Nebius Q2 2026: revenue grew 454%; capital spending was 9.7x revenue

Nebius grew revenue 454%, turned adjusted EBITDA from a $21.0 million loss into a $236.2 million profit, and lifted annualised run-rate revenue from $1.9 billion to $3.0 billion in ninety days.

In the same three months it spent $5.66 billion on property and equipment — 9.7 times the revenue it booked (our arithmetic).

	Q2 2026
Revenue	$582.3M	+454% from $105.1M
— AI cloud segment	$574.9M	~98% of the group
Adjusted EBITDA	$236.2M	−$21.0M a year ago
Loss from operations	$(175.9)M	widened from $(111.2)M
Diluted EPS, continuing ops	$(0.68)	vs ~$(0.72) consensus (FinanceFeeds)
ARR at 30 June	$3.0B	$1.9B at 31 March
Operating cash flow	$2.25B	3.9× revenue (our arithmetic)
Deferred revenue	$5.98B	$4.78B at 31 March
Capital purchases	$5.66B	9.7× revenue (our arithmetic)

Two weeks ago we wrote that SpaceX spending 2.4 times revenue on capital expenditure was the most aggressive build in the market. Nebius is spending roughly four times more aggressively than that, relative to revenue.

Understanding why is the whole report.

Did Nebius issue guidance for Q2 2026?

Not for the quarter — Nebius issued no Q2 revenue or earnings guidance in its SEC filings. Its Q1 2026 shareholder letter (Form 6-K, Exhibit 99.2, filed May 13, 2026) reiterated full-year 2026 targets of $3.0–3.4 billion of revenue, $7–9 billion of year-end ARR and an adjusted EBITDA margin of about 40%, and said a detailed view of guidance would be given on the earnings call; the Q2 letter filed August 12 says the company is “reaffirming our full-year 2026 guidance across all metrics.” Against the year-ago quarter, revenue rose 454% to $582.3 million, and diluted EPS from continuing operations was $(0.68) against $2.05, when Q2 2025 carried a $597.4 million investment revaluation gain. What the reaffirmed range demands of the second half is worked through below.

The difference is contracted demand

SpaceX is building compute ahead of customers. Nebius says it is building behind them.

• ARR of $3.0 billion, up from $1.9 billion three months earlier — a 56% increase in a quarter.
• Deferred revenue of $5.98 billion at 30 June ($979.4 million current, $4,995.8 million non-current), up from $4.78 billion at 31 March. That is cash received or billed for services not yet delivered.
• More than $40 billion of customer commitments and an expected more than $9 billion of customer prepayments in 2026, both stated in the Q2 letter.
• Named counterparties in the filings: a second, $27 billion five-year agreement with Meta and a $2 billion equity investment from NVIDIA (both in the Q1 letter), and Reflection and Cohere among four Q2 deals that averaged more than $1 billion of total contract value.

When a company spends this far ahead of revenue, the question is whether the capacity is pre-sold. For a capacity-constrained business with contracted demand, more capex is more future revenue. That differs from a commitment where the demand is speculative — the same headline (“capex up tenfold”) can mean very different things depending on what sits behind it.

That is the bull case, and it is a real one. The rest of this piece is what it costs.

The cash flow is customers’ money, not earnings

Operating cash flow was $2.25 billion on revenue of $582.3 million — 3.9 times revenue (our arithmetic). No business generates operating cash flow at four times its sales from operations. It came from working capital: the cash-flow statement shows $1,197.0 million from the change in deferred revenue and $1,186.8 million from accounts receivable, against a net loss from continuing operations of $190.4 million.

Deferred revenue is a liability. It is customers paying in advance for compute they have not yet consumed, and Nebius now carries $5.98 billion of it. It is genuine cash, it is available to spend today, and it carries no interest, no dilution and no covenants.

It is also not profit, and it has three properties an investor should hold in mind:

1. It is recognised, not earned again. As that $5.98 billion converts to revenue, it produces income statement revenue but no new cash. The cash already arrived.
2. It is front-loaded per contract. A prepayment is a one-time inflow when a deal is signed. Sustaining $2.25 billion quarters of operating cash flow requires signing new prepaid contracts at an accelerating rate — not merely keeping the existing ones.
3. It is a claim on future capacity. If delivery slips, the obligation does not.

This is the sharpest contrast with CoreWeave, which reported the day before and is the closest comparable business in the market. Both are neoclouds building AI infrastructure at enormous scale. Nebius is not debt-free — but the mix is different:

	CoreWeave	Nebius
Debt at 30 June	$35.1B, up 64% in six months	$8.55B (our arithmetic: $46.7M current + $8,499.0M non-current)
Customer prepayments	—	$5.98B deferred revenue
Interest in the quarter	$640M net interest expense	$119.1M interest expense, $95.0M net of interest income (our arithmetic)
What can go wrong	Refinancing, covenants, rates	Contract cancellation, delivery failure — plus the debt service it already carries

Nebius has also funded the build with equity: the Q2 cash-flow statement shows $2,846.7 million from the sale of treasury shares, and the Q1 letter reports $4.3 billion of convertible notes raised in Q1. The Q2 letter says a first secured financing of $775 million at SOFR + 2.50% was raised in July and that more debt transactions are in progress.

Neither model is obviously better. Prepayment financing is cheaper than debt right up until a customer wants its capacity or its money, at which point the obligation is operational rather than financial — and harder to renegotiate.

The depreciation has not arrived yet

Here is the tension the adjusted number hides.

Adjusted EBITDA was positive $236.2 million. Loss from operations was negative $175.9 million, and it got worse year over year, from a $111.2 million loss. The $412.1 million gap between the two (our arithmetic) is, in the company’s reconciliation, exactly depreciation and amortisation of $259.7 million, share-based compensation of $102.5 million and $49.9 million of acquisition and other transaction-related costs.

Now consider what is coming. Property and equipment, net, was $13.0 billion at 30 June, up from $5.6 billion at December, and capital purchases in the first half were $8.13 billion. The SEC filings give no full-year capital-expenditure figure; BigGo Finance reported that on the Q1 call in May management raised its 2026 capex range to $20–25 billion, from $16–20 billion. Assets of that size do not stay off the income statement.

Run the illustration — and it is an illustration, not a depreciation schedule. Nebius says it depreciates server and network equipment over five years. On a five-year straight-line life, the $13.0 billion already on the balance sheet would generate roughly $2.6 billion of depreciation a year (our arithmetic) — before any second-half spending, and treating all of it as five-year equipment, which it is not entirely. Full-year 2026 revenue guidance is $3.0–3.4 billion. Q2’s own D&A of $259.7 million, already 45% of revenue, annualises to about $1.04 billion (our arithmetic).

Depreciation on the assets already built is, on that assumption, heading toward the size of a full year of revenue.

Adjusted EBITDA excludes depreciation. For a software company that owns nothing, that exclusion is harmless. For a company whose entire business is buying GPUs and data centres, it removes the single largest cost of doing business. The 40.6% adjusted EBITDA margin (our arithmetic) is real on its own terms and it is not a margin an equity holder can spend.

The operating loss widening while adjusted EBITDA swung positive by $257.2 million is the first visible instalment of this. It gets larger from here, mechanically, as this year’s spending starts depreciating.

”Reaffirmed” guidance is doing more work than it sounds

Nebius beat the FactSet revenue consensus — and left full-year revenue guidance unchanged at $3.0–3.4 billion.

Unchanged guidance after a beat can look like conservatism. The arithmetic shows why it should not be read that way here.

• H1 2026 revenue: Q1’s $399.0 million plus Q2’s $582.3 million = $981.3 million.
• Full-year guidance of $3.0–3.4 billion therefore implies $2.02–2.42 billion in the second half — between 2.1× and 2.5× the first half (our arithmetic).
• Nebius exited June at a $3.0 billion ARR, which is a $750 million quarterly run rate. The guidance requires H2 quarters averaging $1.01–1.21 billion (our arithmetic).

So the company must not merely sustain its exit run rate; it must lift it by a third or more within two quarters to reach the bottom of its own range. “Reaffirmed” is not a cushion here. It is a demanding target that depends on capacity arriving on schedule — and the Q2 letter itself says most of the quarter’s deals were signed against capacity arriving in late 2026 and “will contribute primarily to 2027 revenue.”

The stated ARR path says the same thing: $7–9 billion by year-end, from $3.0 billion in June. That is another 2.3–3× in six months, on top of the 56% just delivered.

The targets on the record

Target	What the filings say
FY2026 revenue	$3.0–3.4 billion — reaffirmed (Q2 letter)
FY2026 adjusted EBITDA margin	approximately 40%
Year-end ARR	$7–9 billion
Year-end 2026 contracted power	5 GW, up from more than 4 GW (Q2 letter)
Connected power by year-end 2026	800 MW – 1 GW (Q1 letter)
Deployment pace from 2027	more than 1 GW per year (Q2 letter)
2026 customer prepayments	more than $9 billion expected (Q2 letter)
FY2026 capital expenditure	not in the filings; $20–25 billion per BigGo Finance’s report of the Q1 call

The two lines to hold together are contracted and connected power. Capacity is being contracted far faster than it is being energised: a 5 GW contracted target against 800 MW–1 GW of power actually connected by December. Revenue can only be earned on capacity that is live. The gap between contracted and connected power is, in effect, the revenue that exists on paper and not yet in a data centre.

The bear case, stated fairly

1. Capital spending is 9.7× revenue in the quarter and 8.3× revenue across the first half ($8.13 billion against $981.3 million, our arithmetic). There is no margin for a demand pause.
2. Operating cash flow is customer financing. $2.25 billion against $582 million of revenue came from deferred revenue and receivables, not operations.
3. Depreciation is coming. On the company’s five-year equipment life, the $13.0 billion of property and equipment already on the balance sheet implies roughly $2.6 billion a year of depreciation (our arithmetic) — near this year’s revenue guidance — and adjusted EBITDA excludes all of it.
4. The operating loss widened to $175.9 million from $111.2 million even as adjusted EBITDA turned strongly positive.
5. The guidance is demanding, not conservative. H2 quarters must run a third or more above the June exit rate to reach the bottom of the range.
6. Contracted power far exceeds connected power. Revenue follows energisation, not signatures.
7. There is debt too. $8.5 billion of it, costing $119.1 million of interest in the quarter, with more financing planned.
8. The prior-year comparison is distorted — Q2 2025 contained a $597.4 million investment revaluation gain, so year-over-year profit lines are not like-for-like.

The bull case does not need embellishing: 454% revenue growth, ARR up 56% in a quarter, adjusted EBITDA positive at $236.2 million with the AI cloud segment at a 49.7% margin, NVIDIA as a $2 billion investor, Meta as a $27 billion customer, $5.98 billion of prepaid demand on the balance sheet, and net interest of $95.0 million in the quarter (our arithmetic) where CoreWeave pays $640 million net.

What to check at the Q3 report

• ARR against the $7–9 billion year-end path. It went $1.9B → $3.0B. A third datapoint tells you whether the ramp is on schedule or the year-end target is slipping.
• Connected (energised) capacity in megawatts, against the 800 MW–1 GW year-end figure. This is the physical constraint on revenue and the least spinnable number in the release.
• Deferred revenue. If the balance keeps growing from $5.98 billion, new prepaid contracts are still landing and the cash engine runs. If it flattens or falls while capex continues, Nebius starts funding the build from somewhere else.
• Depreciation and amortisation as a line item. Watch the gap between adjusted EBITDA and loss from operations — $412.1 million in Q2. Widening is expected; how fast is the question.
• Revenue against the implied $1.0–1.2 billion quarterly H2 run rate. Q3 is the first real test of the reaffirmed guidance.
• Capital purchases against the $8.13 billion of the first half, and whether the company puts a full-year figure in writing.
• Debt and interest expense. Debt was $8.5 billion at June and the July secured facility adds $775 million; the more the build leans on borrowing, the closer its funding profile moves toward CoreWeave’s.

The read

Nebius delivered a genuinely exceptional quarter — the growth rate, the ARR step-up and the first positive adjusted EBITDA are all real, and the customer roster is as good as this sector offers. Unlike much of the AI buildout, this capacity is, according to the company, contracted before it is built.

But two things are true at the same time. The $2.25 billion of operating cash flow is customers’ money held in advance, not earnings. And the $13.0 billion of equipment and data centres already on the balance sheet — with billions more being bought this year — will land on the income statement as depreciation at a rate that, on the company’s own five-year equipment life, approaches its entire annual revenue.

For anyone holding NBIS, that makes the tracking job unusually physical. Not the growth rate, which will stay spectacular on these comparisons for several more quarters. Watch energised megawatts against the year-end figure, deferred revenue as the measure of whether prepaid demand is still arriving, and the widening gap between adjusted EBITDA and the operating line — because that gap is where the cost of all this capacity eventually shows up.

---

Sources and method. Company figures are from Nebius’s Q2 2026 Form 6-K filed 2026-08-12 (SEC accession 0001104659-26-094568) — the results release (Exhibit 99.1) and the letter to shareholders (Exhibit 99.2) — for the quarter ended 2026-06-30, reported before the open, read directly from SEC EDGAR: revenue of $582.3 million and its 454% growth from $105.1 million, AI cloud segment revenue of $574.9 million and its 49.7% adjusted EBITDA margin, adjusted EBITDA of $236.2 million against a $21.0 million loss, the $175.9 million loss from operations against $111.2 million a year earlier, D&A of $259.7 million, share-based compensation of $102.5 million and acquisition-related costs of $49.9 million, diluted EPS from continuing operations of $(0.68) against $2.05, the $597.4 million prior-year revaluation gain, ARR of $3.0 billion, operating cash flow of $2,246.1 million including $1,197.0 million from deferred revenue and $1,186.8 million from receivables, capital purchases of $5,657.4 million in the quarter and $8,130.3 million in the half, deferred revenue of $979.4 million current and $4,995.8 million non-current, debt of $46.7 million current and $8,499.0 million non-current, interest expense of $119.1 million and interest income of $24.1 million, net property and equipment of $13,045.2 million, $2,846.7 million from treasury-share sales, the five-year useful life for server and network equipment, the $775 million July secured financing, more than $40 billion of customer commitments, more than $9 billion of expected 2026 prepayments, the 5 GW contracted-power target, the reaffirmation of full-year guidance, and the Reflection and Cohere deals. Prior-quarter comparatives and the full-year guidance itself — Q1 2026 revenue of $399.0 million, adjusted EBITDA of $129.5 million, ARR of $1.92 billion, deferred revenue of $4,778.1 million at 31 March, the $3.0–3.4 billion revenue, $7–9 billion ARR and ~40% margin targets, 800 MW–1 GW of connected power by year-end, the $27 billion Meta agreement, the $2 billion NVIDIA investment and the $4.3 billion of convertible notes — are from the Q1 2026 Form 6-K filed 2026-05-13 (accession 0001104659-26-059872), Exhibits 99.1 and 99.2, also read from SEC EDGAR. Nebius issued no quarterly guidance, so whether the quarter beat or missed its own outlook is judged only against the full-year 2026 guidance and the year-ago quarter.

Not company figures, and attributed accordingly: the revenue consensus of $569.9 million per FactSet (as reported by MarketScreener) and $572.75 million per LSEG data (as reported by Reuters, via Yahoo Finance); the EPS consensus of about $(0.72), as reported by FinanceFeeds; and the $20–25 billion 2026 capital-expenditure range, raised from $16–20 billion, which appears in no SEC filing and is reported here as BigGo Finance described the Q1 earnings call. CoreWeave’s debt and interest figures are from our CoreWeave Q2 2026 analysis. This piece makes no claim about the share-price reaction.

Our arithmetic, labelled as such: the ratio of capital purchases to revenue (9.7× in the quarter, 8.3× in the half) and of operating cash flow to revenue (3.9×); the 40.6% group adjusted EBITDA margin; the $412.1 million gap between adjusted EBITDA and the operating loss; total debt of $8.55 billion and net interest of $95.0 million; total deferred revenue of $5.98 billion and $4.78 billion; the implied H2 range of $2.02–2.42 billion, along with the $1.01–1.21 billion implied quarterly run rate and the comparison to the $750 million quarterly rate implied by exit ARR; the $1.04 billion annualised Q2 D&A; and the depreciation illustration, which applies the company’s five-year equipment life to all $13.0 billion of net property and equipment — a simplification we chose, not a company disclosure, and one that overstates depreciation on longer-lived assets and on assets not yet in service. The interpretation — that the spending is backed by contracted demand, that operating cash flow is customer financing rather than earnings, and that reaffirmed guidance is demanding rather than conservative — is our analysis, not a company statement, and is not a forecast.

Related: CoreWeave Q2 2026, the same business funded far more heavily by debt; SpaceX Q2 2026, where capex ran at 2.4× revenue; AMD Q2 2026, the chip side of the same buildout; and Uber Q2 2026, where the capital commitment is for a business with speculative rather than contracted demand.

Tradune shows what you actually hold across every account, and whether a position has beaten a plain index fund since you bought it. Educational analysis, not investment advice.