A customer buying a large training run needs thousands of processors to communicate quickly. A slow link or a failed job can waste expensive GPU time. Nebius designs the clusters, connects them over a high-speed network and supplies the software that schedules and monitors workloads. In inference, the customer cares about cost per response and consistent latency. Higher utilization helps Nebius spread the cost of its equipment over more paid hours, provided it can still meet the response times it promised.

Large contracts reserve demand in advance. A customer deposit can fund hardware before delivery, and an investment-grade counterparty can make a signed contract easier to borrow against. The trade-off is concentration. A delay serving one large customer can move an entire quarter, while the interest on the financing and a portion of the site costs continue. A contracted megawatt, an energized megawatt and a megawatt accepted for billing represent three different stages.

There is a second route to revenue. A partner can fund the capacity and pay Nebius to enable its architecture and cloud software. That first implementation accounted for about $325 million of Q3 revenue. The contract demonstrates that the platform can earn money on equipment somebody else funds. Management excluded potential asset-light revenue from its initial 2027 outlook.

The distinction matters when valuing the group. AI cloud generated $973.8 million of its $980.8 million in Q3 revenue. Avride and TripleTen also sit inside the group, and it owns stakes in ClickHouse and Toloka. These holdings may have value, but the economics of GPU capacity dominate the revenue forecast. Our valuation assigns no separate premium to a possible sale or flotation of an investment.

Group revenue rose from $146.1 million a year earlier to $980.8 million, ahead of the $907.5 million consensus cited by Reuters. Subtracting the approximately $325 million partner rollout leaves about $655.8 million of other revenue. That residual is a calculation, not a disclosed recurring-revenue segment. It helps explain why extrapolating the headline 68% sequential growth into the next quarter would overstate the evidence.

That mix also matters for profitability. Group adjusted EBITDA was $465.8 million, or 47.5% of sales. The cloud segment reached a 53.2% adjusted margin. At group level, depreciation and amortization alone ran to $354.7 million. Operating income was $50.4 million, interest expense was $112.5 million, and the net loss from continuing operations was $37.5 million. A high adjusted operating margin can coexist with a loss for shareholders who bear the cost of replacing equipment and servicing debt.

Quarterly operating cash flow from continuing operations was negative $273 million. Purchases of property, equipment and intangibles were $7.30 billion. Subtracting those cash purchases gives an outflow of about $7.57 billion before financing on this simple measure. 

Nebius estimates revenue-based payback on comparable one-to-three-year deals improved from 22 months to 19, excluding prepayments. In practical terms, the contracted revenue covers the defined upfront investment three months earlier. Operating expenses, financing costs and taxes still matter. The figure cannot be read as the time until an equity investor has earned back cash.

Annual contract value per megawatt rose from $20–25 million to above $25 million. The newer sample includes longer contracts. Multiplying the Q3 figure by every megawatt of contracted power would overstate the business because it assumes comparable equipment, available power, the same customer terms and immediate utilization at every site.

Roughly 70% in Q2 refers to deal count. More than 90% in Q3 refers to the value of agreements carrying a prepayment. These percentages cannot establish a 20-point increase in the number of customers paying upfront. A comparable measure is the share of qualifying capex covered by prepayments, which moved from 50–60% to above 100% for the selected deals of at least a year. Timing and coverage across the entire construction program remain separate questions.

Vineland, New Jersey, illustrates the cost of a timing error.  Management expected more than half the Microsoft capacity by year-end and the balance in Q2 2027. Underlying contracts remained in place. Customers can still want every GPU, yet an unfinished site pushes revenue out while capital is already tied up. The delay also leaves investors less willing to credit distant capacity announcements.

On the call, management narrowed 2026 revenue to $3.0–3.2 billion, versus $3.0–3.4 billion previously. The shareholder letter describes the low end of the earlier range. Group adjusted EBITDA margin guidance stayed near 40%, while planned capex rose from $20–25 billion to around $27 billion. The earlier December ARR goal of $7–9 billion shifted to a February milestone above $7 billion. ARR annualizes a single month. A $7 billion run rate corresponds to about $583 million in that month, without guaranteeing the same level for the following twelve months.

Management’s August and November guidance side by side. · Couch Investor · 6:25View original ↗

The initial 2027 outlook calls for $12–13 billion of sales and $40–45 billion of capex. Management aims for quarterly pre-tax breakeven in Q4 2027, after financing costs and using a five-year life for compute equipment. It has not promised full-year profitability. To reach the $12.5 billion midpoint requires average monthly sales of about $1.04 billion. Even reaching February’s milestone would leave a substantial ramp for the remainder of the year.

Management expects prepayments and asset-backed funding against current investment-grade contracts to cover 75–85% of 2027 capex. At the $42.5 billion midpoint, the remaining 15–25% is about $6.4–10.6 billion. This is the amount outside those two sources, before considering balance-sheet resources, operating cash flow or new business. The expected $17–20 billion of customer prepayments is already part of the coverage calculation. 

Estimated 2027 capital needs after prepayments and asset-backed financing. · Couch Investor · 9:43View original ↗

A prepayment creates a future service obligation. A lender advances money against expected cash flows and charges interest. Both can reduce immediate reliance on shareholders, but neither makes the build free. Q3 included $5.75 billion of convertible notes and roughly $1 billion of net proceeds from share sales. At the reported $229.10 weighted-average issue price, raising $1 billion requires roughly 4.4 million shares before fees. At $128.64 it takes about 7.8 million. The same financing decision becomes materially more expensive per existing share after a selloff.

Execution risk has two sides. Owning more sites can reduce dependency on a landlord’s power timetable. Nebius then carries more responsibility for permits, construction, cooling and grid access. Expensive GPUs may lose rental value when a newer generation arrives. A competing cloud with spare capacity can cut prices, just as financing costs rise. A customer delaying acceptance magnifies both risks.

The reassuring evidence would be cash collections arriving before major equipment payments, completed sites beginning to bill, and rising operating income after depreciation. A second Vineland delay, lower prepayment coverage, or repeated issuance at depressed prices would weaken the case even if the nominal contract pipeline continued growing.

NBIS closed at $154.69 on November 11 and $128.64 the next day, a 16.84% decline on approximately 69.5 million shares. From the October 9 close of $221.07, the drawdown was 41.8%. Yet the November 12 close was higher than the November 5 close of $120.37. The sequence matters: investors bid the stock up 15.44% immediately before the release, then reversed the gain and more. This was a market repricing the reliability of the build schedule as well as reacting to the earnings number.

For a reproducible momentum check, sum the positive and negative close-to-close changes across the 14 sessions ending November 12. Gains totalled $81.09 and losses $111.19. The simple-window RSI (Cutler method) is 42.2, above the conventional 30 oversold threshold. A charting platform using Wilder smoothing can give a different result. 

The conclusion is narrower than a technical bottom call. By magnitude the drawdown is severe. By this momentum measure the stock had not yet registered an oversold extreme.

Volume shows investors reassessed the news. It does not reveal that the last marginal seller has sold. The $118.81 earnings-day low and $120.37 prior closing reference define an area to watch. A sustained recovery above the pre-release $154.69 close would be stronger price evidence. Neither boundary guarantees support or resistance. A failed capacity milestone could overwhelm both.

Our end-2027 estimate uses equity value divided by revenue. Traditional earnings multiples are unstable near breakeven. We multiply expected 2027 sales by an assumed equity price-to-sales multiple, then divide by projected diluted shares. This equity multiple already reflects our judgment about financing risk. Unlike an enterprise-value multiple, it does not require a second deduction for net debt. It also cannot value debt independently, so the scenarios must change when the funding mix changes.

At $128.64 and 292.9 million shares outstanding on September 30, the indicative equity value is $37.7 billion, or 3.0 times the $12.5 billion guidance midpoint. Applying an illustrative future share count of 315 million lifts that entry multiple to 3.24 times. Our 4.5-times base assumption requires a rerating as delivery and funding risk diminish. 

Simply hitting the sales guide at a 3-times multiple would yield only $119 on 315 million shares. A strong year for Nebius can still disappoint its shareholders.

The bear case assumes $9.5 billion sales, 2.75 times sales and 340 million shares: about $77, or 40% downside. The base case pairs $12.5 billion, 4.5 times and 315 million shares: about $240, or 98% upside. The bull case assumes $13.5 billion, 6 times and 305 million shares: about $284, or 136% upside. These multiples and share counts are editorial assumptions. The bull case requires guidance outperformance, timely power and unusually favorable financing.

Our view is that the selloff created a credible opportunity for investors willing to underwrite construction and funding risk. The return in the base case is meaningful, but the modeled downside is of similar magnitude. The strongest argument for owning the stock is the evidence that customers accept shorter revenue-payback periods and help fund capacity. A 42% drawdown alone says little about prospective returns when the expected share count and cost of borrowing can change this quickly.

We would raise the estimate if the February milestone is reached, the remaining Vineland capacity bills on schedule, and financing disclosures show that growth can be achieved close to our assumed share count. We would cut it if a second delay pushes revenue beyond 2027 or cheaper GPU rentals undermine contract renewals. A move toward $265–270 is plausible in the upside case. At the reporting date, it requires more evidence than the company has delivered.