Nebius: A $40 Billion Backlog, and a Price That Has Already Priced the Bull Case

Both camps read the same Q2 2026 print and reached the opposite conclusion. Nebius Group (NBIS) reported revenue of $582 million, up 454% from a year earlier, a 50% EBITDA margin on the AI cloud segment, and a committed customer backlog that management now puts above $40 billion. Bulls saw a contracted-growth machine that the market is still underpricing. Bears saw a stock that is up 187% year to date trading at roughly twenty times next year's sales while still spending $11 billion on capex against $5 billion of operating cash flow, and concluded the good news is in the price. At $240, one of those two readings has to be wrong.

The disagreement is not about the facts. Both sides sign for the same revenue, the same margins, the same Microsoft and Meta contracts. The fight is over what $40 billion of backlog and a 50% segment margin are worth when the company is still burning capital to deploy them and the stock has already moved to price in flawless execution.

What the two camps share

Anchor the record first.

As of the morning of September 10, 2026, Nebius trades around $240, a market cap of roughly $65 billion, and an enterprise value close to the same number. Year to date the share is up 187%; over the past five sessions it has gained another 20%. Trailing-twelve-month revenue is about $1.35 billion, which makes the headline price-to-sales multiple a staggering 48x. That number looks absurd until you put the forward numbers in the same denominator.

Management guided full-year 2026 revenue to $3 to $3.4 billion. Against a $65 billion market cap, the stock trades at about 19 to 22 times what it expects to bill this year — closer to 20x at the $3.2 billion midpoint. That is the honest multiple, and it is the number the bear case is built on.

On the operating side, Q2 2026 was unambiguous. Total revenue of $582.3 million was up 454% year over year. The AI cloud segment — now 98% of the business — produced $574.9 million of revenue and $285.7 million of segment EBITDA, a 50% margin, while the group posted $236.2 million of adjusted EBITDA against a loss of $21 million a year earlier. GAAP net loss was $190.4 million, but that figure is distorted by a non-recurring $597.4 million revaluation gain recognized in Q2 2025 that is no longer in the base. Run-rate revenue ended June at $3 billion, up from $1.9 billion at the end of March.

The balance sheet tells a second story. Nebius carried $8.0 billion of cash and $17.6 billion of total debt at quarter end, leaving net debt of roughly half a billion. Free cash flow for the trailing twelve months was negative $5.9 billion: $5.3 billion of operating cash flow against $11.1 billion of capex, the second figure driven almost entirely by GPU purchases and data-center fit-outs. Full-year 2026 capex guidance is $20 to $25 billion.

That is the shared floor. Every claim that follows is a read on those same numbers.

Round 1: The bull case — the backlog is contracted revenue, not a hope

The strongest version of the bull argument does not depend on the AI trade being euphoric. It depends on the shape of the contracts.

Nebius's pipeline is not a bag of letters of intent. The two anchor deals are the $17.4 billion five-year Microsoft infrastructure agreement from September 2025 and a $12 billion five-year Meta supply deal signed in March 2026. Beyond those, Q2 closed four landmark deals averaging more than $1 billion each — Reflection, Cohere, an unnamed U.S. frontier lab, and a quantitative-trading firm — at roughly $20 to $25 million per megawatt, with shorter-term deals clearing at $40 to $50 million per megawatt. Management's first capacity auction for Blackwell chips cleared 15% above the previous high rate and 20% above current pipeline pricing. That is not a seller's market in name only; the price action inside the auction is the evidence.

The second bull fact is the prepayment structure. Management projects more than $9 billion of customer prepayments in 2026, with roughly 70% of deals closed in Q2 carrying prepayment terms that cover 50% to 60% of the associated capex. In plain terms, the customers are funding a meaningful share of the data centers they are buying capacity on. Nebius is also using the contracted cash flows to raise asset-backed debt — a $775 million facility at SOFR plus 250 basis points in July — on the strength of the backlog itself. The contracted power target for year-end 2026 was raised from roughly 3 gigawatts to 5 gigawatts, and management says it can sell its entire 2027 capacity at today's midterm pricing and is deliberately holding some back for the higher-margin short-term work.

The bull's cleanest line: this is not a business that needs to hope for a 2027 AI trade. A large fraction of the 2026–2028 revenue path is already in signed contracts with prepayments, and the EBITDA margin is expanding, not contracting, as utilization scales. The 50% segment margin is the number bulls point to as the reason this is not a commodity.

Bull wins this round. The contract stack, the auction clearings, and the prepayment structure are verifiable facts, and they do the heavy lifting the bear has to answer.

Round 2: The bear case — twenty times next year's sales on a business that has not yet printed a dollar of free cash flow

The bear's case does not require the AI trade to die. It only requires the multiple to be the thing it is.

At roughly 20x full-year 2026 revenue, the stock is paying a premium-multiple price for a company whose trailing free cash flow is negative $5.9 billion and whose full-year capex is guided to $20 to $25 billion. Set the two next to each other: at the midpoint, Nebius will generate approximately $1.3 billion of EBITDA (40% margin on $3.2 billion of revenue) while spending approximately $22 billion on capex. Every dollar of EBITDA is funding about one-seventeenth of a dollar of capex. The gap is closed by $9 billion of prepayments, by $2.8 billion of gross equity raised through an ATM program (12.7 million Class A shares at an average of $224), and by asset-backed debt. That is a working capital model as long as demand holds, but it is a model that keeps requiring new capital and keeps issuing new shares while the multiple is at a premium.

The second bear point is more structural: the demand side is real, but the supply side is not Nebius's problem to solve. The Q2 transcript is explicit about the gap between "connected power" and revenue. Nebius is targeting 800 megawatts to 1 gigawatt of connected power by year-end 2026, but much of that capacity is not expected to be active and revenue-generating until the first half of 2027. A site hearing for the Vineland, New Jersey data center was adjourned without a vote. Deployment lags of this kind are exactly where contracted revenue turns into a schedule, and where the $40 billion backlog starts to sound like a five-year number instead of a two-year one.

The third bear point is concentration. Microsoft, Meta, Reflection, Cohere, an unnamed lab, and a quant firm are doing the bulk of the revenue. That is a strong customer list — arguably a stronger one than almost any other neocloud can show. But it is a short one, and the whole business is running on a handful of frontier-lab and hyperscaler decisions. When one of those names restructures its 2027 capex plan, the revenue guide takes the hit the same quarter.

Bear wins this round. The 20x forward multiple on a business with negative trailing free cash flow and a hard capex program is the cleanest bear argument in the whole fight, and it does not depend on the AI sector cooling.

What the $240 price actually demands

This is where the two readings converge into a testable question.

The bull story works if 2027 revenue is a meaningful multiple of 2026, EBITDA margin holds near 40% or expands, the 50% segment margin is the floor rather than the ceiling, the backlog converts on schedule, and dilution stays bounded. The bear story works if any one of those breaks: a capacity lag that pushes contracted revenue into 2028, a prepayment shortfall that forces a larger equity raise, a frontier-lab capex reset that trims the 2027 guide, or an EBITDA margin that compresses as chip generations refresh and pricing power fades.

Notice that the stock has already priced in a version of the bull case. The 187% year-to-date move, the 34% pop on the Q2 print, and the 7.7% jump after Palantir named Nebius its preferred sovereign-AI infrastructure partner in early September are not the behavior of a market that is still underappreciating the business. They are the behavior of a market that is pricing in the 2027 guide and beginning to price in the 2028 one. The stock can keep rising if the 2027 guide comes in above the high end of the $3 to $3.4 billion range, but the expected return from here is much less obvious than it was in June.

The honest framing for a reader who is not already in the position: this is a "business goes up, stock call depends on price" situation. The business is doing real, verifiable work — contracted gigawatts, prepayments, 50% segment margin, Microsoft and Meta on the other side of signed paper. The stock is a different instrument. It is already paying for a large fraction of the bull case to work. The question an investor is answering at $240 is not "will Nebius grow?" — that is largely settled by the backlog. The question is "is the 2027 revenue and margin path already in the multiple?" On the numbers above, the answer is: mostly yes.

The ruling, and what would flip it

The business case goes to the bulls. The stock call, at $240, is a lean-bear or, at best, a hold. The $40 billion backlog and the prepayment structure are real de-riskers, but the multiple has already absorbed most of the de-risking, and the free-cash-flow gap — $22 billion of capex against $1.3 billion of EBITDA — means the equity story stays tied to dilution until utilization and pricing fully clear the cost of the buildout. A new-money position at a 20x-forward multiple on a business with negative trailing FCF carries more downside than upside if 2027 revenue lands at the low end of the guide or EBITDA margin drifts toward 35%.

The tripwire that flips this ruling: the Q3 2026 print, due in early November. Two numbers carry the decision.

First, the group EBITDA margin. If it holds at or above 40% on a revenue quarter at or above $650 million (a natural step-up from $582 million in Q2 if the 5-gigawatt contracted-power path is on track), the multiple has room. If it compresses toward 35% on a similar revenue print, the "expanding margin" pillar of the bull case has failed and the fair multiple comes down with it.

Second, the prepayment cadence. Management projected more than $9 billion of prepayments for 2026. Q3's reported figure, against the first-half run-rate, is the cleanest real-time read on whether the contracted backlog is converting to cash as scheduled. If the annualized run-rate of prepayments is clearly below $9 billion by year-end, the asset-backed financing model gets harder and the equity ATM becomes a bigger part of the story, which at a 20x forward multiple is the bear case winning.

The losing side's earliest confirmation signal is the opposite: a Q3 print with revenue above $700 million, an EBITDA margin of 42% or better, and prepayments on pace for the $9 billion figure would be the point at which the bull case stops being priced and starts being earned.

Until that print, the fair posture for a reader who does not own the stock is to let the multiple do the arguing. The business is the better story. The price is the more fragile one.