Nebius Breaks Above $250 After 30% September Drawdown — Buy on Pullbacks Toward $237–$244

Nebius Breaks Above $250 After 30% September Drawdown — Buy on Pullbacks Toward $237–$244

A backlog above $40B and a 17–21% GPU price increase support the case | That's TradingNEWS

Itai Smidt 10/6/2026 12:24:25 PM

Key Points

  • NBIS trades $251.48, up 8.13% from $232.57, with a day range of $237.00 to $253.11.
  • Q2 revenue was $582.3 million, up 454%, with ARR at $3.0 billion and a 41% adjusted EBITDA margin.
  • 2026 capex is guided to $20–25 billion against $8 billion of cash; 70% of Q2 deals were prepaid.

Nebius Group (NASDAQ: NBIS) traded at $251.48 at 11:19 a.m. Eastern on Tuesday, up $18.91 or 8.13% from Monday's $232.57 close. The stock opened at $237.22, never traded below $237.00, and reached $253.11. Volume was 10.86 million shares before noon against a three-month daily average of 20.08 million. Market capitalization stands at $68.93 billion on 274.1 million shares.

The move reverses Monday's session in full. Shares fell $10.24, or 4.22%, on Monday after opening at $244.03 and sliding to $230.51, a day when coverage of the stock centered on valuation, customer concentration and the durability of its contract backlog. Tuesday's rally carried the price $8.67 above Friday's $242.81 close and to its highest level since the September selloff.

There was no company announcement on Tuesday. The gain came with a broad bid for AI infrastructure. Nvidia set a record at $241.83. Marvell Technology rose 7.54% on its investor day. Constellation Energy gained 9.02% before the open on a 20-year nuclear supply agreement with Google. The S&P 500 printed an all-time high. A stock with a beta of 1.46 and a 178% year-to-date gain through Monday tends to move at a multiple of that kind of tape.

The case argued here is that Nebius has the strongest unit economics among the independent AI cloud providers and one of the most demanding funding profiles in the market, and that the first outweighs the second at this price. Second-quarter revenue grew 454% to $582.3 million. Annualized run-rate revenue reached $3.0 billion and is guided to $7 billion to $9 billion by year-end. The adjusted EBITDA margin was 41%. Customers prepaid for 50% to 60% of the capital cost on 70% of the deals signed in the quarter. The company raised on-demand GPU prices by 17% to 21% on October 1, its second increase in three months.

Against that, capital spending is guided to $20 billion to $25 billion this year, free cash flow over the past twelve months was minus $9.61 billion, most of a backlog above $40 billion comes from two customers, and the stock trades at 51 times trailing revenue.

The 52-week range is $73.52 to $299.86. At $251.48 the shares are 16.1% below the high and 20.3% above the September low of $209. Third-quarter results are due November 10.

The Tape: A 30% September Drawdown, Then a Series of Higher Lows

Nebius has been one of the most volatile large-capitalization stocks of 2026, and the past six weeks compress that history.

The run began with earnings. On August 12 the company reported second-quarter results that beat consensus, and the shares gapped from a $193.23 close to $225.15 in premarket trading, a 16.5% jump. The rally extended to the 52-week high of $299.86.

September undid much of it. A wave of concern over AI safety and the sustainability of infrastructure spending hit the group, and Nebius fell to $209 by mid-month, more than 30% below its peak. A sell rating with an $84 price target added to the pressure. A prominent short seller disclosed that he had converted short positions in the stock into put options expiring before the summer of 2027, framing the trade as a hedge against an AI bubble. CoreWeave, the largest peer, was 53% below its own record at $86.65 on September 29.

The recovery came in steps tied to specific catalysts. On September 17 the company announced price increases, and the stock rose with the group as investors read the move as evidence of pricing power. On September 24 the shares jumped 9% from a $226.61 close after an upgrade to outperform with a $399 target and higher revenue estimates from a second firm. On September 30 a new outperform initiation added 2% in premarket trade. October 1 brought an acquisition, and the stock dipped 1.5% to $232.28 after touching $228.47. October 2 produced a 4% gain to $242.81 in a broad rally.

Monday's 4.22% decline to $232.57 and Tuesday's 8.13% rebound to $251.48 complete the sequence. Across it, each low has been higher than the one before: $209, then $226.61, then $228.47, then $230.51. Tuesday's high of $253.11 is the first print above $250 since the selloff.

Returns over longer windows show the scale. The stock gained 113.72% over six months and 81.72% over one year through Monday. One-month performance was 2.73%, reflecting how much of September's damage had been repaired before Tuesday.

A leveraged exchange-traded fund that targets twice the stock's daily move holds roughly half its portfolio in Nebius exposure. Flows into and out of that product amplify trading in the shares, particularly near the close.

Insiders sold $19.07 million of stock in recent transactions disclosed in regulatory filings. The sales were small against a $68.93 billion market value and drew attention mainly because of their timing after the rebound.

Technical Picture: Support at $237 and $242.81, Open Space to $283.58

The chart has improved materially with Tuesday's move, and the reference levels are clear.

The first support is the gap. Monday closed at $232.57 and Tuesday opened at $237.22, leaving $4.65 of untraded space. The session low of $237.00 sits just under the open. Gaps of this kind on strong volume often hold as support for several sessions. A close back inside it would weaken the breakout.

Above the gap, $242.81 to $244.19 is the next band. It contains Friday's close and Monday's high, the levels the stock failed to hold at the start of the week. Tuesday's rally went straight through them. A pullback that holds above $242.81 would confirm that area has flipped from resistance to support.

Beneath those, the references are Monday's close at $232.57, the cluster of recent lows between $228.47 and $230.51, the September 24 pre-upgrade close of $226.61, and the September low of $209. The pre-earnings close of $193.23 from August 11 marks the bottom of the August gap. The 52-week low is $73.52.

Overhead, Tuesday's high of $253.11 is immediate resistance. The $250 round number has been crossed and will be tested on any intraday fade. Beyond that there is little recent trading history until the average published price target of $283.58, which is 12.8% above the current price. The 52-week high of $299.86 is 19.2% away. A break of that would put the $399 target cited by the most bullish firm in view.

The pattern since mid-September is a rounded base with rising lows, and Tuesday's move is the first decisive push out of it. The measured height of the base, from $209 to roughly $244, is $35. Projected from $244, that points to $279, close to the consensus target.

Volume deserves a note. At 10.86 million shares by 11:19 a.m., the stock was on pace to exceed its 20.08 million average. Monday's decline came on 14.95 million shares, below average. A selloff on light volume followed by a rally on heavier volume is the constructive version of that sequence.

Two cautions apply. The stock has moved more than 4% in four of the past five sessions, alternating direction, so intraday reversals are common. And the 30% drawdown in September shows how quickly sentiment toward this group can change.

The level that matters most on a closing basis is $242.81. Above it, the stock has broken its range. Below $228.47, the pattern of higher lows would fail.

Fundamentals: $582.3 Million in Quarterly Revenue, Up 454%, With a 41% EBITDA Margin

The second quarter was the strongest in the company's short history as a standalone AI cloud business.

Group revenue was $582.3 million for the quarter ended June 30, up 454% from a year earlier and 46% from the first quarter. Consensus had been in a range of $569.9 million to $572.8 million. The core AI cloud business generated $574.9 million, up 514% year over year, and accounted for 98% of the total. First-half revenue was $981.3 million, up 529%.

Annualized run-rate revenue, calculated as the last month of the quarter multiplied by twelve, reached $3.0 billion at the end of June. That compared with $1.9 billion at the end of March and $1.25 billion at the end of 2025, increases of 56% in one quarter and 140% in six months.

Profitability arrived faster than expected. Group adjusted EBITDA was $236.2 million, against a loss of $21 million a year earlier and $129.5 million in the first quarter. The margin was 41%, up from 32%. The AI cloud segment alone produced a margin of 50%. Cost of revenue was $133.6 million, or 23% of sales.

Below that line the picture is less clean. The company reported a net loss of $190.4 million, or $0.68 a share, under standard accounting. Interest expense scaled to $119.1 million as convertible notes and debt raised over the past year flowed through the income statement. Depreciation on a rapidly growing asset base is the other large item.

Management attributed the growth to capacity added in the first quarter, higher utilization, and high-margin revenue from an asset-light model and from Token Factory, its managed inference platform, along with recent acquisitions.

Commercial activity accelerated. The company closed four AI cloud deals in the quarter with an average total contract value above $1 billion each. Total contract value of new wins grew almost fourfold from the first quarter.

Cash flow from operations was $2.3 billion, boosted by customer prepayments. The company ended June with $8.0 billion in cash and equivalents plus $1.1 billion of restricted cash.

A timing point matters for the next report. Capacity that came online late in the second quarter began contributing meaningfully to revenue only in the third. Most of the 2026 buildout is scheduled for the second half. The second quarter's 46% sequential growth was achieved before the bulk of this year's capacity was switched on.

Trailing twelve-month revenue is $1.36 billion, up 507%.

Guidance: $7 Billion to $9 Billion in ARR by December and What It Requires

Full-year targets were reaffirmed across every metric in August, and the arithmetic behind them sets the bar for November 10.

For 2026 the company expects group revenue of $3.0 billion to $3.4 billion, year-end annualized run-rate revenue of $7 billion to $9 billion, an adjusted EBITDA margin of 40%, and capital expenditures of $20 billion to $25 billion. Connected power is targeted at 800 megawatts to 1 gigawatt by December. Formal guidance for 2027 is planned for later this year.

One target was raised. Contracted power for year-end was lifted to 5 gigawatts from more than 4 gigawatts. In February the figure had been more than 3 gigawatts, and the original plan called for 2.5. The target has doubled in under a year.

The revenue path is steep. With $981.3 million booked in the first half, the $3.2 billion midpoint requires $2.2 billion in the second half. That is 2.3 times first-half revenue. If the third quarter lands near $900 million, the fourth would need to deliver $1.3 billion.

The run-rate target is steeper still. From $3.0 billion at the end of June to the $8 billion midpoint is an increase of 167% in six months. It depends on bringing several hundred megawatts online on schedule. The company ended 2025 with 170 megawatts of active power, so reaching 800 megawatts to 1 gigawatt means multiplying the installed base by five to six times in twelve months.

Management has described the unit economics. Capacity yields $20 million to $25 million of annual revenue per megawatt. At the midpoint, 900 megawatts of connected power would support revenue in the range of $18 billion to $22 billion a year at full utilization, well above the run-rate target, which implies that much of the year-end capacity will still be ramping in December.

The Vineland, New Jersey site, which serves the Microsoft agreement, remains in the 2026 plan. Delivery there is among the largest single contributors to the second-half ramp.

Reaffirming the range in August mattered because doubts had circulated ahead of the report. Holding all four metrics while raising contracted power answered them for one quarter.

For November 10, three numbers will be scrutinized. Third-quarter revenue needs to show the acceleration the full-year guide implies, likely above $850 million. Run-rate revenue at the end of September should be in the area of $4.5 billion to $5 billion to keep the year-end range credible. And connected power needs to have risen substantially from midyear.

From 2027 the company plans to deploy more than 1 gigawatt a year. One sell-side projection has run-rate revenue approaching $22 billion by the end of 2027.

Contracts and Backlog: More Than $40 Billion, Two Customers

The backlog is the foundation of the bull case and the source of its most cited risk.

Customer commitments exceed $40 billion. Two agreements make up most of it. Microsoft signed a contract valued at $17.4 billion, rising to as much as $19.4 billion with additional capacity, reported to cover more than 100,000 of Nvidia's GB300 chips. Meta Platforms signed an initial $3 billion agreement over five years in November 2025 and a second in March 2026 worth up to $27 billion, of which $12 billion is committed in the first five years.

Beyond the two anchors, disclosed customers include Reflection, Cohere and a U.S. quantitative trading firm. Palantir named Nebius its preferred sovereign AI infrastructure provider. In late September, AIB Data Centers signed a binding agreement to supply 50 megawatts of critical capacity for the company's use.

The contracts fall into three tiers. Short-term deals of three to six months carry premium pricing. Core contracts run one to three years. Long-term agreements with investment-grade counterparties provide the base load and the collateral for financing. Management has said the short-term tier is earning higher prices as demand outstrips supply.

Prepayments are what distinguish the model. Seventy percent of deals closed in the second quarter included upfront customer payments, a record, and those payments covered 50% to 60% of the associated capital expenditure. The payback period on capital and related operating costs for second-quarter deals was one year. The company expects more than $9 billion of customer prepayments in 2026.

When a customer funds half the build before the first chip is installed, the provider needs less external capital, and the customer has a financial stake in taking delivery. Financing secured in the second quarter was priced at 2.50 percentage points over the benchmark overnight rate, a low spread for a company at this stage.

Concentration is the risk on the other side. If most of $40 billion comes from two buyers, a change in either one's capital plans would be felt immediately. Both Microsoft and Meta are also developing their own accelerators and building their own data centers. Microsoft has committed more than $60 billion across several independent cloud providers, partly because such contracts are booked as operating expenses over time and keep the spending off its own capital budget. That accounting preference could change.

The counterargument is contractual. These are multi-year take-or-pay commitments from two of the best credits in the world, with cash paid in advance. They are obligations the customers have already funded in part.

Capital Intensity: $20 Billion to $25 Billion of Capex Against $8 Billion in Cash

The funding requirement is the number that makes cautious investors hesitate, and it deserves a plain accounting.

Capital expenditure in the second quarter was $5.66 billion, compared with $510.6 million a year earlier. Full-year guidance is $20 billion to $25 billion. The first half consumed a portion of that, leaving the larger share for the second half. Trailing twelve-month free cash flow is minus $9.61 billion.

The sources of funds are four. The first is cash on hand: $8.0 billion plus $1.1 billion restricted at the end of June. The second is customer prepayments, expected to exceed $9 billion this year. The third is operating cash flow, which was $2.3 billion in the second quarter alone. The fourth is external financing, including asset-backed debt secured against long-term contracts and convertible notes.

Added together, the first three cover a large part of the plan. Cash of $9.1 billion, prepayments above $9 billion and second-half operating cash flow that should exceed the first half's bring the total toward the low end of the capex range before any new borrowing. The debt-to-equity ratio is 98.6%.

Spending is also sequenced to follow demand. Management has said it secures as much power as it can because the cost of doing so is small, builds as fast as capital allows, and fills sites with GPUs only in line with contracted or clearly visible demand. Chips account for the large majority of the cost of a deployed megawatt, so the heaviest outlay comes only after a customer has signed.

The risks are execution and timing. At $5.7 billion a quarter, any delay in bringing a site online means capital is deployed without revenue to match. Construction, grid connection and chip delivery each have to arrive on schedule. Interest expense of $119.1 million a quarter will rise as more debt is drawn.

Credit markets have turned more selective toward the sector. A widely read analysis in September argued that the most popular part of the AI trade could become its weakness, citing risky contracts and rising debt costs. An AI data-center loan was marketed this week at an 11% yield. Nebius's borrowing at 2.50 points over the benchmark reflects the quality of its counterparties, but the direction of spreads across the industry is wider.

A pricing concern has been raised as well. If GPU rental rates fall faster than assets depreciate, returns on data-center investment would compress. The company's two price increases in three months are the current evidence against that.

The balance sheet can carry the 2026 plan. The 2027 plan, at more than 1 gigawatt of new capacity a year, will require the prepayment model to scale with it.

Pricing Power and Product: A 17% to 21% Increase and the Inferize Acquisition

Two recent decisions show where management sees its advantage.

On October 1, Nebius raised pay-as-you-go prices across its on-demand GPU cloud. The increases were 17% to 21% on instances built on Nvidia's H100, H200, B200 and B300 chips, and 25% on AMD EPYC central processor rates. It was the second price rise in three months.

Raising prices on older chips is the notable part. The H100 is two generations behind Nvidia's current products. In a normal hardware cycle, rental rates on prior-generation equipment fall as newer models arrive. Rates rising instead means demand for compute of any vintage exceeds what is available. For a provider, higher prices on fully or partly depreciated hardware flow almost entirely to margin.

The market initially worried the increase might drive customers away. Over the following two weeks that reading reversed, and the price change came to be seen as evidence of strong demand. A large peer made similar moves, and commentary described the independent cloud providers as flexing their pricing power.

The second decision was an acquisition. On October 1 the company bought Inferize, an inference optimization firm whose technology shortens the time needed to launch and scale large models and makes inference workloads elastic. The price was not disclosed; one report put it at up to $150 million. The team has joined Token Factory, the company's managed inference platform. The deal follows the purchase of the agentic search company Tavily for $400 million in February and earlier additions of Eigen AI and technology from Clarifai.

The strategic direction is up the stack. Renting bare GPU capacity is a business in which providers compete mainly on price and availability. Managed inference, where the customer pays per token and the provider handles optimization, earns a higher margin and is harder to switch away from. Management cited Token Factory and the asset-light model as contributors to second-quarter margin expansion.

That positioning is what an initiating firm meant last week when it described Nebius as not just another neocloud, pointing to infrastructure scale, software capability, customer relationships and low-cost access to capital.

Ownership of the physical layer supports the margin too. More than 75% of contracted capacity sits in the company's own data centers, with sites in Finland, New Jersey, Missouri, the United Kingdom, France and Iceland. The Finnish facility is planned at 310 megawatts and the Missouri campus at 1 gigawatt. Owning the building removes a landlord's margin and gives control over design and timing.

The company also has a strategic alliance with Nvidia, which is an investor, an arrangement that has helped secure chip allocations.

Valuation: 51 Times Trailing Sales, 8.6 Times Year-End Run Rate

The stock is expensive on what the company has earned and reasonable on what it has guided, and the gap between the two is the whole debate.

At $251.48, market capitalization is $68.93 billion. Trailing twelve-month revenue is $1.36 billion, giving a price-to-sales ratio of 50.7. On the midpoint of 2026 revenue guidance, $3.2 billion, the multiple is 21.5. On June's run rate of $3.0 billion it is 23.0. On the midpoint of the year-end run-rate target, $8 billion, it is 8.6.

Earnings-based measures are not informative. Trailing net income of $42.4 million includes non-operating gains, and the second quarter alone showed a $190.4 million loss. Published price-to-earnings figures range from 189 to more than 1,500 depending on the share count and earnings base used.

A more useful frame is cash earnings against the growth rate. Adjusted EBITDA in the second quarter was $236.2 million, or $945 million annualized. If the company reaches an $8 billion run rate at a 40% margin, annualized EBITDA would be $3.2 billion, and the stock would trade at 21.5 times that figure. If one sell-side projection of a $22 billion run rate by the end of 2027 were met at the same margin, EBITDA would be $8.8 billion and the multiple under 8.

Analyst views are widely dispersed. Nine firms have an average rating of buy and an average target of $283.58, which is 12.8% above the current price. The highest target is $415 and the lowest is $84. Recent calls include an upgrade to outperform with a $399 target, an outperform initiation that argued the valuation does not reflect long-term earnings power, a neutral rating with a $224 target, and a sell at $84. A spread of $331 between high and low on a $251 stock shows how far apart the assumptions are.

Against the main peer, the comparison favors Nebius on growth and balance sheet. CoreWeave reported second-quarter revenue of $2.575 billion, up 112%, with a backlog of $104 billion and 1.5 gigawatts of active power. Its revenue is 4.4 times larger. Nebius is growing four times faster, carries $8 billion of cash, and funds half its buildout with customer money. CoreWeave leans almost entirely on leased facilities and carries more debt.

The bear case on valuation is that the multiples only work if every milestone is hit. A stock at 51 times trailing sales has no cushion for a missed quarter. The shares fell 30% in September on sentiment alone.

On guided numbers the price can be justified, and it leaves little room for a miss.

The Sector: A Strong Day for AI Infrastructure

Nebius's 8.13% gain came on a day when the evidence for AI spending was favorable across the supply chain.

Nvidia traded at a record $241.83 with a market value of $5.839 trillion, within 3% of becoming the first $6 trillion company. Its second-quarter data-center revenue was $89.0 billion, up 117%, and it has guided the current quarter to $108 billion. Nvidia's results named Nebius among the partners already running racks of the Vera Rubin platform, the next chip generation.

Hon Hai, the largest assembler of AI servers, reported September-quarter revenue up 47% to $95.48 billion on Monday and said its AI business would keep expanding through year-end. Taiwan Semiconductor hit a record.

Marvell Technology rose 7.54% to $291.71 as it laid out a long-range plan anchored by a $120 billion custom chip agreement with Google.

Power was the day's other theme. Constellation Energy announced a 20-year agreement with Google covering 890 megawatts of new nuclear capacity and a separate 2,700-megawatt supply deal. Electricity is the binding constraint on data-center growth, and Nebius's guidance is expressed in megawatts for that reason. One recent analysis described the AI buildout as needing power it cannot get.

Financing headlines pointed the same way. Banks launched a record $60 billion chip-backed financing tied to Broadcom and Anthropic. OpenAI is in talks on a $30 billion funding round. Those are the ultimate buyers of the compute that Nebius and its peers rent out.

Demand indicators for the independent providers have been positive for three weeks. Coverage in mid-September cited two signs that investors need not fear an AI slowdown: strong demand reported by the providers and rising prices for AI chips. CoreWeave received a new Street-high price target of $250 on Monday.

The group traded poorly on Monday regardless. CoreWeave fell 2.49% to $87.39, IREN dropped 3.07% to $40.48 and Applied Digital lost 2.68% to $24.70. Tuesday's rebound in Nebius came as the broader market set a record.

The macro backdrop is mixed for a capital-intensive growth company. The 10-year Treasury yield is 5.27%, near a 24-year high, and the Federal Reserve raised rates in September. High rates raise the cost of the debt that funds data centers. Offsetting that, lower oil prices on Tuesday pulled yields down, and equity investors have been willing to pay for earnings growth.

Nebius is a member of the Nasdaq-100, so index-tracking funds hold it and trade it with flows into technology.

Competition: CoreWeave, the Hyperscalers and the Custom-Chip Question

Nebius competes on three fronts, and its position differs on each.

Among independent AI clouds, CoreWeave is the largest. It had 1.5 gigawatts of active power at midyear, 4.2 gigawatts contracted, 51 data centers and a $104 billion backlog with a further $25 billion added early in the third quarter. Its anchor customers are Meta, OpenAI and Anthropic. Nebius is targeting up to 1 gigawatt connected by December against 5 gigawatts contracted, so the two will be closer in contracted scale by year-end than their current revenue suggests.

The business models differ. CoreWeave relies almost entirely on colocation, leasing space in facilities owned by others. Nebius owns more than three-quarters of its capacity. CoreWeave has financed growth mainly with debt. Nebius has used equity, convertibles and customer prepayments. One comparison after second-quarter results concluded that Nebius's cash cushion, prepayment-funded builds and revenue streams beyond GPU rental make it the more defensible of the two if large customers squeeze the market.

Other independents include Lambda, Crusoe with 4.9 gigawatts contracted, Nscale in Europe, IREN and Applied Digital. Microsoft has spread more than $60 billion across several of them.

The second front is the hyperscalers themselves. Microsoft, Meta, Google and Amazon are customers and competitors at once. They rent from independent providers because they cannot build fast enough and because rental keeps spending off their capital budgets. If their own construction catches up with demand, the need for third-party capacity falls. Take-or-pay contracts protect existing commitments. Renewals and new deals would be at risk.

The third front is custom silicon. The large technology companies are designing accelerators with Broadcom and Marvell to reduce reliance on Nvidia. Nebius's fleet is built on Nvidia hardware. If workloads migrate to in-house chips running in the hyperscalers' own facilities, demand for rented Nvidia capacity could soften. The time frame for that is years, and so far custom chips have added to total compute demand without displacing merchant GPUs.

Nebius has two differentiators. One is geography. It has a strong European footprint and is positioned for customers who need data processed within national borders, a market where regulatory barriers limit U.S. competitors. The Palantir designation for sovereign AI infrastructure reflects that. The other is software. The company inherited a large engineering organization from its predecessor business and builds its own server designs, cloud platform and inference tooling.

It also holds two non-core businesses: Avride, which develops autonomous driving and delivery robots, and TripleTen, an education technology platform. Together they contributed 2% of second-quarter revenue.

Risks: Concentration, Funding, Execution and a 30% Drawdown Three Weeks Ago

A stock that has nearly tripled this year carries risks in proportion, and five are material.

Customer concentration comes first. Most of a backlog above $40 billion is attributable to Microsoft and Meta. Both are building their own capacity and designing their own chips. A decision by either to slow spending would affect future bookings even if existing contracts are honored.

Funding is second. Capital spending of $20 billion to $25 billion in one year is three times the company's cash balance. The plan depends on prepayments continuing at record levels and on debt markets staying open at reasonable spreads. Credit investors have become more demanding across the sector. Trailing free cash flow of minus $9.61 billion means the company cannot fund itself without customers and lenders.

Execution is third. Reaching 800 megawatts to 1 gigawatt of connected power from 170 megawatts in twelve months requires construction, grid connections and chip deliveries to land on schedule across several countries. Guidance implies second-half revenue 2.3 times the first half. A delay at one large site would show up immediately in the numbers.

Valuation is fourth. At 51 times trailing sales, the shares price in the guided ramp. The spread between the highest and lowest analyst targets is $331. The stock fell more than 30% in September without any change in company fundamentals.

The fifth is the durability of GPU rental pricing. Rates are rising now. If supply catches up, they could fall faster than the hardware depreciates, which would compress returns on assets already purchased. A critical analysis this week argued that the two largest independent providers face a pricing problem that could undermine data-center returns.

Additional items include rising interest expense, already $119.1 million a quarter; dilution from convertible notes; regulatory exposure across multiple jurisdictions; and a visible short position by a well-known investor through puts expiring before mid-2027. Insider sales of $19.07 million were disclosed recently.

Macro conditions matter more for this company than for most. A 10-year yield at 5.27% raises the hurdle for long-duration growth assets and the cost of infrastructure debt. A beta of 1.46 means the stock would fall harder than the market in a correction.

Set against those, the company has $8 billion in cash, a 41% EBITDA margin, operating cash flow of $2.3 billion in one quarter, contracts with two of the strongest counterparties in the world, and prices that are going up. None of the risks is visible in reported results today. Each would matter if growth slowed.

Verdict: Buy for Risk-Tolerant Investors, With $283.58 the Target and $228.47 the Line

Nebius is a high-growth, high-capital business whose reported numbers currently support the bullish view, at a price that requires them to keep doing so.

The positives are measurable. Revenue grew 454% last quarter. Run-rate revenue rose from $1.25 billion to $3.0 billion in six months and is guided to $7 billion to $9 billion by December. The adjusted EBITDA margin reached 41%. Seventy percent of new deals came with prepayments covering half or more of the capital cost. On-demand prices rose 17% to 21% on October 1. The backlog exceeds $40 billion. Contracted power guidance has doubled in a year to 5 gigawatts. The company owns most of its data centers and is adding higher-margin inference software.

The negatives are also measurable. Capital spending will be $20 billion to $25 billion this year against $8 billion of cash. Free cash flow is deeply negative. Two customers account for most of the backlog. The stock trades at 51 times trailing revenue and dropped 30% in three weeks last month.

On the chart, Tuesday's 8.13% gain to $251.48 broke the stock out of a three-week base with rising lows. Support sits at $237.00 and at $242.81.

The rating is buy for investors who can tolerate large swings. The first target is $283.58, the average of published estimates and 12.8% above the current price. Beyond that, the 52-week high of $299.86 is 19.2% away. A third-quarter report on November 10 that shows run-rate revenue near $5 billion and connected power on track would bring the $399 target into the discussion.

The position is wrong on a close below $228.47, which would break the sequence of higher lows since mid-September. A return to $209 would mean the September low had failed to hold. From $251.48, the invalidation level is 9.2% below and the first target 12.8% above. That ratio is modest, which argues for entering on pullbacks toward $237 to $244, where the risk falls to between 4% and 6% and the upside to the target rises to between 16% and 20%.

Position size should reflect the volatility. This stock has moved more than 4% in four of the last five sessions.

The decisive test is execution in the second half. If revenue in the third and fourth quarters reaches the $2.2 billion that guidance implies, the stock will be trading at under 9 times its own year-end run rate with half its growth capital paid for by customers. If the ramp slips, the valuation will reprice quickly, as it did in September.

That's TradingNEWS