-098924. Form 6-K note settlement documents, filed 24 August 2026, accession 0001104659-26-100347. Form 6-K annual general meeting results, filed 26 August 2026, accession 0001104659-26-101076.
Peer filings. CoreWeave 10-Q Q2-2026 (0001769628-26-000366) and 8-K results release (0001769628-26-000362), both 11 August 2026; FY2025 10-K (0001769628-26-000104). IREN 10-Q for the quarter ended 31 March 2026 (0001878848-26-000026) and 8-K of 13 August 2026 (0001140361-26-032638). Applied Digital FY2026 10-K, 29 July 2026 (0001144879-26-000048). Cipher Digital 10-Q Q2-2026, 4 August 2026 (0001819989-26-000041). TeraWulf 10-Q Q2-2026, 5 August 2026 (0001083301-26-000166).
Hyperscaler and supply-chain filings. Microsoft FY2026 10-K (0001193125-26-323660). Amazon FY2025 10-K (0001018724-26-000004) and Q2-2026 10-Q (0001018724-26-000026). Alphabet FY2025 10-K (0001652044-26-000018) and Q2-2026 10-Q (0001652044-26-000071). Meta FY2025 10-K (0001628280-26-003942) and Q2-2026 10-Q (0001628280-26-050705). Oracle FY2026 10-K (0001193125-26-277521). Nvidia Q2 FY2027 10-Q, filed 26 August 2026 (0001045810-26-000075), and Form 13F-HR filed 14 August 2026 (0001045810-26-000065).
Non-filing sources, each carrying its own reliability. Share prices and market capitalisations from stockanalysis.com as of 26 August 2026, except Applied Digital which is 21 August 2026 from Yahoo Finance. GPU rental pricing from Silicon Data, Thunder Compute and GPUSmith aggregations, which are commercial indices and not audited. Market sizing from Synergy Research (2 April 2026) and Gartner (23 June 2026), which do not reconcile. CoreWeave CDS and loan repricing from Bloomberg and Reuters coverage dated 28 July and 3 August 2026. Power and interconnection data from PJM via Canary Media, Carbon Direct (May 2026) and Data Center Knowledge (12 May 2026). Hyperscaler guidance from earnings calls dated 10 June, 22 July, 29 July and 30 July 2026. Colour palette anchored on sec-api.io's declared theme colour #2b6aca; the remaining tones were derived, not sampled.
Not verified. Nebius does not disclose customer concentration percentages. Q3-2025 and Q4-2025 capital expenditure are company approximations from the shareholder letters, not audited line items. Q2-2025 exit ARR of $0.43B is implied from a disclosed growth rate rather than reported directly. FY2025 depreciation and amortisation is $417.9M per the audited 20-F versus $404.0M in the February 2026 shareholder letter; the 20-F figure is used throughout. FY2026 revenue, adjusted EBITDA and capital expenditure guidance were given on earnings calls rather than in the filings, so the only guided figures used here are the year-end ARR range of $7-9B, contracted power of 5GW, connected power of 800MW to 1GW and customer prepayments above $9B, all of which appear in the written shareholder letters. IREN has not filed FY2026 results; its data is five months stale relative to the calendar-quarter peers.
Nothing in this note is investment advice. The scenarios are a model, not a forecast, and every forward figure is an assumption stated in section 7.1 rather than a company projection.
Nebius Group N.V. (NASDAQ: NBIS) · Financial analysis, ratios, scenarios and sensitivity · Prepared 26 August 2026 · Filing data via sec-api.io
