Nebius Group Reveals Premium Pricing Power as Short-Term Capacity Fetches Double the Rate of Multiyear Deals
Q2 2026 earnings call, August 12, 2026 — revenue up 454% to $582 million as management discloses new deal economics and raises capacity targets
Nebius Group used its second-quarter call to pull back the curtain on the pricing dynamics driving its business, and the numbers point to a market where compute scarcity is translating into outsized economics for suppliers willing to hold back capacity rather than lock it up early. CEO Arkady Volozh disclosed that the company closed four landmark deals this quarter, each averaging more than $1 billion, with Reflection, Cohere, an unnamed scaled U.S. neolab, and a large U.S.-based quant trading firm. Those midterm, one-to-three-year contracts yield $20 million to $25 million per megawatt with upfront payments covering 50% to 60% of associated capital expenditure.
What stood out more was the second tier of business Nebius is now running deliberately. For customers with immediate, time-boxed needs — typically six months or less — the company is negotiating short-duration deals in the $40 million to $50 million per megawatt range, "and sometimes above." Volozh was explicit that this is a choice, not a constraint: "We could sell today our entire 2027 capacity on these terms if we wanted to. But we are not doing this. We see that we can achieve higher value by retaining some capacity to serve shorter-term and immediate client needs." That is a notable strategic signal — Nebius is intentionally under-committing forward capacity to capture premium pricing as the market tightens, a bet that only works if demand continues to outstrip new supply.
Auction Pricing Confirms the Squeeze
The company's first capacity auction, launched this quarter, provided real-time validation of that thesis. Marc Boroditsky, who runs go-to-market, said the auction cleared at the highest price the company has ever seen for Blackwell-generation chips, 15% above the previous ceiling and 20% above the existing pipeline for the same hardware. He described the process as being about pure price discovery in a market where "the standard reference points — competitive pricing, analyst opinions and even prediction markets — are all over the place. In a market where we have several buyers for every GPU, we let the market tell us directly." The winning bidder has already indicated it plans to participate in future rounds, and Nebius is treating the auction and the short-term training deals as parallel tools for gauging fair value as it heads into 2027 contract negotiations, rather than as one-off promotions.
Asset-Light Partnerships Open a New Capacity Lever
Perhaps the most consequential structural disclosure was the introduction of an asset-light partnership model, under which outside partners finance, build and operate data center facilities while Nebius supplies the software platform and customer demand in exchange for high-margin, capital-light revenue. Volozh said the company has received "dozens of inquiries" from potential partners since announcing the model, driven by GPUs increasingly being treated as an investable asset class. He framed it as a capital discipline tool for a company that is "still a startup" and cannot build in every region simultaneously. Management was careful to caveat that the model is early stage, but positioned it as a lever that could meaningfully expand 2027 capacity without proportionate balance-sheet strain.
Financial Results: Growth Accelerating Into Margin Expansion
Group revenue grew 454% year-over-year to $582 million, up 46% quarter-over-quarter, with the core Nebius AI cloud business growing 514% to $575 million and now representing 98% of group revenue. Annualized run-rate revenue reached $3 billion at the end of June, up 598% year-over-year and 58% sequentially. CFO Dado Alonso noted this growth came even before the majority of 2026 capacity additions have come online, with most of that impact still to hit in the back half of the year.
Profitability improved sharply alongside growth. Group adjusted EBITDA reached $236 million versus a $21 million loss a year ago, with margin expanding to 41% from 32% in the prior quarter. The Nebius AI segment alone generated a 50% adjusted EBITDA margin, with the gap to the group figure attributed to continued investment in Avride and TripleTen, both still early-stage. Alonso said the company sees "a clear path to capturing more margin, not just this year, but in 2027 and well beyond," as owned data center capacity comes online and begins improving unit economics in the second half of next year.
Capacity Build-Out: 5 Gigawatt Target and the Vineland Overhang
Nebius raised its year-end 2026 contracted power target to 5 gigawatts, positioning itself, in Volozh's words, as "one of just a few companies in the world able to build more than 1 gigawatt of new capacity a year," a pace it intends to sustain into 2027. Andrey Korolenko, overseeing infrastructure build-out, said the company is exceeding its own internal 2026 power projections and is securing hundreds of megawatts of behind-the-meter generation capacity, aided by its partnership with Bloom Energy on fuel cells. Management reaffirmed its 800 megawatt to 1 gigawatt connected power guidance for year-end, though Korolenko flagged that revenue contribution lags connected power by several months given the sequence of commissioning, networking, cluster build and customer onboarding.
On the specific overhang around the Vineland, New Jersey site, where a public hearing on an amended layout plan adjourned without a vote, Tom Blackwell characterized the delay as a normal part of the regulatory process tied to the project's switch to Bloom-based on-site power. He said the company remains "confident that the layout complies with all the applicable local, state and federal laws and regulations," and Korolenko added that building construction is already complete with engineering fit-out progressing, and that the Bloom transition is not expected to affect the project timeline. Investors should note this remains a watch item given the political sensitivity around U.S. data center siting more broadly.
Funding: Prepayments and Asset-Backed Debt Doing the Heavy Lifting
Customer prepayments hit an all-time high in the quarter, with roughly 70% of deals closed in Q2 including an upfront payment. Nebius now expects more than $9 billion of prepayment funding in 2026 alone, directly reducing external financing needs. The company also issued 12.7 million Class A shares through its at-the-market program at a weighted average price of $224, raising roughly $2.8 billion, with 12.3 million shares still available under the program — which Alonso described as "a flexible funding tool" rather than a firm commitment. In July, Nebius closed its first asset-backed debt facility for $775 million, priced at SOFR plus 250 basis points against deployed GPU infrastructure and contracted cash flows from an investment-grade customer, with more than $40 billion of additional backlog available to borrow against on similar terms. Alonso noted the company carries almost no corporate-level debt today, leaving that as an additional lever it has not yet tapped.
2027 Setup and Competitive Positioning
Management reaffirmed full-year 2026 guidance across the board — annualized run-rate revenue of $7 billion to $9 billion, group revenue of $3 billion to $3.4 billion, adjusted EBITDA margin of approximately 40%, and capital expenditures of $20 billion to $25 billion — while declining to formalize 2027 guidance yet. Alonso pointed to the quarter's deal pricing above $20 million per megawatt with sub-two-year payback periods as a baseline for early 2027 pricing, adding that capacity deployment next year will significantly exceed 2026 levels. Asked about xAI's move to sell compute at premium prices, Volozh dismissed any read-through to Nebius's own strategy, framing the broader AI cloud market as expanding from hundreds of billions of dollars annually toward a trillion-dollar scale that hyperscalers cannot fully service alone: "Somebody needs to build the rest of the market, and that will be us." On customer wins, Boroditsky said all four landmark deals were competitive, earned wins against incumbent hyperscaler relationships, with one customer telling the company its proof-of-concept was "quite literally the best POC they have ever had."
Inference and Open-Source Strategy
Roman Chernin detailed the pace of open-model support through Token Factory, citing day-zero availability for releases including Nemotron Ultra, GLM 5.2, Kimi K3, DeepSeek Flash and Minimax 3 within the past four weeks alone, with GLM 5.2 — dubbed by some as a "DeepSeek moment 2" — achieving a 100% quality score in independent benchmarking. Tavily, acquired earlier this year, grew its developer base to 2.5 million from 1 million in February and achieved enterprise security certification during the quarter. Chernin argued the strategic logic is that enterprises are increasingly customizing open-weight models against proprietary data, a trend that favors an infrastructure provider offering flexibility over closed-model lock-in.