Executive assessment
Nebius presents a credible financing-dependence hypothesis, but an unproven default and pricing opportunity. The strongest protection thesis is a simultaneous interruption of customer funding and external capital while committed spending remains high. A broad claim that disappointing AI economics automatically causes Nebius to default does not survive the contractual and financing evidence.
The investment argument has four necessary premises. First, economic disappointment must reach Nebius through lower new orders, weaker renewal pricing, customer funding constraints or delivery disputes. Second, cash commitments must remain difficult to reduce. Third, usable cash and financing remedies must be insufficient before obligations fall due. Fourth, the resulting event and creditor loss must be covered by protection purchased at a favorable all-in cost. The first two establish vulnerability; all four are required for an investable CDS position.
The decisive cash flow finding
Reported operating cash generation is heavily supported by customer funding. The diagnostic on page 2 isolates that dependence without treating legitimate prepayments as fictitious revenue. The risk is that replacement advances disappear while Nebius still owes delivery and service on advances already spent. This is a liquidity-timing problem that can coexist with positive EBITDA.
Conditional investment conclusion
In the stress scenario, modeled unrestricted cash falls to $997m in Q4 2026 and a cumulative $2,853m funding deficit appears in Q1 2027 before remedies. The defensive scenario remains positive through Q2 2027. These are analyst-selected scenarios, not forecasts, user-supplied estimates or assessed probabilities. Quarter-by-quarter spending, advances and current liquidity must be verified before assigning a likely failure date.
The preferred research target is Nebius Group N.V. unsecured parent credit. Protection on a subsidiary GPU loan would cover different assets and obligations. Parent CDS availability remains unverified: Reuters reported no publicly tradable Nebius CDS on October 15. A bespoke market may exist, but neither its existence nor its price is established here. [15]
What would make the thesis actionable
Obtain a dated cash-use schedule, demonstrate a deficit after realistic remedies, identify the covered legal obligation and obtain a firm two-way protection quote. The report supplies a reproducible stress framework and a conditional premium ceiling; it does not manufacture an executable trade from an economic concern.
Cash flow quality and funding dependence
Historical operating and investment bridge
| USD millions | Q1 2026 derived | Q2 2026 reported | H1 2026 |
|---|---|---|---|
| Operating cash flow | 2258.0 | 2246.1 | 4504.1 |
| Equipment and intangible purchases | -2472.9 | -5657.4 | -8130.3 |
| Operating cash less these purchases | -214.9 | -3411.3 | -3626.2 |
Source [1]. Q1 equals H1 less Q2. This cash investment measure is not maintenance capex; expansion dominates the business model. A deficit during contracted expansion can be rational if subsequent receipts and financing are dependable.
Cash reconciliation and diagnostic
| H1 control [2] | Amount |
|---|---|
| Opening cash including restricted | 3721.6 |
| Operating | 4504.1 |
| Investing | -8288.8 |
| Financing | 9160.4 |
| Foreign exchange | 0.8 |
| Closing including restricted | 9098.1 |
| Restricted portion | 1056.0 |
| Unrestricted cash and equivalents | 8042.1 |
H1 deferred-revenue cash-flow adjustment: $4,395.0m. Subtracting it from operating cash leaves $109.1m; also removing the $427.9m receivables inflow leaves negative $318.8m. Deferred revenue includes potentially refundable amounts. [2]
These subtractions are diagnostic, not normalized free cash flow. Deferred-revenue movements combine advances and recognition of previously received cash; removing them is not the same as estimating operations with no new advances. The residual also includes other working-capital movements, taxes and cash interest. A forecast needs gross customer receipts, recognition, refunds and service costs separately.
Management described three contract durations: three-to-six-month transactions, core contracts averaging one-to-three years, and longer investment-grade agreements. It expected more than $9bn of 2026 prepayments. [3] The short case concerns what happens after those funding receipts, especially if renewal economics deteriorate faster than spending can adjust.
Avoid four accounting errors
Do not add advances to operating cash flow when they are already included. Do not count previously received deferred revenue again as future cash. Do not subtract the entire deferred-revenue liability from liquidity as if all customers can demand repayment today. Do not infer a current cash balance by adding August financing to June cash without deducting subsequent payments.
Parent debt and maturity exposure
All eight convertible series below are parent obligations. This issuance-based pro forma assumes the announced $400m exchange of each June 2025 series completes and no other conversion or repurchase. It is not a trustee-certified outstanding balance. Original principal, carrying value and maturity redemption amount are different quantities. [4–9]
| Maturity and series | Original principal | Coupon | Maturity cash* |
|---|---|---|---|
| June 2029 / June 2025 | 100 | 2.00% | 120 |
| February 2030 / August 2026 | 3450 | 0.50% | 3795 |
| September 2030 / September 2025 | 1581.25 | 1.00% | 1818.44 |
| March 2031 / March 2026 | 2587.5 | 1.25% | 3105 |
| June 2031 / June 2025 | 100 | 3.00% | 125 |
| September 2032 / September 2025 | 1581.25 | 2.75% | 1818.44 |
| March 2033 / March 2026 | 1750 | 2.625% | 2100 |
| February 2034 / August 2026 | 2300 | 4.50% | 2875 |
| Total | 13450 | Annual coupon 263.33 | 15756.88 |
*Contractual maturity accretion applied to assumed residual original principal, absent conversion, redemption or repurchase; excludes coupons. Sources: June terms [4], September issuance and terms [5,6], March pricing and closing amounts [7,2], August closing and terms [8,9].
The maturity uplift is $2,306.9m above this original-principal pro forma. The annual coupon calculation is sum(original principal × coupon). It excludes subsidiary loans, fees, contingent interest and customer-financing accretion. Accretion is a future repayment burden, not an extra current cash coupon.
The timing implication for protection
About $5.61bn of parent maturity payments in this pro forma fall in 2030, followed by $3.23bn in 2031. Protection expiring before those dates primarily expresses an early liquidity or acceleration thesis. A five-year tenor from October 2026 reaches the main 2030–2031 concentration but leaves the 2032–2034 maturities outside its life; the confirmation must specify the exact termination date.
A strong stock price can facilitate conversion or equity-financed repayment. In a downturn, that relief can weaken as investors prefer cash repayment. This creates a conditional interaction between credit and equity markets; conversion is not guaranteed debt forgiveness, and a falling share price alone is not a credit event.
Outstanding balance reconciliation is a required gate
The June financial table reports $8,499.0m carrying value and $10,041.8m in its principal column. [2] Its older-note figures do not mechanically reconcile to original issue sizes multiplied by disclosed final accretion factors. Obtain trustee balances and schedules before using the pro forma for settlement, leverage or recovery. If the August exchanges did not settle, add $800m original principal, $980m maturity cash and $20m annual coupon to the table.
Secured debt and contractual commitments
The GPU financing sits in separate borrowers
Nebius Compute II LLC and Nebius Compute II Oy entered a $775m secured facility, maturing October 31, 2030, at Term SOFR plus 2.50%. Security covers substantially all borrower assets and their shares, subject to exceptions. The disclosed DSCR covenant is 1.15x; a minimum-liquidity covenant also applies. Parent support covers specified bad acts and certain manager and colocation performance obligations, not an expressly disclosed blanket repayment guarantee. [10]
At an assumed 4% SOFR and unchanged full balance, annual cash interest would be $50.375m; each 100bp increase adds $7.75m. These are sensitivities, not current rates or amortization forecasts. The actual remaining principal schedule, hedges, cash sweeps, liquidity minimum and guarantee triggers require the facility documents.
Commitments beyond bond principal
| June disclosure [2] | USD millions |
|---|---|
| Operating lease liability / undiscounted payments | 1,625.3 / 2,175.1 |
| Uncommenced leases / prepayments | 12,054.1 / 612.4 |
| 2027 power and related minimum commitments | 541.8 |
| Equipment and intangibles acquired but unpaid | 973.3 |
Uncommenced leases were expected to start in 2026–2027, with terms up to twelve years. [2] They are not all immediately due. The correct risk measure is the dated unavoidable payment schedule, after enforceable cancellation or renegotiation rights. Adding all nominal leases, power commitments and bonds into one near-term funding gap would be misleading.
In the model, routine rent, electricity, taxes and cash interest belong in residual operating cash. Equipment invoices belong in cash capex when paid. Neither the unpaid equipment amount nor total future leases are added again as a one-off outflow. Separately model only incremental refunds, collateral calls or other exceptional cash demands not already in operating cash or capex.
How subsidiary stress can reach parent notes
The August 2026 indenture for the 2030 notes includes cross-default provisions for at least $100m of borrowed debt of the parent or a Significant Subsidiary, involving nonpayment after applicable grace or acceleration, followed by a 30-day notice cure process. Significance is defined in the instrument. Parent-note interest nonpayment has a 30-day grace period. [11, section 7.01]
A facility covenant breach is therefore not automatically a parent default. Test borrower significance, acceleration, notice, cure and guarantee scope. Even a parent bond event of default is not automatically a payable CDS event: the protection contract has its own definitions and settlement conditions. Obtain legal review of the selected reference obligation and any subsidiary transmission route.
Customers and the failure modes that matter
The central analytical question is which cash flows remain collectible after disappointing AI economics. Model each cohort separately: protected hyperscaler commitments, AI-native term contracts and short-duration capacity sales. A portfolio-wide revenue haircut conceals these differences.
Meta is an important defense against the short
The March 2026 agreement comprises $12bn of dedicated capacity and an arrangement for up to $15bn of specified unsold capacity. Meta must buy qualifying unsold capacity for the remainder of five years from initial deployment. The filing also discloses delivery-related fee discounts, termination and service-level provisions. [12]
This protection can interrupt the link from weak third-party demand to lost cash receipts. Its force depends on covered clusters, available capacity, delivery, pricing, exclusions and enforceability. Do not treat the full $27bn as incremental cash on top of third-party sales into the same capacity. The strongest adverse case is failure to deliver qualifying capacity or economic erosion outside the covered pool.
| Failure mode | Mechanism benefiting protection | Evidence needed |
|---|---|---|
| Advance receipts slow | New funding stops while past obligations remain | Gross advance receipts and dates; refund rights; customer financing |
| Delivery slips | Payments delayed; discounts, costs or termination claims | Milestones, acceptance certificates, cure periods, contractor remedies |
| Renewal pricing falls | Shorter contracts reset before assets are repaid | GPU cohort utilization, prices and renewal margins |
| Secured financing tightens | Lower advance rates or trapped cash reduce parent access | Eligibility, DSCR tests, sweeps, guarantees and collateral valuation |
| Equity and converts close | Refinancing becomes expensive as conversion relief weakens | Firm financing terms and actual issuance capacity |
| Asset realizations disappoint | Unsecured recovery falls after senior claims | Asset ownership, liens, forced-sale bids and subsidiary liabilities |
These are proposed causal mechanisms, not findings that the events have occurred. The most damaging combination is correlated: weaker AI customer funding reduces advances, slows collections, lowers used-equipment values and closes capital markets together. A single isolated price decline may be absorbed.
Depreciation and replacement spending
Nebius extended server and network useful lives from four to five years in 2026. [13] That changes accounting expense, not physical durability or contractual collections. The investment question is whether equipment remains commercially competitive and cash-generative long enough to repay financing after power, refresh and replacement spending. Test new and older GPU cohorts separately; an accounting write-down itself is not a CDS trigger.
Quarterly liquidity stress through June 2027
This is an unrestricted consolidated cash sensitivity, not parent-only cash or a GAAP forecast. Opening June cash is $8,042.1m. Q3 includes $5,680m estimated net August proceeds and assumes the full $775m facility becomes usable cash: $6,455m financing in total. [2,9,10] Facility draw status, fees and cash restrictions need verification; removing that assumption reduces each closing balance by $775m.
All figures below are analyst assumptions. Q3 is elapsed but unreported in the reviewed statements. No other equity, debt, asset sale or restricted-cash release is assumed. Advance cash is entered separately from residual operating cash; that residual includes all other customer collections less operating cash costs, interest and taxes. No automatic repayment of all deferred revenue is assumed.
| Inputs by Q3 26 / Q4 26 / Q1 27 / Q2 27 | Defensive | Stress | Severe |
|---|---|---|---|
| New advances | 2500 / 2500 / 2000 / 2000 | 1500 / 500 / 500 / 500 | 500 / 0 / 0 / 0 |
| Residual operating cash | 250 / 250 / 400 / 400 | 0 / 0 / -100 / -100 | -100 / -200 / -200 / -200 |
| Cash capex paid | 6000 / 6000 / 5000 / 5000 | 7000 / 8000 / 4000 / 3500 | 8000 / 9000 / 3000 / 2000 |
| Exceptional cash demands | 0 / 0 / 0 / 0 | 0 / 500 / 250 / 0 | 0 / 1000 / 500 / 0 |
Closing cash = opening cash + financing + advances + residual operating cash − capex − exceptional demands. Later capex reductions in the adverse cases assume partial management response; they do not establish that all modeled spending is legally unavoidable.
| Closing cash before remedies | Q3 2026 | Q4 2026 | Q1 2027 | Q2 2027 |
|---|---|---|---|---|
| Defensive | 11,247.1 | 7,997.1 | 5,397.1 | 2,797.1 |
| Stress | 8,997.1 | 997.1 | (2,852.9) | (5,952.9) |
| Severe | 6,897.1 | (3,302.9) | (7,002.9) | (9,202.9) |
Negative values are cumulative unfunded uses, not feasible cash holdings. A $1bn analytical cash floor is a sensitivity, not a legal covenant. Stress first falls $2.9m below that floor in Q4 2026, a rounding-level result that should not be treated as a precise signal. It requires $3,852.9m of cumulative remedies to restore the floor by Q1 2027; $2,852.9m avoids negative cash.
What the cases establish
The model identifies the joint conditions necessary for pressure. It does not establish their probability or prove a funding gap. Actual fungibility may make parent cash lower than consolidated cash; new financing or elective spending cuts may make liquidity materially stronger. Reconcile bank balances and committed payments before upgrading a scenario into a forecast.
Reverse stress and recovery analysis
How much adverse spending can liquidity absorb
June cash plus assumed Q3 financing totals $14,497.1m. Maintaining a $1bn buffer permits $13,497.1m of cumulative net uses from July onward. For H2, the general test is: cash capex + exceptional demands − advances − residual operating cash > $13,497.1m. Beyond this point additional financing, receipts or spending reductions are required to preserve the analytical buffer.
At stress H2 capex of $15bn, $0.5bn exceptional demands and zero residual operating cash, advances or other remedies must total $2,002.9m to retain the buffer. The assumed $2bn advances almost exactly meet that requirement. Zero advances under the same spending produce a $1,002.9m cash deficit. This is a threshold calculation, not evidence that advances will stop.
The practical rescue test
| Possible remedy | How it changes the conclusion |
|---|---|
| Defer $1bn of genuinely optional capex | Adds $1bn liquidity; verify penalties and lost contracted revenue |
| Collect $1bn more advances | Adds liquidity now but creates future service obligations |
| Raise $1bn equity net | Adds liquidity without debt service; dilution is not creditor loss |
| Raise $1bn secured debt net | Adds liquidity, but may encumber assets or trap future cash |
| Sell noncore assets | Count net proceeds only after timing, tax, restrictions and claims |
| Improve collections | Add only receipts not already in the scenario |
An unavoidable funding gap exists only after feasible remedies are insufficient or too late. Strategic customers and suppliers may have incentives to fund completion. The short case must explain why they would refuse, why aid would be inadequate, or why its terms would still impair the reference debt.
Unsecured recovery needs an entity waterfall
Value subsidiary collateral against secured claims and local liabilities first. Only legally distributable residual value and enforceable intercompany recoveries reach the parent. Then add parent cash and realizable investments, subtract administration and senior claims, and allocate among equal-ranking creditors. Customer refund claims may compete; their entity and priority require legal analysis. Do not count both gross subsidiary enterprise value and the parent equity interest in it.
Illustration only: assume $8bn realizable value accessible to parent unsecured creditors after all senior and subsidiary claims, and $14bn allowed equal-ranking claims. Recovery is 57.1%. At $5bn value it is 35.7%; at $11bn it is 78.6%. These are not Nebius valuations. The debt denominator at a future credit event must use the then-accreted allowed claim, not automatically today’s original principal or final maturity amount.
New secured borrowing can extend survival while reducing assets available to unsecured creditors. Whether that helps a CDS buyer depends on tenor, premium carry and eventual settlement recovery. A refinancing that postpones failure beyond protection expiry can defeat the trade even if later losses become larger.
Protection economics and execution gates
Reuters reported no publicly tradable Nebius CDS on October 15. [15] No firm quote, reference-entity code, deliverable list, dealer commitment or secondary exit was verified. A bespoke protection contract is a research candidate. A short convertible is not an equivalent substitute because equity delta, volatility, borrow costs and recall risk affect its return.
Illustrative premium ceiling
Use a simple two-outcome model to make the price discipline explicit. Assume five-year protection, a covered default at year three with probability p, recovery R, no upfront payment, and running annual premium s paid until default or expiry. Ignore discounting, accrued settlement premium, counterparty losses and trading costs. Expected profit per dollar of notional = p(1−R) − s[3p + 5(1−p)].
| Assumed five-year default probability | 20% recovery | 40% recovery | 60% recovery |
|---|---|---|---|
| 10% | 167 bp | 125 bp | 83 bp |
| 20% | 348 bp | 261 bp | 174 bp |
| 30% | 545 bp | 409 bp | 273 bp |
Each cell is the break-even annual spread under those assumptions, not a market quote or probability estimate. For p=20%, R=40% and $10m notional, expected protection receipts are $1.2m. At 200bp, expected premium is $0.92m, leaving $0.28m before costs; at 300bp, expected premium is $1.38m, leaving negative $0.18m. A costly contract can be unattractive despite a correct bearish thesis.
For a contract with upfront cost U per dollar notional and expected other costs K, subtract U + K from the numerator when solving for s. Standard-coupon/upfront quotes require survival-weighted discounting and the exact settlement conventions. Spread widening only creates a realizable trading profit if an executable unwind exists; it is not a contractual default payout.
Required dealer and legal diligence
Request indicative three- and five-year two-way terms on Nebius Group N.V., with exact expiry, reference obligation, entity identifier, notional minimum, upfront plus running premium, reference-obligation substitution, convertible eligibility and accreted-principal treatment. Specify credit events and restructuring elections, failure-to-pay thresholds and grace periods, deliverable obligations, physical or cash settlement, valuation disputes and fallback when no auction occurs.
Separately verify counterparty eligibility, ISDA and collateral documentation, margin liquidity, collateral interest, dealer creditworthiness and unwind costs. A protection seller exposed to the same AI downturn can introduce wrong-way risk. Loan syndicate membership does not demonstrate willingness to sell CDS. No counterparties were contacted.
Decision rule for Theseus
Advance to a proposed trade only when current cash and payment evidence supports a loss distribution, the proposed contract captures that loss, and a firm price is below the risk-adjusted value after carry and execution costs. Until then classify Nebius as a researched parent-credit watchlist candidate, with delivery and advance-funding risks as the leading hypotheses.
Evidence that would confirm or reject the thesis
| Priority | Evidence to obtain | Decision consequence |
|---|---|---|
| 1 | Cash by legal entity; restricted balances; July-to-date cash bridge | Replaces stale consolidated starting cash |
| 1 | Gross advances received versus scheduled; refund and termination terms | Tests whether customer funding is dependable |
| 1 | Weekly equipment, construction and lease cash commitments; cancellation rights | Separates unavoidable needs from optional growth |
| 2 | Delivery milestones, acceptance, service credits and customer correspondence | Tests the proposed disruption mechanism |
| 2 | Full secured facility; draw balance, amortization, covenants and sweeps | Measures cash trapping and parent recourse |
| 2 | Trustee-certified note balances; exchange settlement; accretion tables | Corrects the provisional debt map |
| 3 | GPU-cohort margins, renewals, residual bids and refresh spending | Tests long-run repayment and recovery |
| 3 | Firm CDS terms or convertible borrow with depth and exit costs | Determines whether the thesis can be implemented |
Evidence that would strengthen the protection case
Advances repeatedly miss contractual dates; delivery slips while invoices continue; capacity-backed financing is refused or materially repriced; cash becomes trapped at operating subsidiaries; short-duration renewals lose margin after power and refresh costs; and equity issuance no longer covers the dated cash requirement. The strongest evidence would be a reconciled funding deficit that remains after identified remedies, accompanied by a contract covering the exposed parent debt.
Evidence that would defeat or postpone it
On-time deployment, reliable hyperscaler collections, repeat secured financing, conversion or equity issuance, and successful reduction of uncommitted spending can preserve liquidity. Asset sales may provide another buffer. If these actions carry the company through the spending peak, a near-term protection trade can lose its full premium even if industry-wide AI returns eventually disappoint.
What the current research proves
It establishes a financing-sensitive business model, specific legal pathways worth testing, a provisional maturity map and explicit liquidity thresholds. It does not establish present insolvency, a probable default quarter, an unconditional customer refund obligation, an asset liquidation value or mispriced protection. Maintaining that distinction is necessary to turn Theseus’s macro thesis into instrument-level underwriting.
Research limitations that affect valuation
The latest statements examined are June 2026 accounts with later disclosed financings. The model does not assume October cash equals June cash. Private customer schedules, the full secured-loan agreement and a current dealer market are necessary inputs to a final investment decision. Numerical scenarios are intentionally transparent sensitivities; no invented user estimates or probability weights have been assigned.
Source register
[1] Q2 earnings release; operating cash and cash purchases
[2] June 2026 financial statements; cash flow and notes 4, 8, 11, 12
[3] Q2 shareholder letter; contract duration and funding
[4] June 2025 original note terms
[5] September 2025 closing and exercised options
[6] September 2025 accretion and maturity terms
[7] March 2026 convertible pricing and terms
[8] August 2026 closing and announced exchanges
[9] August 2026 pricing and estimated net proceeds
[10] July 17 facility filing; borrowers and parent support
[11] August 2026 indenture for 2030 notes; defaults and remedies
[12] March 16 Meta agreement disclosure
[13] Operating and financial review; useful life and liquidity
[14] August 2026 indenture for 2034 notes
[15] Reuters October 15; Nebius CDS availability
Calculations are reproducible from the displayed inputs. Debt totals are issuance-based pro formas; source accounting tables use different measurement bases. All monetary values are USD millions except where stated. CDS: credit default swap. DSCR: debt-service coverage ratio. SOFR: secured overnight financing rate. EBITDA: earnings before interest, tax, depreciation and amortization. A basis point is 0.01 percentage point.