This is investment research, not personal financial advice.
Nebius Group (NASDAQ:NBIS) rallied 34.1% on 12 August, from $193.23 to $259.20, after reporting Q2 revenue of $582.3 million, up 454% from a year earlier. The market added roughly $17.7 billion to the company's equity value in one session. Revenue growth was the headline. The more consequential disclosure sat elsewhere: management said new cloud contracts imply a roughly 22-month payback on the capital committed to serve them (Nebius 2026a; Yahoo Finance 2026).
That claim changes the question. Demand is no longer the immediate constraint. Nebius ended Q2 with more than $7 billion of annualised run-rate revenue, or ARR, under contract and $24 billion of backlog. It is now trying to turn scarce power, GPUs and network capacity into repeatable cash returns while building far ahead of recognised revenue. The 34.1% reaction looks directionally justified because Q2 supplied evidence that newly installed capacity can earn attractive project returns. The size of the move, however, prices a large portion of that evidence as durable before fleet-level cash conversion has been demonstrated.
A 22-month claim did more work than a 454% headline
The Q2 release contains three layers of evidence. First, reported revenue accelerated: $582.3 million for the quarter and $981.3 million for the half, against $105.1 million and $156.0 million a year earlier. Second, the core Nebius cloud segment produced $444.6 million of Q2 revenue and $222.0 million of adjusted EBITDA, a 49.9% margin. Third, the contract book expanded faster than reported revenue. ARR rose 96% from the end of Q1 to more than $7 billion, while backlog reached $24 billion (Nebius 2026a).
The disclosed commercial terms explain why the market treated this as more than a beat. Management described an average contract duration of 3.6 years and average prepayment of about 30%. For recently contracted capacity, the estimated capex payback period was 22 months. A contract that pays back its direct capital before the halfway point of a 43-month average term leaves room for financing costs, shared infrastructure, replacements and a residual return. It also means prepayments can partly fund the build rather than merely reassure lenders.
But the 22-month measure is a project estimate, not a consolidated return on invested capital. The company did not publish the numerator and denominator in enough detail to reconstruct it. It is unclear how much common data-centre infrastructure, land, power reservation, software development and central overhead sits outside the contract-level calculation. Nor does it reveal renewal pricing once an accelerator generation ages. Those omissions matter because H1 spending was immense: $8.13 billion of purchases of property, equipment and intangibles, against $981.3 million of revenue (Nebius 2026b).
The reaction therefore has a split verdict. A positive repricing was warranted: contract economics, prepayment and cloud margin all improved the evidence. A $17.7 billion one-day increase asks investors to accept that the project metric will survive at fleet level. That part remains unproven.
The business is a capacity allocator with a software layer
Nebius is often described as an AI cloud company. Economically, it is a capacity allocator. It secures power, data-centre shells, accelerator systems and high-speed networking, then wraps those assets in software that lets customers train and run models. The software layer matters because bare GPU rental invites price comparison. Cluster orchestration, networking, storage, inference tools and developer interfaces can improve utilisation and make a workload harder to move. Yet the asset base still sets the cadence: capacity must be financed and installed before it can earn revenue.
The group also owns Avride, an autonomous-driving and delivery-robot business, and TripleTen, a technology education platform. These activities contribute strategic optionality but not the present earnings engine. In Q2, Avride generated $9.3 million of revenue and a $42.0 million adjusted EBITDA loss. TripleTen produced $23.7 million of revenue and a $17.7 million loss. The Nebius cloud segment supplied nearly all group revenue and more than all group adjusted EBITDA (Nebius 2026a).
The reinforcement loop is straightforward. More contracted capacity supports larger hardware orders and stronger access to constrained components. A broader installed base gives the software stack more workloads, which can improve scheduling and utilisation. Better utilisation raises the return per connected megawatt, strengthening the case for more power contracts and cheaper funding. Backlog and prepayments can then reduce the cash gap between construction and service commencement.
The loop can reverse. If deployments slip, the company carries debt and depreciating hardware before billing begins. If a large customer delays, backlog does not become cash on schedule. If accelerator supply improves across the industry, rental prices may fall while financing obligations remain fixed. A full-stack platform softens those risks; it does not remove them.
Management expects 1 gigawatt of connected power by year-end within 2.5 gigawatts already under contract. The scale is material, but contracted power is not the same as energised, commissioned and revenue-producing capacity. The gap between those categories is where construction execution, utility timing and capital discipline show up.
Four years of accounts show the cost of arriving early
The history below uses the continuing Nebius-era business as restated in the 2024 and 2025 annual reports. That distinction is essential. The predecessor group contained Russian operations that were divested; comparing old consolidated revenue with today's cloud platform would create a false growth record. Adjusted EBITDA is company-defined. Capex is filed purchases of property, equipment and intangibles. The final column is an author-computed gross-debt-to-cash ratio, used because conventional net-debt-to-EBITDA is meaningless while EBITDA is negative or changing by hundreds of millions (Nebius 2024; Nebius 2025).
| Period | Revenue ($m) | Adjusted EBITDA ($m) | Capex ($m) | Gross debt / cash (x, computed) |
|---|---|---|---|---|
| 2022 | 13.5 | (124.5) | 14.6 | 2.52 |
| 2023 | 9.8 | (282.8) | 82.9 | 0.06 |
| 2024 | 91.5 | (266.4) | 807.5 | 0.00 |
| 2025 | 529.8 | (64.9) | 4,066.0 | 1.12 |
| H1 2026 | 981.3 | 365.7 | 8,130.3 | 1.06 |
The table is less a normal operating history than a capital formation record. Revenue grew almost sixfold in 2025 while the EBITDA loss narrowed. H1 2026 then crossed into positive adjusted EBITDA. At the same time, annual capex rose from $14.6 million in 2022 to $4.07 billion in 2025, followed by twice that amount in six months. The accounting inflection is genuine. So is the financing intensity.
Group adjusted EBITDA also needs separating from cloud economics. H1 adjusted EBITDA was $365.7 million, but Q2 alone contributed $236.2 million. The core segment's Q2 margin approached 50%; group costs and loss-making options pulled the consolidated number lower. That is still powerful operating leverage. It is not yet an owner-return measure because depreciation, interest, working capital and the build programme sit below or outside adjusted EBITDA.
The H1 balance sheet held $8.04 billion of cash and $1.06 billion of restricted cash. Current and non-current debt totalled about $8.55 billion. Cash roughly matched gross debt, but this apparent balance is the result of a financing wave rather than accumulated free cash. During the half, Nebius issued $4.34 billion of convertible notes, raised $2.85 billion through treasury shares and received $2.0 billion from pre-funded warrants (Nebius 2026b). Capital access has been a competitive advantage. It is also part of the per-share cost of growth.
Owner cash is negative after a $4.4 billion working-capital assist
The statutory cash-flow statement is where the Q2 excitement meets the build. H1 operating cash flow was $4.50 billion. Purchases of property, equipment and intangibles were $8.13 billion. The simple owner-cash bridge is therefore:
| H1 2026 cash bridge | $bn |
|---|---|
| Operating cash flow | 4.50 |
| Less: property, equipment and intangible purchases | (8.13) |
| OCF less capex, author-computed | (3.63) |
That deficit is severe but needs interpretation. Deferred revenue increased by $4.40 billion during the half, largely reflecting customer prepayments. Strip out that working-capital inflow and operating cash was close to flat before capex. Prepayments are economically valuable: they lower funding needs and signal commitment. They are not profit. Nebius still owes future computing service against the cash received (Nebius 2026b).
This distinction makes the payback claim central. If the 22-month period measures all direct capital and the contracts remain firm, today's negative owner cash is the front end of an attractive annuity. If it excludes a material share of common infrastructure or assumes utilisation that is not achieved, prepayments merely delay the financing problem.
ROIC would give false precision here. The denominator is expanding too quickly, a large share of the installed base has not reached steady utilisation, and the return numerator is flattered by prepayments while depreciation trails capex. The useful interim measures are capex payback, adjusted EBITDA per connected megawatt, cash generation after removing deferred-revenue movements, and the amount of new equity or debt required per dollar of incremental ARR.
On the evidence available, incremental economics look stronger than the consolidated accounts. Q2 cloud adjusted EBITDA of $222.0 million on $444.6 million of revenue supports management's claim that energised capacity can earn high margins. The missing bridge is from that segment margin to cash after all shared capital and financing costs.
Scarcity is a moat until supply catches up
Nebius currently has three defensible advantages. The first is power. AI infrastructure projects are constrained by grid connections, permitting, cooling and equipment lead times. Contracting 2.5 gigawatts gives Nebius a larger future footprint than its current revenue would suggest. The second is deployment capability. Revenue growth from $91.5 million in 2024 to $529.8 million in 2025, then $981.3 million in H1 2026, shows that the company can convert at least part of its plan into live service. The third is capital access. Few young platforms could fund an $8.1 billion half-year build while retaining cash close to gross debt.
The financial evidence supports a widening capacity moat and a stable software moat, but not a permanent one. CoreWeave reported $2.58 billion of Q2 revenue and $4.65 billion for H1, showing that Nebius is not alone in raising capital and converting AI demand into contracted infrastructure (CoreWeave 2026). Hyperscalers possess broader customer relationships, internal chip programmes and lower funding costs. Hardware suppliers also capture part of the scarcity rent.
Customer concentration is the main counter-evidence. Large multi-year contracts can transform reported ARR overnight, but they also concentrate renewal and counterparty risk. The prepayment structure reduces credit exposure while increasing delivery obligations. A small number of contracts can make both growth and margin appear smoother than the underlying asset life.
Technology ageing creates another leak. A 22-month payback is attractive only if the useful revenue period comfortably exceeds it and refresh capital does not absorb the residual. New accelerator generations can command premium pricing; older systems may migrate to inference or lower-priced workloads. Nebius has not yet shown a full hardware cycle as a standalone Western cloud operator.
The macro backdrop cuts both ways. Financing costs remain material for long-duration infrastructure, while demand for AI compute is expanding fast enough to attract every major cloud and specialised provider. Rate data provide no reason to treat capital as free (FRED 2026). The moat will be visible in achieved returns after financing, not in the size of the construction budget.
Management has earned operating credit, not a blank cheque
Management's 2025 and H1 2026 execution deserves credit. The company moved from a complex corporate separation to a scaled cloud build, crossed into positive adjusted EBITDA and signed contracts that materially exceed current run-rate revenue. Raising capital before the demand arrived also reduced immediate liquidity risk.
The capital-allocation audit is tougher. The group is funding three different propositions: a hyperscale AI cloud, autonomous mobility and technology education. Cloud economics are becoming legible. Avride and TripleTen still consume adjusted EBITDA. Those options may become valuable, but every dollar assigned to them competes with data-centre capacity whose contract returns are already visible.
Share issuance also changes the owner outcome. The market snapshot uses about 268 million shares, including the effect of pre-funded instruments where appropriate. Future convertibles, stock compensation and additional equity can make enterprise growth outrun per-share value. Scenario work must therefore use rising diluted share counts rather than hold today's denominator constant.
The strongest anti-thesis is not that demand disappears. It is that management correctly identifies enormous demand, earns good project-level margins, and still produces modest per-share returns because the programme requires repeated external capital. That path would reconcile excellent operational headlines with disappointing owner cash.
The valuation now discounts a much larger 2030 company
A conventional near-term earnings multiple is a poor fit. FY2026 revenue is guided to $4.5-$4.7 billion, yet current capex is building capacity for later years, and reported profit is distorted by financing and fair-value items. A staged enterprise-value-to-EBITDA framework is more useful, with explicit revenue, steady-state margin, exit multiple and dilution assumptions. It remains a coarse method; the ranges are wide on purpose.
The shares closed at $259.20. On 268 million shares, equity value is about $69.5 billion. June cash of $8.04 billion and debt of roughly $8.55 billion put enterprise value near $70.0 billion before restricted cash and other adjustments. Capitalise that value to 2030 at 12%, then apply a 12-15 times EBITDA multiple and a 32-36% margin. The quoted valuation implies roughly $21-$29 billion of 2030 revenue, depending on the exact combination. From the midpoint of FY2026 guidance, that is about 46-59% annual growth for four years.
The four scenarios use that reverse calculation as a cross-check, not an anchor:
| Case | Operating frame | Per-share range |
|---|---|---|
| Severe downside | $15-$18bn 2030 revenue; 18-25% margin; 8-10x; heavy dilution | $70-$145 |
| Bear | $20-$25bn; 25-30% margin; 10-12x; continued external funding | $160-$280 |
| Base | $27-$34bn; 32-36% margin; 12-15x; moderate dilution | $320-$520 |
| Bull | $34-$40bn; 38-42% margin; 15-18x; 22-26 month paybacks persist | $620-$950 |
These are author estimates in US dollars. They subtract approximately $0.5 billion of June net debt and use diluted share counts that rise from about 280 million in the severe case to roughly 310 million in the bull case, reflecting the paradox that faster construction can require more funding. The ranges are not forecasts. They map the operating outcomes consistent with different per-share values.
Two variables dominate. At $27 billion of 2030 revenue and a 32% EBITDA margin, moving the exit multiple from 10 to 15 times changes the undiscounted enterprise value from about $86 billion to $130 billion. At a fixed 12 times, moving margin from 25% to 40% shifts that value from $81 billion to about $130 billion. Financing and dilution then determine how much reaches each share.
The post-result valuation sits above the bear range and below the base range. It does not assume failure. It assumes Nebius can become one of the world's larger AI infrastructure operators while preserving a substantial part of today's project economics. The 454% revenue increase makes that outcome conceivable. The cash-flow statement shows why it is not yet established.
Three disclosures will decide whether the rally was proportionate
The first crux is payback reconciliation. Q3 and FY2026 reporting should show whether management repeats the 22-month measure and provides a fleet-level bridge. A period above 30 months, or disappearance of the disclosure, would weaken the return case more than a modest revenue miss. A stable measure alongside broader customer mix would strengthen it.
The second is owner cash after prepayments. H1 OCF less capex was negative $3.63 billion even after a $4.40 billion deferred-revenue inflow. The next two halves will show whether the deficit narrows as commissioned capacity begins billing, or whether each wave of ARR requires another financing wave. This is the cleanest test of self-funding.
The third is core cloud margin through deployment. Q2's 49.9% segment adjusted EBITDA margin is strong. Sustaining more than 35% while bringing new sites online would indicate that software, utilisation and pricing survive scale. Two quarters below that level would point to construction drag, weaker utilisation or price competition.
There are also hard balance-sheet signals. Track gross debt and the fully diluted share count together. Debt alone understates the cost when convertibles and pre-funded warrants finance the build; share count alone misses fixed claims. Compare the combined increase with ARR, connected megawatts and owner cash.
Source notes and the missing payback bridge
Confidence is high for the filed numbers, market move and company identity. The Finance API sidecar matched Nebius Group N.V., NASDAQ:NBIS and the 12 August close, and its filing index was reconciled to EDGAR. The main missing information is analytical rather than documentary: the company has not published enough detail to reconstruct contract-level payback, allocate shared infrastructure to each cohort or measure renewal economics through a complete GPU cycle.
The 34.1% rally was not a simple celebration of triple-digit growth. Q2 reduced the risk that Nebius is building capacity without attractive demand. It did not settle whether attractive contracts can cover the full $8.1 billion build, financing costs and dilution. At $259.20, the market is pricing the 22-month project claim as the opening evidence of a durable fleet return. The next cash-flow statements will show whether that inference holds.
References
- Nebius Group N.V., Q2 2026 results and shareholder materials (Nebius 2026a).
- Nebius Group N.V., H1 2026 unaudited condensed statements (Nebius 2026b).
- Nebius Group N.V., Q1 2026 financial statements (Nebius 2026c).
- Nebius Group N.V., Q2 2026 Form 6-K capital disclosures (Nebius 2026d).
- Nebius Group N.V., 2025 Annual Report on Form 20-F (Nebius 2025).
- Nebius Group N.V., 2024 Annual Report on Form 20-F (Nebius 2024).
- SEC issuer record and XBRL company facts (SEC 2026a; SEC 2026b).
- Yahoo Finance, NBIS daily market data for 12 August 2026 (Yahoo Finance 2026).
- AFP, report on Nebius Q2 results and cloud outlook, 12 August 2026 (AFP 2026).
- CoreWeave, Inc., Q2 2026 Form 10-Q (CoreWeave 2026).
- Federal Reserve Bank of St. Louis, US short-rate series (FRED 2026).