Nebius Q3 2026: The Market Overreacted

On Thursday, $NEBIUS(NBIS)$ released its Q3 2026 results, and the stock closed at $128.64.

In short, Nebius reported:

• Revenue, profit and margins beat

• Higher prices

• Prepayments covering more than 100% of the capex

• 2026 revenue at the low end of guidance

• 2027 revenue Guidance of $12-13B

The market hated this report, but in my opinion, it's an overreaction!

Nebius lost $7.6B of market value in a day, because 2026 revenue will now land 4-10% below what analysts expected. The power at a leased site in New Jersey arrived late, so part of this year's revenue moves into 2027.

Also, the stock had jumped 29% in the 4 trading days before the report, mostly after CoreWeave's strong results, so Thursday's fall only took it back to where it was a week ago.

Meanwhile, customers signed contracts at higher prices than ever and agreed to prepay the capex behind them. The CEO, Arkady Volozh, summed it up at the start of the call:

“The demand for AI is real and substantial. Everything else, like building the capacity and financing it is just a task. Tasks can be hard, but they all can be solved.” Arkady Volozh, CEO, Q3 2026 Earnings Call

Today I want to walk you through the results, the Microsoft delay, why I expect more delays across the whole industry, and how Nebius is paying for its growth. I am also working on a rebuilt valuation model, which you will find in my newest valuation report for premium subscribers sometime next week.

Let's dig in.

1. Q3 2026 Financial Results

• Revenue: $981M, +571% Y/Y, vs $907M estimate. BEAT

• ARR: $6.4B, +113% Q/Q

• ADJ EBITDA: $466M, a 47.5% margin, vs $349M estimate. BEAT

• EPS: -$0.13 vs -$0.63 estimate. BEAT

• GAAP Net Income: -$38M

Nebius beat on every line, although part of the beat came from a one-off payment.

1.1. Revenue

Nebius reported revenue of $981M, +571% Y/Y and +68% Q/Q!

However, $325M of it came from the new asset-light model, and management calls this revenue non-recurring. I explained the model in my Q2 review. A partner pays for the buildings and the GPUs, while Nebius brings its cloud software and the customers. In Q3, the first partner paid Nebius to set up its platform on capacity the partner funds.

Without this payment, revenue was $656M, +13% Q/Q.

As you can see in the chart above, this is still growth against the reported Q2. However, the 6-K shows that Nebius booked $84M of similar partnership revenue earlier this year. If that sat in Q2, core revenue actually grew 32% Q/Q, even though capacity at leased sites arrived late, mostly in New Jersey.

As a result, the CFO said on the call that 2026 revenue will be $3.0-3.2B, against the $3.0-3.4B range before.

“Demand has never been stronger and the changes in our 2026 guidance are just about timing, not demand.” Dado Alonso, CFO, Q3 2026 Earnings Call

Nebius made $1.96B in the first 9 months, so the new guidance points to Q4 revenue of $1.04-1.24B. Analysts expected $1.46B, so the whole miss for the year lands in Q4, and this gap is the main reason the stock fell.

1.2. ARR

ARR reached $6.4B, +113% Q/Q.

As a reminder, ARR is the revenue of the last month of the quarter multiplied by 12. So Nebius made $533M in September, against $250M in June.

However, most of the $325M one-off landed in September. Nebius made $448M in July and August together, or $224M a month, so without the one-off, the run rate was between $2.7B and $3B. The CFO said on the call that asset-light deals "can add choppiness to ARR".

Nebius had guided for $7-9B of ARR by December. It now expects ARR of more than $7B by February, as New Jersey reaches full capacity. That means recurring monthly revenue has to more than double in 5 months.

From 2027, Nebius will stop guiding for ARR.

“It is based on 1 month of the quarter, and we don't build our business around any 1 month.” Dado Alonso, CFO, Q3 2026 Earnings Call

Instead, it will guide for revenue, pre-tax earnings and capacity, and I like this change. ARR made sense when Nebius was a start-up, but now 1 payment from 1 partner can move it by billions, since every dollar that lands in September counts 12 times. I prefer pre-tax profit, as it includes the depreciation and interest that ARR ignores.

1.3. ADJ EBITDA

Group ADJ EBITDA was $466M, a 47.5% margin!

In the chart above, you see how quickly this turned around, as ADJ EBITDA was -$5M a year ago. The AI cloud business earned $518M, a 53.2% margin, up from 50% in Q2. The gap between the 2 margins comes from Avride and TripleTen, the smaller Nebius businesses.

The one-off helped here too, since setting up software on a partner's hardware probably costs Nebius very little. If we assume the one-off cost Nebius almost nothing, the rest of the business earned a 21% ADJ EBITDA margin, well below Q2. The CFO explained that new sites cost money before they earn anything:

“New sites carry rent, power and people before they build.” Dado Alonso, CFO, Q3 2026 Earnings Call

The company kept its 40% margin guidance for the year, which points to a Q4 margin in the mid-30s. So Q4 should be a big step up from the core 21% in Q3, as the sites that cost money in Q3 start to earn revenue.

1.4. Net Income

Nebius made a GAAP operating profit of $50M, its first since it became an AI cloud company.

Depreciation is by far the largest cost of a GPU cloud, at $355M in Q3. But the profit came from the one-off payment, and without it Nebius would still have had an operating loss. Excluding the one-off, depreciation was 54% of revenue, up from 45% in Q2, as new GPUs start to depreciate before they earn their full revenue.

The pre-tax loss was $63M, which included a $50M one-time charge for swapping convertible notes (bonds that can turn into shares) early in August. Without the charge, the pre-tax loss was only $13M, while after a tax benefit, the net loss was $38M.

2. Microsoft

Microsoft news sent the stock down 15% on 5 November, a week before the results.

Nebius serves its $17.4B, 5-year Microsoft contract from a leased data center in Vineland, New Jersey. In the 2 weeks before the report, the news about this site got worse:

• 28 October: Microsoft's capex came in at $45.2B, below the $50.7B analysts expected, because some partners were late.

• 4 November: Business Insider published a satellite analysis by Aterio, which estimated that 140 MW in Vineland could slip into 2027.

• 5 November: Bloomberg reported that Microsoft is watching projects from Nebius, Iren and Nscale more closely, even sending its own people to New Jersey.

“Capital expenditures were $45.2 billion as some third-party capacity deliveries we expected in Q1 were delayed.” Amy Hood, Microsoft CFO, 28 October 2026

In its Q3 report, Nebius confirmed the delay.

According to the slide above, more than half of the Microsoft capacity will be delivered by the end of 2026. The rest moves to Q2 2027.

The grid in New Jersey can't deliver enough electricity yet, so the landlord, DataOne, ran the site on temporary gas generators. In September, New Jersey fined DataOne $1.07M for running 62 of them without air permits. DataOne disputes the finding, but said it would apply for the permits. Then, in late Q3, it started to retire some of the generators earlier than planned.

“In late Q3, DataOne decided to wind down the temporary generation faster than it was expected. We are expediting the Bloom deployment as much as possible, but we still will have a gap due to this situation.” Andrey Korolenko, Chief Product and Infrastructure Officer, Q3 2026 Earnings Call

The permanent fix is Bloom Energy fuel cells, which already produce power on the site.

“The power is on the site. Of course, it took a couple of months of delay, but we are back on track. Electricity is there. The project is still going.” Arkady Volozh, CEO, Q3 2026 Earnings Call

Most importantly, Microsoft didn't cancel anything.

Nebius kept delivering new tranches to Microsoft through Q3, with more planned for November and December. Even Bloomberg wrote that Microsoft hasn't ruled out new neocloud deals.

I estimate that Microsoft pays Nebius $11M per MW a year. If Aterio's 140 MW is right, $1.5B of yearly revenue arrives 4 to 6 months late. That fits an ARR target that moved by 2 months and a revenue guidance at the low end.

Furthermore, Microsoft pays far less than the newer customers.

Q3 deals pay more than $25M per MW, so the same MW now earns more than twice as much. Nebius signed the Microsoft deal in September 2025, when it was a small company that needed a famous customer. In March, the CEO told investors this:

“We treat this business as temporary” Arkady Volozh, CEO, Morgan Stanley conference, March 2026

Banks lend against contracts with investment-grade customers, the safest borrowers like Microsoft, and the $775M facility in July was secured on exactly this kind of contract. So for me, the Q2 2027 date matters more than the lost revenue, as Nebius needs the contract running to borrow against it.

3. Delays

New Jersey won't be the last delay, at Nebius or anywhere else in this industry.

“Building at this scale is hard and delays are part of our industry right now. We are not immune. For example, some of our colocation sites, including New Jersey, are behind schedule. Customers have been kept informed and none of that capacity is canceled.” Arkady Volozh, CEO, Q3 2026 Earnings Call

In the past 3 weeks, the same delays showed up across the industry:

• Microsoft's CFO raised late deliveries from partners on an earnings call for the third time.

• Iren moved its $4B ARR target from December to the end of January, possibly February.

• Microsoft cancelled a $3B deal with Nscale in the UK, because the grid connection there slipped into the 2030s.

Up to half of the US data centers planned for 2026 are late or at risk!

The research firm Sightline Climate tracks 140 of these projects and expects 30-50% of them to be delayed or cancelled. In March, only 5 GW of the 16 GW planned for 2026 was under construction.

Simply put, GPUs can be ordered and installed in months, while a grid connection takes years. Transformers, the big boxes that bring the grid's voltage down to what a data center can use, now have waiting times of 2 to 4 years. So operators plug in whatever power they can find, like the gas generators and fuel cells in Vineland, and then the permits for those become the next hurdle.

Moreover, politics is making it harder.

In the 3 November elections, voters rejected data centers in 40 of 41 local ballot measures. In Texas, the governor-elect campaigned on "an immediate moratorium on data centers to last until the legislature can meet to put in place real protections".

“As of today, we don't see or anticipate any meaningful impact on time lines coming from the shift in the political landscape.” Tom Blackwell, Chief Communications Officer, Q3 2026 Earnings Call

So I expect more headlines like the one from New Jersey, at Nebius and at its competitors. However, Microsoft walked away from Nscale because its grid connection slipped into the 2030s, while in Vineland the fuel cells already run, so Microsoft keeps the same price and gets the rest of its capacity in Q2 2027.

In fact, these delays push GPU prices up!

If half of the planned capacity comes late, every GPU that is already running becomes more valuable. Customers who can't wait pay more and prepay to reserve their spot, which is what Nebius saw in Q3. CoreWeave's CEO said the same on its Q3 call, that slower construction could mean GPU pricing "may stay higher for longer".

4. Customer Pricing

Nebius raised its prices again!

We see that Q3 contracts pay more than $25M per MW a year, against $20-25M

in Q2 and $12M at the start of 2026. Per GPU-hour, prices rose more than 20% in 3 months.

Instead of only a fixed price list, customers can now bid for capacity or rent it at a live price:

• Reserve auctions, where customers bid for capacity, cleared more than 15% above the minimum price Nebius set.

• Spot prices, the live price for capacity that Nebius can take back at short notice, at times went above the normal on-demand rate.

• Short-term clusters of 3 to 6 months rent for more than twice the price of a reserved contract.

This goes against one of the biggest bear arguments.

In September, Rothschild Redburn started covering Nebius with a Sell rating and an $84 target, calling neocloud earnings "cyclically inflated". Some coverage read that as a bet on falling GPU prices. Since then, every price I could find has gone up, and CoreWeave sees the same.

“Pricing for long-dated customer contracts is going up to $20 million-$25 million per megawatt” Nitin Agrawal, CoreWeave CFO, Q3 2026 Earnings Call

Customers committed $12B to Nebius since the start of Q3!

Yet the backlog grew only 3% to $38.7B. An analyst from Citi asked about this on the call, and the Chief Revenue Officer explained that Nebius raised its prices in September, so several deals were signed after the quarter ended.

“At the start of September, we increased prices. We added higher prepaid terms and contract duration to a new set of capacity that we were releasing. As a result, since the start of Q3, customers committed more than $12 billion of contract value, including multibillion-dollar 5-year agreements.” Marc Boroditsky, Chief Revenue Officer, Q3 2026 Earnings Call

The backlog rose $1.2B in Q3 while Nebius booked $1B of revenue, so only $2.2B of new contracts made it in. That means almost $10B of the $12B was signed in the 6 weeks after the quarter ended, so I expect the backlog to jump in Q4.

The customer list is getting stronger too. Databricks, Datadog and Snowflake signed multi-year deals for the first time. Palantir also picked Nebius as its preferred AI cloud for sovereign AI. Cognition, the company behind the AI coder Devin, will run its first large Vera Rubin clusters on Nebius.

Nebius still depends on a few big customers, as 3 made up 70% of Q3 revenue. However, the largest, at 33%, looks like the asset-light partner, so every big new name like these lowers that risk.

4.1. Prepayments

On new Q3 deals, customers prepay more than the whole capex!

Prepayments on Q3 deals covered more than 100% of the capex behind them, including the building and the network. In Q2, they covered 50-60%. Furthermore, more than 90% of the contract value signed in Q3 came with a prepayment.

“Our customers are paying more, they pay earlier and they're committing for longer, and we are getting our capital back faster.” Arkady Volozh, CEO, Q3 2026 Earnings Call

It's like a tenant paying the first years of rent before the landlord has built the house, so the landlord never needs a mortgage.

Nebius pays no interest on this money and issues no shares for it, so it's the cheapest funding it can get.

The payback period also fell to 1 year and 7 months, from 1 year and 10 months in Q2. According to the letter, this covers the capex and running costs of comparable 1 to 3-year deals, without any help from the prepayment. So Nebius earns its money back in 19 months on a contract that runs up to 3 years. Absolutely incredible.

Looking at the chart above, deferred revenue, reached $9.4B, from $6B in June. Nebius expects $10B of prepayments in 2026 and $17-20B in 2027.

Still, most of the $3.4B added in Q3 hasn't arrived as cash yet. Receivables rose from $0.3B to $4B, and 75% of that sits with 1 large customer. That's why OCF was -$273M in Q3, after 2 strong quarters.

5. Capacity

Nebius kept its target of 5 GW of contracted power by the end of 2026, and it already has more than 4 GW.

Since the Q2 report, it added:

• Missouri: 300 MW more, taking the site to 1.1 GW.

• Alberta, Canada: a 100 MW site, with delivery in 2027.

• South Carolina: a 50 MW colocation (a leased site), its 8th US site.

• India: 2 colocations, its first sites in Asia-Pacific.

Most importantly, from 2027 most of the new capacity comes from sites Nebius owns.

The map above shows that a little more than 10% of today's capacity is in owned sites. Of the next GW Nebius adds, more than half will be owned, and of the GW after that, more than 75%. Missouri, Alabama and 2 sites in Finland start delivering in 2027, adding up to more than 3 GW over the next 3 to 3.5 years. Nebius has already secured the power and the components for 2027.

In my Q2 review, I wrote that owned sites are 75% of contracted capacity. Contracted power includes sites that are still being built, while most of what earns revenue today is leased, so a power issue at 1 landlord's site was enough to cut this year's revenue.

In my opinion, this is the right move after New Jersey. At a leased site, Nebius can't control the landlord's permits or power, while on its own sites, it handles the permits and the construction itself. The company says those sites are on time or even ahead of schedule.

“Where we rely on third parties, there have been delays.” Nebius Q3 2026 Shareholder Letter

Owned sites are also more profitable, as Nebius doesn't pay a landlord's margin.

5.1. Asset-light

The asset-light model brought in $325M of revenue in Q3!

The first phase of the first asset-light partnership closed in Q3, and that is the one-off in this quarter's revenue. Nebius gets paid twice, once when it sets up its platform on the partner's capacity and again when that capacity comes online, which should start in 2027.

The cash should follow in the coming quarters, as the customer that looks like the partner still owed Nebius about $400M at the end of September.

It's 1 partner so far, but I like that the partner pays for the buildings and the GPUs, as funding is what investors worry about most right now.

6. 2027 Guidance

Nebius expects revenue of $12-13B in 2027, 4x this year!

Analysts expected $12.3B, so the first guidance lands right on consensus. But management says it built a lot of room for delays into this number.

“On the capacity part, we took a very, very conservative approach about the timing, which give us a lot of comfort. As a matter of fact, a significant part of this 2027 capacity will be owned by us and not by third parties, which give us control over the delivery.” Dado Alonso, CFO, Q3 2026 Earnings Call

Moreover, the guidance includes no revenue from new asset-light deals, so any new partner comes on top.

“We didn't add any potential revenue from asset-light model and other partnership for 2027.” Ophir Nave, COO, Q3 2026 Earnings Call

Nebius also expects to reach pre-tax breakeven in Q4 2027!

That's after interest, with GPUs depreciated over 5 years. For a company planning

$40-45B

of capex next year, this is a big promise. I think it's doable, as Q3 deals pay more than twice the Microsoft price while new customers prepay the capex.

“On the demand side, we have no doubt that this capacity will be sold. Clients are looking for capacity already for 2028 and prices for capacity in '27 are going up.” Dado Alonso, CFO, Q3 2026 Earnings Call

7. Cash and Funding

Nebius ended Q3 with $7.4B of cash and $14.2B of debt.

Net debt is $6.8B, and on top of that Nebius holds $1.2B of restricted cash, which is set aside for specific uses. In October, Citi gave Nebius a $3.5B loan that runs for 1 year, which Nebius can refinance with asset-backed loans against its Microsoft and Meta contracts.

The whole sector fell because investors worry about who will pay for all these data centers. So let's see where Nebius got its money this year.

Nebius raised $28B in 2026, half of it was debt, mostly convertible notes, while less than $6B came from selling shares. Customers committed $7.8B of prepayments, and about $4.5B of that had arrived as cash by September.

7.1. Capex

Capex was $7.3B in Q3, against $5.7B in Q2.

Nebius raised its 2026 capex guidance to $27B from $20-25B. After $15.4B in the first 9 months, that leaves $11.6B to spend in Q4. Nebius has $7.4B of cash and $3.5B from Citi, while customers still owe it $4B, so it should be able to pay for most of Q4 without selling more shares.

For 2027, Nebius expects capex of $40-45B!

Prepayments and loans against its contracts with customers should cover 75-85% of it. If Nebius spends $42.5B and customers and lenders cover 80% of it, it still has to find $8.5B from other sources, likely debt or equity.

For 2027 as a whole, prepayments cover 44% of capex, less than on the new Q3 deals, partly because the Microsoft and Meta capacity and the older contracts were signed on weaker terms. Loans against those contracts cover most of the rest.

Also, Nebius can change this mix. When borrowing gets expensive, it sells more capacity on long contracts with big prepayments, while in cheaper times it can borrow more and keep capacity for short-term deals that pay more.

“At the end of the day, when your prepayment on your contracts cover more than 100% of the CapEx, you're in a great position.” Ophir Nave, COO, Q3 2026 Earnings Call

7.2. Dilution

Nebius had 292.9M shares in September, +23% Y/Y.

In Q3, Nebius paid some holders of its convertible notes to swap them for shares early, which added 15.8M new shares. Nebius also sold 4.6M shares through its ATM program, at an average of $229.10, raising $1B. Today, at $129, the same $1B would cost 78% more shares.

8. Conclusion

In conclusion, I think the market overreacted to a few months of delay and ignored the best contracts Nebius has ever signed.

In August, everything looked easy. Prices had doubled and customers were prepaying, so the stock hit $278 after the Q2 report. Then Microsoft said some partners were late, while the midterms turned data centers into a political target.

On Thursday, Nebius confirmed the delay part of the Microsoft capacity moved to Q2 2027, and 2026 revenue will end slightly below its previous guidance of $3.4B.

However, since the start of Q3, customers committed $12B at more than $25M per MW and agreed to prepay. With new customers prepaying the build, I don't think Nebius has the funding problem the market feared in October.

There will be more delays, at Nebius and at everyone else who is waiting for the grid to connect their data centers. Q4 revenue will miss estimates by up to 29%, but the delayed Microsoft revenue only moves into 2027, while the new contracts lock in the higher prices for up to 5 years. Meanwhile, Nebius is moving most of its build to its own sites, where it controls the power and the permits.

The stock is now at $129, half of where it was after the Q2 report!

Whether it's cheap depends on what Nebius can earn by 2030. In my newest valuation report for premium subscribers, I show what the new prices and

$40-45B

of capex in 2027 do to my 2030 earnings, and how much dilution the $8.5B funding gap could bring.