Nebius Group (NBIS $171.45 +0.71%) reports Q3 2026 results in mid-November; the company has not confirmed a date yet, and last year's Q3 came on November 11. The stock goes into the print down 25.2% in a week and 42% below its June closing high, after CoreWeave (CRWV $62.13 +1.65%)'s debt scare spread to every company that rents out GPUs. We rebuilt Nebius's quarterly model from its filings and earnings calls to see what Q3 and Q4 are likely to show.
Our answer is bullish. Q3 revenue should beat consensus, because the June exit run-rate alone covers most of the quarter. Full-year revenue lands at the top of guidance even on a conservative ARR forecast, and any year-end ARR above about $7.9B would push 2026 revenue past the top of the range. Meanwhile the sell-off priced Nebius like CoreWeave, a company with a very different balance sheet, while $37.5B of contracted backlog points to a much larger 2027.
The data behind this piece
Explore the primary data Equibles carries for NBIS:
Why Q3 should beat consensus
Revenue figures are continuing operations, restated for the 2025 deconsolidation of Toloka, so every year adds up to the full-year totals in the company's annual report.
Since 2026, Nebius defines ARR as AI cloud revenue in the last month of the quarter multiplied by 12. That makes ARR a monthly revenue figure in disguise:
- March 2026: ARR of $1.92B, so about $160M of AI cloud revenue in March.
- June 2026: ARR of $3.0B, so about $250M in June.
- Q2 AI cloud revenue was $574.9M, about $192M a month on average. June ran 30% above the quarter's average, because most of the new capacity arrived late in the quarter.
If AI cloud revenue simply stayed at the June level, Q3 would bring in $750M. On top of that come the two Microsoft tranches delivered between mid-July and mid-August, and other capacity that the CFO said was "deployed late in Q2" and would "begin contributing to revenue in Q3". We model monthly growth of 9% to 12% through Q3, slower than the roughly 16% a month Nebius averaged from March to June.
Our monthly AI cloud revenue path
| $M | Mar | Jun | Jul F | Aug F | Sep F | Oct F | Nov F | Dec F |
|---|---|---|---|---|---|---|---|---|
| AI cloud revenue | 160 | 250 | 280 | 305 | 342 | 380 | 440 | 600 |
| ARR (month x 12, $B) | 1.92 | 3.00 | 3.36 | 3.66 | 4.10 | 4.56 | 5.28 | 7.20 |
March and June are implied by the reported ARR; every other month is our estimate. Our September figure gives a Q3 exit ARR of $4.1B.
The upside case is simple: growth does not have to slow. At the March to June pace of 16% a month, AI cloud revenue would run at about $290M in July, $336M in August and $390M in September. That is about $1,024M of Q3 revenue, 13% above consensus and above the top of the analyst range, with a Q3 exit ARR of about $4.7B.
Our forecast for Q3 and Q4 2026
Q1 and Q2 are reported. Q3 and Q4 are our estimates. Dollar figures are in millions.
| $M | Q1 2026 | Q2 2026 | Q3 2026F | Q4 2026F | FY 2026F | FY 2026 guidance |
|---|---|---|---|---|---|---|
| AI cloud revenue | 389.7 | 574.9 | 927.0 | 1,420.0 | 3,311.6 | |
| Other segments, net of eliminations | 9.3 | 7.4 | 7.5 | 7.5 | 31.7 | |
| Group revenue | 399.0 | 582.3 | 934.5 | 1,427.5 | 3,343.3 | 3,000 to 3,400 |
| Analyst consensus revenue | 904.8 | 1,456.5 | 3,340.1 | |||
| ARR at period end ($B) | 1.92 | 3.0 | 4.1 | 7.2 | 7.2 | 7 to 9 |
| Adjusted EBITDA | 129.5 | 236.2 | 373.8 | 599.6 | 1,339.1 | |
| Adjusted EBITDA margin | 32% | 41% | 40% | 42% | 40.1% | about 40% |
| Capital expenditure | 2,472.9 | 5,657.4 | 6,500.0 | 7,870.0 | 22,500 | 20,000 to 25,000 |
Other segments are TripleTen, the education business, and Avride, the autonomous driving unit, less intercompany eliminations. Consensus is the average of 17 analysts for each quarter and 21 for the full year, captured on October 18. We do not forecast earnings per share: in the first half alone, a $780.6M non-cash revaluation of the ClickHouse stake and a $115.9M acquisition charge moved net income more than operations did.
ARR upside is revenue upside
Nebius has held both 2026 targets since February and reaffirmed them "across all metrics" in August: $3.0B to $3.4B of revenue and $7B to $9B of year-end ARR. Because ARR is December's AI cloud revenue multiplied by 12, the two targets are linked. Every extra $1B of year-end ARR is about $83M more revenue in December, and that revenue also counts toward 2026.
Holding October and November at our $380M and $440M, here is what each year-end ARR outcome means for full-year revenue:
| Year-end ARR | December AI cloud revenue ($M) | Q4 group revenue ($M) | FY 2026 revenue ($M) | Against the $3.4B top of guidance |
|---|---|---|---|---|
| $7.2B (our base case) | 600 | 1,427.5 | 3,343.3 | Inside, near the top |
| $7.5B | 625 | 1,452.5 | 3,368.3 | Inside, near the top |
| $8B | 667 | 1,494.2 | 3,410.0 | Above |
| $8.5B | 708 | 1,535.8 | 3,451.6 | Above |
| $9B | 750 | 1,577.5 | 3,493.3 | Above |
The break-even point is a year-end ARR of about $7.9B. Anything above it, which is the upper half of management's own ARR range, puts 2026 revenue above guidance. The revenue range is a forecast, not a ceiling.
We keep $7.2B as our base case because the size of the December step depends on the exact day new capacity starts billing. But the direction of the risk is up. Management said the Q2 contracts, signed at more than $20M of revenue per megawatt, "start coming online from late Q4 and onwards", and the first short-term premium contract goes live in Q4. Each block that starts billing early in December lifts both ARR and 2026 revenue.
Even our cautious base case meets both targets: revenue near the top of its range and ARR inside its range. A print in the lower half of the ARR range would reflect when capacity went live, not weaker demand, and capacity that starts late in December bills for full months from January.
ARR before 2026 was defined on total revenue in the last month of the quarter, not AI cloud revenue alone. By the end of 2025 AI cloud was about 94% of revenue, so the change makes little difference to the trend.
Margins: running ahead of plan
On the Q1 call in May, the CFO guided to a lower margin in Q2, a return to Q1 levels in Q3, and a higher margin in Q4, because new capacity carries start-up costs before it bills. Q2 broke that pattern: group adjusted EBITDA margin rose from 32% to 41%, and AI cloud margin from 45% to 49.7%.
We model 40% for the group in Q3, a point below Q2, because Q3 carries a full quarter of costs for capacity that is still ramping. Q4 rises to 42% as the October 1 list-price increase and the first short-term premium contract start contributing. Nebius raised on-demand prices by 17% to 21% on October 1; the H100 moved from $3.85 to $4.50 an hour, and the B300 from $7.85 to $9.50. On-demand is a small part of revenue, which is why we add only one point.
Pricing on contracted capacity matters more. The four large contracts signed in Q2, with Reflection, Cohere, an unnamed U.S. AI lab and a U.S. quantitative trading firm, averaged more than $1B each at $20M to $25M of annual revenue per megawatt. About 70% of the deals included prepayments covering 50% to 60% of the related capital spending. Management says most of that revenue arrives in 2027.
2027 is where the backlog lands
The bigger story is next year. Most of what Nebius signed in 2026 bills from 2027, and the contracted backlog already dwarfs this year's revenue.
| Contract | Value | Timing | Status |
|---|---|---|---|
| Microsoft, Vineland, New Jersey | ~$17.4B over 5 years | Tranches through 2026 | "Delivered all capacity tranches … to date" (August 12) |
| Meta, first order | ~$2.9B over 5 years | Delivered December 2025 to Q1 2026 | Fully in service |
| Meta, second agreement | Up to ~$27B | $12B dedicated from early 2027; up to $15B of capacity Meta must buy if unsold | On track for early 2027 |
| Four Q2 contracts | >$1B each on average | Capacity arriving late 2026 | Mostly 2027 revenue |
| First short-term capacity deal | Not disclosed | Goes live in Q4 2026 | Priced at a premium |
Remaining performance obligations, the contracted revenue not yet recognized, reached $37.5B at June 30, up from $21.3B at the end of 2025. Of that, 36%, about $13.5B, is expected within 24 months, roughly four times our 2026 revenue forecast. The $37.5B equals about 73% of the market capitalization, and most of it bills from 2027, when the second Meta agreement's $12B of dedicated capacity also starts. Analysts expect 2027 revenue of $12.29B, more than three and a half times 2026. Three customers, which the filing does not name, made up 24%, 21% and 14% of Q2 revenue: 59% from three buyers.
Not CoreWeave: the balance sheet behind the build
Nebius guided to $20B to $25B of 2026 capital expenditure, up from $16B to $20B in May. It spent $8.13B in the first half, so the second half needs $11.9B to $16.9B. Our model uses the $22.5B midpoint, and treats the guide as cash purchases of property and equipment.
Second-half operating cash flow of $5.5B is our estimate. It rests on customer prepayments: Nebius expects "over $9 billion" of them in 2026, and deferred revenue rose by $4.4B in the first half, which leaves at least $4.6B for the second half. Year-end cash of about $5.6B excludes the 12.3M shares left under the at-the-market program, worth about $2.0B at $165.47, and any new project debt.
Debt and convertible notes, pro forma for July and August
| Instrument | Principal ($M) | Coupon | Maturity | Conversion price |
|---|---|---|---|---|
| June 2025 notes, after the August exchange | 200.0 | 2.00% / 3.00% | 2029 / 2031 | $51.45 |
| September 2025 notes | 3,162.5 | 1.00% / 2.75% | 2030 / 2032 | $138.75 |
| March 2026 notes | 4,337.5 | 1.25% / 2.625% | 2031 / 2033 | $183.22 / $180.31 |
| August 2026 notes | 5,750.0 | 0.50% / 4.50% | 2030 / 2034 | $313.46 / $324.65 |
| Senior secured term loan | 775.0 | SOFR + 2.50% | October 2030 | n/a |
| Total | 14,225.0 |
Most of the notes accrete to between 110% and 125% of face value at maturity, so the cost is higher than the coupons suggest. The August notes add about $121M a year of cash coupons; interest expense was already $119.1M in Q2. That is a very different position from CoreWeave's: about $50B of debt, a record cost of default insurance on October 15, and bonds yielding above 16%. Nebius's debt is mostly low-coupon convertible notes, its gross debt was covered by cash at June 30 pro forma for the new financing, and customers prepay much of the build. The CFO also pointed out that Nebius has "almost no corporate-level debt", which leaves room to borrow as the business scales.
Share count and dilution
| Shares (M) | Count |
|---|---|
| Class A and Class B outstanding, June 30 | 271.9 |
| Issued in the August note exchange | 15.8 |
| NVIDIA pre-funded warrant (March, $2B) | 21.1 |
| Basic share base | 308.8 |
| Convertible notes in the money at $165.47 (June and September 2025 series) | 26.7 |
| Convertible notes out of the money (March and August 2026 series) | 41.9 |
| Remaining at-the-market program | 12.3 |
| Total if everything converts or is sold | 389.7 |
Note shares use the initial conversion rates. Nebius sold 12.7M shares through the at-the-market program in Q2 at an average of $223.60, raising about $2.8B; any sales after June 30 are not in the table.
How the market is positioned: a sell-off from outside
The October 12 to 16 sell-off came from outside the company. Nebius published no news that week. CoreWeave fell by the same 25.2%, as its bonds traded toward distressed levels and a hot inflation print revived expectations of rate hikes. D.A. Davidson's Gil Luria told CNN that the neocloud companies are "the subprime tranche of the AI trade, and that's why they're selling off first."
What the positioning data shows:
- Estimates did not move. The Q3 revenue consensus was $907.6M on September 18 and $904.8M on October 18. Analysts left their numbers unchanged through a 34% drop from the October 6 close.
- Short interest is below its peak. 44.7M shares were short on September 30, 17.6% of Class A shares, with 2.71 days to cover. That is 27% below the July 15 peak of 61.1M, although still 86% higher than a year earlier. Settlement data for mid-October is due on October 26.
- The worst day was not driven by short sellers. On October 16 short sales were 39.3% of reported volume, the lowest daily reading since mid-September.
- Institutions piled in, with hedges. The number of 13F filers holding Nebius rose from 733 to 1,069 in Q2, with 407 new positions, including Lone Pine, Situational Awareness and Norges Bank. Capital Research and Janus Henderson sold out completely.
- NVIDIA's 9.3% stake is the warrant. Its July 13G reported 22.26M shares, against 1.19M on its 13F. The difference is the 21.07M-share pre-funded warrant NVIDIA bought for $2B in March.
- Insiders are sellers. Insiders sold about $318M of stock in 2026, including the COO's 500,000 shares on October 5 at $231 to $244, under a 10b5-1 plan. The only open-market purchase was 296 shares, about $50,000, bought by director Arne Grimme on October 15 and 16.
Put exposure on 13F filings doubled in Q2 to 50.3M shares, and the put-to-call ratio rose from 0.90 to 1.52 in three quarters. Most of it sits with market makers and multi-strategy funds, so it reflects hedging as much as bearish bets, but puts grew faster than calls while long holders were adding.
| Key data | Value |
|---|---|
| Share price, October 16, 2026 close | $165.47 |
| 52-week closing range | $73.87 to $286.69 |
| Change, year to date | +97.7% |
| Shares: Class A and Class B plus the NVIDIA warrant | 308.8M |
| Market capitalization, on that share count | ~$51.1B |
| Gross debt, pro forma | $14.2B |
| Cash, June 30 pro forma for July and August financing | $14.5B |
| Enterprise value | ~$50.8B |
| Enterprise value / FY 2026 consensus revenue ($3.34B) | 15.2x |
| Enterprise value / FY 2027 consensus revenue ($12.29B) | 4.1x |
| Customer prepayments (deferred revenue), June 30 | $6.0B |
| Remaining performance obligations, June 30 | $37.5B |
Market capitalization here counts both share classes, the August exchange shares and NVIDIA's warrant, so it is higher than a figure based on Class A shares alone.
At about 4.1x 2027 consensus revenue, the price assumes little of the step-up that the backlog already contracts. Analyst targets run from Rothschild & Co Redburn's $84, with a Sell rating since September, to Northland's $410. The full-year 2027 revenue consensus of $12.29B spans $8.8B to $18.0B, so the range of outcomes is still wide.
What management said
“We closed our largest AI Cloud deals on our strongest terms to date, at prices that represent a step-change in the economics of our business.”
Arkady Volozh, founder and CEO, Q2 2026 shareholder letter
“We could sell our entire 2027 capacity on these terms today. We are deliberately not doing so because we see higher value in retaining some capacity for immediate customer needs.”
Arkady Volozh, founder and CEO, Q2 2026 shareholder letter
“We expect over $9 billion in customer prepayments in 2026. We are reaffirming our full-year 2026 guidance across all metrics.”
Arkady Volozh, founder and CEO, Q2 2026 shareholder letter
“The majority of these deals were signed against capacity arriving in late 2026, and will contribute primarily to 2027 revenue.”
Nebius, Q2 2026 shareholder letter
“To date, we've delivered all capacity tranches to Microsoft under the contract, including two tranches over the past month.”
Nebius, Q2 2026 shareholder letter
“We continue to expect annualized run rate revenue of $7 billion to $9 billion, group revenue of between $3 billion and $3.4 billion, group adjusted EBITDA margin of approximately 40%, and capital expenditures of $20 billion to $25 billion. And we remain confident in our ability to accelerate capacity deployment in the second half of the year and expect capacity deployed late in Q2 to begin contributing to revenue in Q3.”
Dado Alonso, CFO, Q2 2026 earnings call
“We're negotiating deals for $40 million to $50 million per megawatt range, and sometimes above, under this model. One of such deals has been signed just recently by the way. Both deal terms I just described are all coming online later this year. So they will not have a material effect on our 2026 revenue guidance.”
Arkady Volozh, founder and CEO, Q2 2026 earnings call
“You should expect the deals we closed in this quarter with more than $20 million per megawatt deals and less than 2-year payback period to start coming online from late Q4 and onwards.”
Dado Alonso, CFO, Q2 2026 earnings call
“The dynamics we expect across both capacity and pricing make us extremely excited about 2027, although we will provide a formal guidance later this year.”
Dado Alonso, CFO, Q2 2026 earnings call
“[W]e still expect to meet this guidance from 800 megawatts to 1 gigawatt of connected power this year.”
Andrey Korolenko, Chief Infrastructure and Product Officer, Q2 2026 earnings call
“We currently have almost no corporate-level debt, so that represents an additional source of capital that we can evaluate as the business scales.”
Dado Alonso, CFO, Q2 2026 earnings call
“These investments unlock growth by increasing capacity substantially from Q2 to Q3, leading us to be confident in our adjusted EBITDA margin returning to Q1 levels in Q3 before moving even higher in Q4.”
Dado Alonso, CFO, Q1 2026 earnings call
The Q1 call quote comes from a speech-to-text transcript; the Q1 2026 earnings call is on Equibles.
Our assumptions
| Assumption | Q3 2026F | Q4 2026F | Basis |
|---|---|---|---|
| AI cloud revenue, first month of quarter ($M) | 280 | 380 | June exit of ~$250M plus July Microsoft tranches |
| AI cloud revenue, second month ($M) | 305 | 440 | Late-Q2 capacity ramping |
| AI cloud revenue, third month ($M) | 342 | 600 | Q2 contracts start billing in December |
| Exit ARR ($B) | 4.1 | 7.2 | Third month x 12 |
| Other segments, net of eliminations ($M) | 7.5 | 7.5 | Q2 run-rate |
| Group adjusted EBITDA margin | 40% | 42% | Q2 was 41%; FY guide about 40% |
| Capital expenditure ($M) | 6,500 | 7,870 | $22.5B FY midpoint less $8.13B first half |
| Second-half operating cash flow ($M) | 5,500 for H2 | At least $4.6B of prepayments plus EBITDA, less interest | |
| August convertible notes, net proceeds ($M) | 5,680 | Company estimate with full option exercise | |
| Secured term loan ($M) | 775 | Closed July 2026 | |
| New at-the-market share sales | none | none | 12.3M shares remain available |
| Price increase on October 1 | n/a | +1 point of margin | On-demand only, small share of revenue |
What could break the bull case
- Capacity timing. Every month a data hall slips moves revenue out of 2026. Our Q4 depends on Q2 contracts billing from December.
- Customer concentration. Three customers made up 59% of Q2 revenue.
- Funding costs. Credit markets are repricing GPU-backed borrowers. CoreWeave's bonds yield above 16%, and the next Nebius debt deal will show what lenders charge.
- Dilution. The at-the-market program and in-the-money notes add shares at the current price.
- Pricing durability. Today's GPU prices reflect a shortage. Market rates for H100s are around $2.80 to $3.60 an hour, below Nebius's new $4.50 list price.
- Delivery penalties. The Microsoft contract allows liquidated damages for late delivery, and Microsoft can terminate a tranche after a grace period.
Catalysts we are watching
- Q3 results: revenue against the $904.8M consensus, and Q3 exit ARR, where our estimate is $4.1B and the upside case is $4.7B.
- Whether the $7B to $9B ARR guide is narrowed, and where: with our October and November path, a midpoint of $7.9B or more would point to revenue above the current guide.
- Formal 2027 guidance, which the CFO said would come "later this year".
- Connected power against the 800 MW to 1 GW year-end target.
- New debt financing, and its price compared with CoreWeave's.
- Short interest for the October 15 settlement, published October 26.
- The second Meta agreement coming online in early 2027.
Sources
- Q2 2026 results, shareholder letter and guidance: 6-K filed August 12, 2026 (accession 0001104659-26-094568)
- Q2 2026 interim financial statements, segments, backlog, convertible notes and share sales: 6-K filed August 12, 2026 (accession 0001104659-26-094844)
- Q1 2026 results and letter: 6-K filed May 13, 2026
- Q4 and full-year 2025 results: 6-K filed February 12, 2026
- Annual report for 2025: 20-F filed April 30, 2026
- NVIDIA $2B investment: 6-K filed March 11, 2026
- Meta agreement of up to $27B: 6-K filed March 16, 2026
- Microsoft agreement: 6-K filed September 8, 2025
- Senior secured term loan: 6-K filed July 17, 2026
- August convertible notes and note exchange: 6-K filed August 20, 2026 and 6-K filed August 24, 2026
- GPU list prices: Nebius pricing page
- Q2 2026 earnings call, August 12, 2026 (public transcript of the company webcast)
- Q1 2026 earnings call transcript, share prices, analyst consensus, short interest, short volume, 13F and 13G holdings and insider trades: Equibles
- Forecast and assumptions: Equibles Research model, October 19, 2026
This article is our analysis of public filings and our own financial model. It is not investment advice, and forecasts can be wrong. Equibles staff may hold positions in the securities discussed.