> ## Content Index > Fetch the complete content index at: https://www.sotkn.com/llms.txt > Use this file to discover other available public pages before exploring further. # What Does Nebius Do? The Neocloud Prepay Model - URL: https://www.sotkn.com/stack/neoclouds/what-does-nebius-do/ - Published: 2026-09-28T04:46:20.000Z - Updated: 2026-09-28T04:46:21.000Z - Description: Nebius sells specialist AI compute, funds a large share of the halls with customer prepayments, and gets paid as that capacity goes live — connecting contracted megawatts is the stuck step. - Author: Sean - Tags: #layer-neoclouds *Nebius sells specialist AI compute, funds a large share of the halls with customer prepayments, and gets paid as that capacity goes live — connecting contracted megawatts, not booking the next deal, is the stuck step.* `NEOCLOUDS` · `Data as at 12 Aug 2026` · `Next update: after Q3 2026 results` **Affects:** Neoclouds · Hyperscalers · US It is a [neocloud](https://www.sotkn.com/explained/neocloud-explained/): a specialist operator that supplies AI compute rather than a full cloud platform. For the side-by-side, see [neocloud vs hyperscaler](https://www.sotkn.com/explained/neocloud-vs-hyperscaler/). For the dated rate-card print, see the [Wire on the October PAYG hike](https://www.sotkn.com/wire/nebius-raises-gpu-prices-october-2026/). For the neighbour fleet, see [What Does CoreWeave Do?](https://www.sotkn.com/stack/neoclouds/what-does-coreweave-do/). **In short:** Nebius buys accelerators, stands them in halls, and sells cluster time on contracts that often arrive with cash up front. Q2 2026 AI-cloud revenue was $574.9m, with a 50% adjusted EBITDA margin and about $40bn of committed backlog. Deferred revenue was $6.0bn. The model works if contracted megawatts become connected halls — not if the contract list is merely long. *Education only. Figures are from company filings dated as shown.* ## Numbers that matter AI cloud revenue $574.9m +514% YoY · 99% of group $582.3m · quarter ended 30 Jun 2026 Growing Committed backlog \~$40bn Customer commitments · 12 Aug 2026 letter Booked Connected vs contracted power 0.8–1 → 5 GW YE 2026 connected guidance vs contracted target · Q2 call Stuck Deferred revenue (prepay) $6.0bn $979m current + $4,996m non-current · 30 Jun 2026 Funded Capex vs revenue (Q2) \~9.7× $5.66bn PPE spend vs $582.3m revenue Front-loaded Adj. EBITDA vs net loss (Q2) $236.2m / $(190.4)m Group 41% adj. EBITDA; AI cloud 49.7% · continuing ops loss Gap open Sources: Nebius Q2 2026 results and letter, 12 Aug 2026. --- ## What does Nebius actually do? Three steps, in order. Neoclouds · Nebius **Contract. Prepay. Connect.** 01 ### Contract Multi-year AI cloud deals and short premium windows. A signature is not a hall. 02 ### Prepay Customers fund 50–60% of related capex on most Q2 deals. Cash arrives before revenue. 03 ### Connect Megawatts, GPUs, onboarding. This is the stuck step — contracted power is 5× the live target. Source: Second Order reading of Nebius Q2 2026\. Educational only. Not a buy list. **1\. It contracts scarce capacity.** Nebius sells GPU clusters to labs, enterprises, and large offtakers. The Q2 letter (12 Aug 2026) said four landmark deals averaged more than $1bn of total contract value each, at $20–25m of annual revenue per megawatt — and that it could sell its entire 2027 capacity on those terms today, and is deliberately not doing so. From 1 October 2026 the public on-demand card lifts selected NVIDIA GPUs by 17–21% ([Nebius prices](https://nebius.com/prices?ref=sotkn.com), checked 28 Sep 2026). Commitment discounts of up to 35% still apply. On-demand is the margin of the book, not the book. **2\. It takes cash before the hall is live.** That is the funding difference versus [CoreWeave](https://www.sotkn.com/stack/neoclouds/what-does-coreweave-do/). About 70% of Q2 deals included customer prepayments covering 50–60% of the related capex. The company expects more than $9bn of customer prepayments in 2026\. On the balance sheet at 30 June 2026, deferred revenue was $979.4m current and $4,995.8m non-current. **3\. It connects halls and bills when capacity is accepted.** Chief product and infrastructure officer Andrey Korolenko said on the Q2 call that the year-end 2026 connected-power target remains 800 MW to 1 GW, and that contracted power is now guided to 5 GW — almost all of it coming online over the next three to three-and-a-half years. Connected power is not yet revenue. Commissioning, networking, cluster build, platform, and onboarding still sit in between. | Step | What it means | Where the risk sits | | -------- | --------------------------------------------------------- | ---------------------------------------------------------------- | | Contract | Multi-year and short AI-cloud deals; \~$40bn commitments | Concentration, conversion timing, mix of dedicated vs merchant | | Prepay | Customer cash covering a slice of capex; deferred revenue | Prepays that slow, or that have to be refunded if delivery slips | | Connect | MW, GPUs, customer onboarding | Power, sites, chips, and the months between connected and billed | --- ## How does Nebius make money? It gets paid for compute that is live and recognised as revenue. Three things decide how fast the AI-cloud line can grow: - **Delivery** — a booked megawatt earns nothing until the cluster is on - **Utilisation and price** — Q2 deals averaged above $20m of annual revenue per MW, with a $40–50m/MW opportunity on short-term capacity - **Mix** — AI cloud is already 99% of group sales; Avride and TripleTen sit beside it, they do not set this layer On 12 August 2026 Nebius printed $582.3m of group revenue for the quarter ended 30 June 2026, up 454%. Nebius AI cloud was $574.9m, up 514%. Annualised run-rate revenue — last month of the quarter times twelve — reached $3.0bn. Group adjusted EBITDA was $236.2m (41%). The AI cloud generated $285.7m of adjusted EBITDA, a 49.7% margin. **Our arithmetic.** $574.9m is 98.7% of group sales (574.9 ÷ 582.3). AI-cloud adjusted EBITDA of $285.7m on $574.9m is 49.7%. Group adjusted EBITDA is $49.5m lower than the AI-cloud line ($285.7 − $236.2) — other businesses and group costs on the same scorecard. Loss from operations was $(175.9)m. Net loss from continuing operations was $(190.4)m. Underneath the 41% group margin, two clocks still close the gap to the bottom line. | | Q2 2026 | As a share of group adj. EBITDA | | ----------------------------- | --------- | ------------------------------- | | Group adjusted EBITDA | $236.2m | — | | Interest expense | $119.1m | **50%** | | Depreciation and amortisation | $259.7m | — | | Reported continuing-ops loss | $(190.4)m | — | **Our check:** 119.1 ÷ 236.2 = 50.4%, which we round to 50%. Interest takes half of group adjusted EBITDA. The rest of the gap is depreciation — writing down GPUs and halls — plus the items adjusted EBITDA is defined to skip. The letter says the company uses a **five-year** useful life for server and network equipment, up from four years before 2026\. That is a company accounting choice. We do not treat it as settled physics. Cash tells the prepay story more clearly than the income statement. Operating cash flow from continuing operations was $2,246.1m. Inside that, deferred revenue rose $1,197.0m — **our check:** 53% of quarterly operating cash flow (1,197.0 ÷ 2,246.1). Purchases of property and equipment were $5,657.4m, or **9.7×** quarterly revenue (5,657.4 ÷ 582.3). The offset is not CoreWeave-scale recourse debt. Cash was $8,042.1m; debt $8,545.7m ($46.7m current + $8,499.0m non-current) — **$0.5bn of net debt**. In July, after the quarter closed, Nebius raised a $775m secured facility at SOFR + 2.50%, backed by deployed GPUs and contracted cash flows from an investment-grade customer. Who gets paid: Nebius, for live AI-cloud hours. Money sticks at delivery. Prepay is a claim on future delivery until the hall is accepted. --- ## What limits it? Connection first. Power and sites second. Concentration third. **1\. Contracted power is not connected power.** Year-end 2026 connected guidance is 800 MW to 1 GW; contracted power is 5 GW. **Our arithmetic:** at the top of the connected range, contracted is 5× live. Korolenko said almost all of that contracted figure comes online over three to three-and-a-half years, and that several months still sit between a connected hall and recognised revenue. A contract signed against 2027 capacity does not invoice at the AI-cloud rate today. See [speed-to-power](https://www.sotkn.com/explained/speed-to-power-explained-for-investors/) on the [US door](https://www.sotkn.com/markets/us/). **2\. Remaining commitments are not this year's sales.** About $40bn of customer commitments can be borrowed against. Most Q2 landmark deals, the letter said, were signed against capacity arriving in late 2026 and will contribute primarily to 2027 revenue. Full-year 2026 group revenue is still guided at $3.0–3.4bn. **Our read:** $40bn is a conversion clock, not 2026 AI-cloud. **3\. Two large offtakers sit on the book.** The Q3 2025 letter put the Microsoft Vineland agreement at $17.4–19.4bn; the Q2 2026 letter said all those tranches had been delivered. On 16 March 2026 Nebius signed a second Meta agreement worth up to about $27bn: $12bn of dedicated NVIDIA Vera Rubin capacity from early 2027, plus a $15bn unsold-capacity order. Buildout for that second Meta deal is “on track to come online in early 2027.” Those tickets fund the ramp. They also concentrate who the fleet serves. **4\. Capex is the cash cost of the same bind.** Full-year capex is guided at $20–25bn. Q2 already spent $5.66bn. Property and equipment, net, rose from $5,553.3m at year-end 2025 to $13,045.2m at 30 June 2026\. The company is adding an asset-light path — partners deploy the Nebius stack in their own halls — because capital and connected megawatts both bind. The honest uncertainty is conversion. A company can be raising on-demand prices and still miss a revenue guide if halls slip. It can also print $40bn of commitments that do not show up in this year's AI-cloud. Those are different clocks. We are not going to pretend a 21% PAYG hike settles both. --- ## Who else does this? Roles, not a league table. | Company | What it does | What makes it different | | ------------------- | ----------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Nebius** | Specialist AI cloud; prepay-heavy funding | \~$40bn commitments; $6.0bn deferred revenue; 5 GW contracted target vs 0.8–1 GW connected; Meta/Microsoft offtake | | **CoreWeave** | Largest specialist AI compute supplier | Scale and the largest contracted backlog; funds more with debt. See [What Does CoreWeave Do?](https://www.sotkn.com/stack/neoclouds/what-does-coreweave-do/) | | **Applied Digital** | Builds and operates sites | Closer to the landlord end than the compute end | | **Hyperscalers** | Broad cloud; also buy specialist overflow | Own balance sheets, own chips. See [Oracle](https://www.sotkn.com/stack/hyperscalers/what-does-oracle-do/) and [hyperscaler, explained](https://www.sotkn.com/explained/hyperscaler-explained/) | The question that separates them is who has cash in before the hall is live, who can connect the next megawatt, and whether the rate card is still rising after capacity arrives. --- ## What would change this view? The read on the model weakens if: - Connected power at year-end 2026 prints well below the 800 MW–1 GW range while contracted power keeps rising — the conversion gap would widen - On-demand and short-contract prices fall for a full quarter while installed capacity keeps rising — the opposite of the [October hike](https://www.sotkn.com/wire/nebius-raises-gpu-prices-october-2026/) - Prepayments drop well below the \~70% of Q2 deals, a sign buyers no longer fear running short - A named offtaker — Meta or Microsoft — defers or cuts in a dated print - [Hyperscalers](https://www.sotkn.com/stack/hyperscalers/) absorb overflow so specialist utilisation stays weak even when headline GPU demand looks firm - First Vera Rubin deployments attach late because packaging or [HBM](https://www.sotkn.com/explained/hbm-explained/) is short — the stuck step would then sit in [semiconductors](https://www.sotkn.com/stack/semiconductors/) It stays intact while connected power tracks the 800 MW–1 GW year-end band, deferred revenue and prepay language stay large, AI-cloud adjusted EBITDA holds near 50%, and on-demand prices do not reverse the October column. --- ## Practical takeaways - Nebius's model on this layer is **Contract → Prepay → Connect**: deals swell commitments; customers fund a slice of the build; megawatts have to go live before AI-cloud revenue prints - Q2 2026: $574.9m AI-cloud revenue (+514%), \~$40bn commitments, $6.0bn deferred revenue, group adj. EBITDA $236.2m versus a $(190.4)m continuing-ops loss - Interest takes \~50% of group adjusted EBITDA; servers are now depreciated over five years - The stuck step is **connecting contracted power** — 5 GW booked versus 0.8–1 GW connected guidance — not “did Nebius sign another deal” - Next hard read: Q3 2026 results — connected megawatts, prepay, and whether the 1 October rate card held --- ## FAQ **What does Nebius do?** Nebius is a specialist AI cloud. It buys accelerators, stands them in data halls, and sells cluster time on contracts — including multi-year deals with large offtakers. It does not sell a full hyperscaler platform. **What does Nebius do in simple terms?** It owns the expensive machines that train and run AI models, takes cash from customers before many halls are live, and charges for the compute once the cluster is actually running. **Is Nebius profitable?** Not on a reported continuing-operations basis. It posted a net loss of $190.4m in the quarter ended 30 June 2026 while reporting $236.2m of group adjusted EBITDA and $285.7m of AI-cloud adjusted EBITDA (49.7% margin) on $574.9m of AI-cloud revenue (12 Aug 2026). **How is Nebius different from CoreWeave?** Both sell AI compute. CoreWeave funds more of the fleet with debt. Nebius funds a larger share with customer prepayments — $6.0bn of deferred revenue at 30 June 2026 — and still loses money on a GAAP continuing-ops basis while it connects halls. See [What Does CoreWeave Do?](https://www.sotkn.com/stack/neoclouds/what-does-coreweave-do/). **What is deferred revenue here?** Cash customers have already paid for capacity that has not yet been recognised as revenue. At 30 June 2026 that stock was $6.0bn. It is a prepay clock, not this quarter's AI-cloud sales. **Should I buy Nebius shares?** We do not make buy or sell calls, and nothing here is a recommendation. What we can tell you is what the company does, how it gets paid, and which figures would change the picture. The decision is yours. --- ## Keep reading on the map - **The layer:** [Neoclouds](https://www.sotkn.com/stack/neoclouds/) — who gets paid when short contracts bind - **The term:** [Neocloud, explained](https://www.sotkn.com/explained/neocloud-explained/) - **The split:** [Neocloud vs hyperscaler](https://www.sotkn.com/explained/neocloud-vs-hyperscaler/) - **The Wire:** [Nebius raises GPU prices from 1 October](https://www.sotkn.com/wire/nebius-raises-gpu-prices-october-2026/) - **Where it happens:** the [US](https://www.sotkn.com/markets/us/) — [speed-to-power](https://www.sotkn.com/markets/us/us-speed-to-power-september-2026/) is the long clock under the halls ## Members' Analysis: if the halls connect, does the bind move to power-at-the-site? The free part: $6.0bn of deferred revenue against a 5 GW contracted-power target is what a prepay-funded delivery bind looks like in a dated print. The members cut is live: [Where does the stuck step move?](https://www.sotkn.com/analysis/where-does-the-stuck-step-move/). This page stays the free Stack read of the Nebius model. ## Sean Writes Second Order by TKN — a plain-English map of AI infrastructure and semiconductors for non-specialist investors. Focus: who gets paid, where money sticks, and which step is stuck. Not investment advice. [Author page →](https://www.sotkn.com/author/sean/) · [Editorial standards →](https://www.sotkn.com/editorial-standards/) ## How we check this Figures come from company filings and company releases, each dated below. Capacity from the earnings call is the company's spoken print, not our estimate. Rate-card figures are from the public prices page on the date shown. Where we work something out ourselves, the arithmetic is shown in full. **Last reviewed:** 28 Sep 2026 · **Next review:** after Q3 2026 results. Spot an error? Tell us and we will correct it. [Editorial standards →](https://www.sotkn.com/editorial-standards/) ## Sources 1. Nebius, [Q2 2026 financial results](https://nebius.com/newsroom/nebius-reports-second-quarter-2026-financial-results?ref=sotkn.com) ([PDF](https://assets.nebius.com/assets/dfe7a7f3-771e-4653-94e8-8f86bf126b1d/PR.pdf?ref=sotkn.com)), 12 Aug 2026 — revenue, adj. EBITDA, continuing-ops loss, interest, D&A, capex, cash, debt, deferred revenue, operating cash flow 2. Nebius, [Letter to shareholders Q2 2026](https://assets.nebius.com/assets/a6ecfd85-a6cb-4967-8ef7-9a25bd261f9c/SHLQ226.pdf?ref=sotkn.com), 12 Aug 2026 — AI-cloud $574.9m and $285.7m adj. EBITDA (49.7%), \~$40bn commitments, $20–25m/MW, prepays, five-year useful life, Microsoft delivered, second Meta buildout, 5 GW contracted target 3. Nebius Q2 2026 earnings call, 12 Aug 2026, via [investor hub](https://nebius.com/investor-hub?ref=sotkn.com) — 800 MW–1 GW connected-power guidance 4. Nebius, [Meta AI infrastructure agreement](https://nebius.com/newsroom/nebius-signs-new-ai-infrastructure-agreement-with-meta?ref=sotkn.com), 16 Mar 2026 — up to \~$27bn 5. Nebius, [Microsoft AI infrastructure agreement](https://nebius.com/newsroom/nebius-announces-multi-billion-dollar-agreement-with-microsoft-for-ai-infrastructure?ref=sotkn.com), 8 Sep 2025; $17.4–19.4bn from the [Q3 2025 letter](https://assets.nebius.com/assets/7f8a9169-0d2b-469d-87e7-44342ea7fcd2/SHLQ3%20%284%29.pdf?ref=sotkn.com) 6. Nebius, [AI Cloud pricing](https://nebius.com/prices?ref=sotkn.com), checked 28 Sep 2026 — on-demand columns from 1 Oct 2026 7. Reuters, [Nebius hikes AI cloud prices again](https://www.reuters.com/technology/nebius-hikes-ai-cloud-prices-again-demand-computing-power-soars-2026-09-17/?ref=sotkn.com), 17 Sep 2026 — reported as the second PAYG hike in about three months 8. Second Order, [Nebius GPU prices from 1 October](https://www.sotkn.com/wire/nebius-raises-gpu-prices-october-2026/) — Wire dated 21 Sep 2026 All external links `rel="nofollow noopener"`.