The Nebius stock prediction still circulating is a September ARR multiple, and that model is not what the next 49 sessions are pricing. Nebius closed Wednesday, 21 October 2026, at $171.57. The last 90 Nasdaq log-returns put a bull at $322.33, a base at $145.45 and a bear at $85.57 through 31 December 2026. The bull is $2.33 above the older $320 figure. That is a different close and a different model, not a reused target, and the bear is $85.57 rather than $122.
What the coverage misses is the sign of the drift. The mean log-return is negative, so the base sits 15.2 percent under the close, while the 12 August gap from $193.23 to $259.20 stays inside the sample and keeps the bull wide. The investor-events page, fetched 22 October, lists no upcoming event and no third-quarter release. The closes used here are Wednesday’s completed prints. A partial Thursday bar is not the spot.
Key facts
- Wednesday 21 October 2026 official close: $171.57, down 27.7 percent from the 25 September close of $237.33 in the same file. Source: Nasdaq historical prices, saved 22 October 2026, Nasdaq.
- The 20-day average is $214.57 and the 50-day average is $221.76. The close is 20.0 percent under the first and 22.6 percent under the second. Prior five-session range: $160.25 to $198.60. Same Nasdaq file.
- Second-quarter group revenue was $582.3 million, up 454 percent. AI cloud revenue was $574.9 million, about 98 percent of the group, and ARR was $3.0 billion at 30 June. Source: shareholder letter, 12 August 2026, SEC exhibit 99.2.
- CFO Dado Alonso reaffirmed $7 billion to $9 billion of year-end ARR, revenue of $3 billion to $3.4 billion, an adjusted EBITDA margin of about 40 percent, and capital expenditure of $20 billion to $25 billion. Source: 12 August 2026 call, The Motley Fool.
- Second-quarter capital expenditure was about $5.7 billion against operating cash flow of $2.3 billion. Cash was $8 billion at 30 June. GAAP net loss was $190.4 million. Same call and letter.
- Will Sommer of Gartner said AI providers are expected to spend nearly $1 trillion on infrastructure in 2026 and $6 trillion more from 2027 to 2030. Source: 14 October 2026, The Register.
- HUMAIN and DataVolt said Nebius would take up to 50 megawatts at Oxagon, half of the first 100 megawatts under construction, phased from the fourth quarter of 2027. No contract value. Source: 21 October 2026, ZAWYA and HUMAIN on X.
What just happened, and why the obvious reading is wrong
The obvious reading is that the contract story broke. The tape does not say that. From $237.33 on 25 September to $171.57, Nebius fell $65.76, or 27.7 percent. It is also 40.2 percent under the 18 June close of $286.69, the high close in this file, and still 105.0 percent above the 31 December 2025 close of $83.705.
Wednesday’s bar was ordinary. The open was $176.46, the high $178.85, the low $164.00, and the close $171.57 on 19,693,930 shares, $0.12 above Tuesday. The violent print in the window was Friday 16 October, from $188.41 to $165.47. The 14 October close of $186.01 is the session FinanceFeeds reported when the Gartner wording and a CoreWeave debt scare hit together.
Both averages are overhead. The close is $43.00 under the 20-day line of $214.57 and $50.19 under the 50-day line of $221.76. It is $11.32 above the prior five-session low of $160.25 and $27.03 under the prior five-session high of $198.60.
The 6 October close was $249.87, the print in the Astera, Nebius and CoreWeave tape note. From there to $171.57 is $78.30, or 31.3 percent, across the eleven sessions that followed. Nebius did not cut the $7 billion to $9 billion range in those sessions. The newsroom’s latest results item is still the 12 August release. The price moved. The guidance line on the company site did not.
Who is exposed
The shareholder letter names the book that closed in the second quarter. Four AI-cloud deals averaged more than $1 billion of total contract value each, at $20 million to $25 million of annual contract value per megawatt. Reflection and Cohere are named. A US lab and a US quantitative trading firm are not.
Microsoft and Meta are delivery dates, not fresh dollar totals. The letter says every Microsoft tranche under the contract has been delivered, including two in the month before 12 August, and that a second Meta agreement was on track for early 2027.
Nvidia is the supplier. Nebius had its first Vera Rubin NVL72 racks in test. On 21 October, Nico Caprez, VP of AI infrastructure growth at Nvidia, said “NVIDIA AI infrastructure is productive, durable and fungible across models, workloads and customers.” Euro Beinat, chief AI officer at Prosus, said: “With Nebius Token Factory, we have doubled throughput, while autoscaling helps us handle peaks without holding GPU capacity in reserve.” Both lines are in the 21 October release.
Oxagon points the other way. Nebius would be the customer. HUMAIN chief executive Tareq Amin said: “The real measure of AI infrastructure is not how much capacity you build, but the demand you can secure and the value you can deliver.” Up to 50 megawatts, inside a 360-megawatt first phase, would start in phases from the fourth quarter of 2027. There is no lease and no price. Finnish AI Region reported a Caverion substation of about 15 million euros for a new 70-megawatt hall beside the 75-megawatt Mäntsälä site, aimed at autumn 2027. Neither project changes Wednesday’s close.
What the tape and the letter actually show
The chart uses only the Nasdaq official daily closes saved for this piece. The last bar is Wednesday 21 October 2026. The 20-day and 50-day lines are simple averages of that file. The lower panel draws the bull, the base and the bear from the formula below, not from a broker target.
| What the tape shows | What the primary document shows |
|---|---|
| Close $171.57 on 21 October, 27.7 percent below $237.33 on 25 September | Revenue $582.3 million in the second quarter, up 454 percent. ARR $3.0 billion at 30 June. Letter, 12 August 2026 |
| Close 20.0 percent under the 20-day average and 22.6 percent under the 50-day average | Alonso reaffirmed $7 billion to $9 billion of year-end ARR and $3 billion to $3.4 billion of 2026 revenue |
| Prior five-session range $160.25 to $198.60. Wednesday finished inside it | About 70 percent of second-quarter deals included prepayments, covering 50 to 60 percent of the tied capital expenditure |
| 12 August close jumped from $193.23 to $259.20. That log-return is 17.3 percent of the sum of squared deviations | Most of those deals were signed against late-2026 capacity and will contribute primarily to 2027 revenue |
| Bull $322.33, base $145.45, bear $85.57 from this formula, not from an analyst note | Four deals averaged more than $1 billion of total contract value, at $20 million to $25 million per megawatt |
Read the rows across. The left column is a falling price. The right column is a contract book the company has not withdrawn. The letter says those large deals mostly hit revenue in 2027, so a lower October price is not, on these documents, a cancelled Microsoft tranche. The second-quarter results note is the session that put the 12 August gap on the tape. The calculation keeps that day.
What the company has said, and what it has not
Arkady Volozh, founder and chief executive, wrote on 12 August: “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.” The letter adds that Nebius could sell its entire 2027 capacity on those terms and is deliberately not doing so, because it wants capacity left for immediate demand.
On 21 October, in a clip Fireside Alpha posted, Volozh said: “Yes, the prices are still there, demand is still there.” He called financing, building and chips tasks, and said missing demand would be the problem. A second clip says preemptible capacity launched the day before and “the prices went up.”
Alonso put the numbers on the reaffirmation. “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.” Nebius has not posted a third-quarter release. The events page still says there are no upcoming events. Outside calendars have pointed at 22 October. They are not the company’s date, and they are not an input to the bull or the bear.
The financing tension
Operating cash flow in the second quarter was $2.3 billion against capital expenditure of about $5.7 billion. Cash at 30 June was $8 billion. The firm was spending ahead of operations, against a $20 billion to $25 billion full-year budget. The letter expects more than $9 billion of customer prepayments in 2026, which is 36 to 45 percent of that budget. Upfront cash covered 50 to 60 percent of the capital expenditure on the second-quarter deals. The rest is the at-the-market program, asset-backed debt and corporate debt.
Through 30 June, Nebius had sold 12.7 million Class A shares at a weighted average of $223.60, about $2.8 billion gross, with 12.3 million left. At $171.57 those remaining shares would raise about $2.11 billion, against about $2.75 billion at $223.60. The gap is about $640 million, roughly 23 percent less per share still in the program.
In July the company raised $775 million at SOFR plus 2.50 percent, secured by deployed GPUs and an investment-grade contract. The newsroom lists a later closing of convertible notes with aggregate gross proceeds of approximately $5.75 billion. The letter also moved server and network equipment to a five-year useful life, from four years. Depreciation and amortization were $259.7 million, 45 percent of revenue, larger than group adjusted EBITDA of $236.2 million. The GAAP net loss was $190.4 million. The transcript ties the year-ago profit to a $597.4 million revaluation gain that did not repeat.
Sommer told The Register that datacenter operators would need $8.5 trillion of revenue through 2030 to cover spending, financing and a 12 percent return, against no more than $3.6 trillion of attributable AI revenue. He also said: “Not everyone is in trouble. Some will outperform, some will underperform, and some will have to exit the market.” A five-year life lowers today’s depreciation. It does not create the 2027 revenue the letter has already deferred.
Bull, base and bear
The title numbers are this distribution. They are not a street target, and they are not the September FinanceFeeds pair of a $320 bull, a $224 base and a $122 bear, which was an ARR multiple on the $237.33 close. No broker target is imported into the headline.
The inputs are the last 90 daily log-returns in the saved file, ending Wednesday. The mean, mu, is -0.00336996. The sample standard deviation, sigma, divides by 89 and is 0.07578236. N is 49 because the close is 21 October 2026 and the horizon runs through 31 December. The square root of 49 is 7.
Those 49 sessions are a calendar. October has seven weekday sessions left after Wednesday: 22, 23, 26, 27, 28, 29 and 30 October. November has 20 if Thanksgiving on 26 November is out and the 27 November early close is kept, including 30 November. December has 22 if Christmas on 25 December is out and the 24 December early close is kept. Seven plus 20 plus 22 is 49. Early closes count. Full closures do not.
The base compounds the mean only. $171.57 times e to the power of 49 times -0.00336996 is $171.57 times e to the power of -0.165128, which rounds to $145.45, 15.2 percent under the spot. The bull adds 1.5 standard deviations. The exponent is -0.165128 plus 1.5 times 0.07578236 times 7, or 0.630587, and the price rounds to $322.33, 87.9 percent above the spot. The bear subtracts one standard deviation. The exponent is -0.695605, and the price rounds to $85.57, 50.1 percent under the spot. The negative drift did not pull the bull under the close, so the drift stays in all three cases.
The rounded bull is $2.33 from $320, and the rounded bear is not within $1 of $122, $130, $160, $120 or $144. This pair is not those older pairs.
| Case | Price | Versus $171.57 | What has to be true |
|---|---|---|---|
| Bull | $322.33 | plus 87.9 percent | The path lands 1.5 sample standard deviations above the drifted mean over 49 sessions. Pricing holds in the sense Volozh described on 21 October, and the market does not fade the August gap already inside sigma. |
| Base | $145.45 | minus 15.2 percent | Only the mean log-return compounds. No extra shock. The negative drift of the last 90 sessions keeps compounding through year-end. |
| Bear | $85.57 | minus 50.1 percent | The path lands one standard deviation under the drifted mean. Financing gets tighter at the lower tape, or the write-down narrative sticks, without an August-style upside day. |
What happens next
By 30 October 2026, seven of the 49 sessions will have printed. If Nebius still has not posted a third-quarter release, the reference price stays a completed close and the distribution does not get a new mean. A release that does land would be the first test of Alonso’s line that late-second-quarter capacity shows up in third-quarter revenue. A premarket print is not a bar on this chart.
By 27 November 2026, an early close that still counts, the Oxagon framework and the Mäntsälä substation should still be 2027 assets. Neither is in the 2026 revenue guide. A rally on those headlines alone would be paying for capacity the letter said contributes primarily in 2027. The base of $145.45 does not assume that pull-forward.
By 31 December 2026 the formula ends. The bull of $322.33 needs the one-and-a-half-sigma upside. The bear of $85.57 keeps the negative drift. The separate operating test is whether exit ARR is inside $7 billion to $9 billion and 2026 revenue is inside $3.0 billion to $3.4 billion. The letter says most of those large contracts hit revenue in 2027, so a print inside the guide can sit next to a price still under both averages. The distribution is the price path. The guide is not.
Frequently asked questions
Where do the sessions through 31 December 2026 put a Nebius bull and a bear that are not the $320 and $122 pair?
The bull is $322.33 and the bear is $85.57, with a base of $145.45, off the Wednesday 21 October close of $171.57. The figures use 49 sessions, a mean log-return of -0.00336996 and a sample standard deviation of 0.07578236. The September pair was an ARR multiple on a $237.33 close. This pair is the volatility formula, and the $2.33 gap versus $320 is not a reused target.
Did Nebius report third-quarter results on 22 October 2026?
Not on the company pages fetched for this piece. The investor-events page lists no upcoming events, and the newsroom’s latest results item is still the 12 August second-quarter release. Outside calendars have pointed at 22 October. Those calendars are not used here as the company’s date. A release that appears later would be a new document, and it would not change the Wednesday close that feeds the formula.
Why is the base below the close if the bull is almost $322?
Because the average daily log-return over the last 90 sessions is negative. Compounded for 49 sessions, that drift alone takes $171.57 to $145.45. The bull adds 1.5 standard deviations on top of the same drift, and this stock’s standard deviation is wide, in part because the 12 August gap is still in the sample. A wide bull and a base under the spot can be true at the same time.
How much of the risk is the 12 August session?
The close rose from $193.23 on 11 August to $259.20 on 12 August, a log-return of 0.2937. That day is 17.3 percent of the sum of squared deviations in the 90-return window. It is the largest single move, and it widens sigma, which widens the bull and the bear. It is not most of the variance. The calculation keeps it. Taking it out to make the range look calmer would be a different model.
Does the 50-megawatt Oxagon framework raise the 2026 guide?
No. HUMAIN and DataVolt described a preliminary framework, not a signed lease, with no contract value. Nebius would be an anchor customer for up to 50 megawatts, half of the first 100 megawatts now under construction, with phased operations from the fourth quarter of 2027. That is capacity Nebius might buy later. It is not revenue Nebius books this year, and it is not an input to $322.33 or $85.57.
Is the Wednesday close the price to use if the stock is trading on Thursday morning?
Yes, for this formula. The spot is the last completed official daily close in the Nasdaq file, $171.57 on 21 October 2026. A premarket print, a partial Thursday bar, or an after-hours print is not a close and is not the last bar on the chart. The historical file fetched for this piece ends on Wednesday. Any live print needs its own clock and its own source, and it does not move the bull, the base or the bear.
This is not financial advice.