Nebius Group's Q2 2026 earnings call revealed a pricing structure that reframes AI compute economics: the company charges $40 million to $50 million per megawatt for contracts lasting up to six months, compared to $20 million to $25 million per megawatt on one- to three-year deals.
CEO Arkady Volozh explained the gap on the call. "Our model was built to build capacity in advance, but not to pre-sell it in advance. That is why we have this available capacity now that can allow us to allocate it to shorter term and much higher margin contracts."
The strategy is deliberate supply discipline. Volozh said Nebius could sell its entire 2027 capacity at the long-term rate today but is choosing not to, keeping inventory available for the higher-priced short-term market. At $40 million to $50 million per megawatt, a single 100-megawatt cluster generates $200 million to $250 million in revenue over a six-month contract.
Short-term customers differ from the core contract base. These are organizations with urgent compute needs on hard deadlines—a lab finishing a training run, a generative video company managing a traffic surge—that cannot wait in a hyperscaler's standard provisioning queue.
The long-term contracts anchor the business. Nebius closed four deals in Q2 averaging above $1 billion each, totaling more than $4 billion. Customers include Reflection, Cohere, a U.S. entity called Neolabs and a large quantitative trading firm. Those agreements include prepayment terms covering 50 to 60 percent of associated capital expenditure, effectively shifting a significant portion of Nebius's hardware financing burden to customers at signing.
Mark, Nebius's commercial lead, described the wins as competitive victories against hyperscalers. He cited one deal closed in May in which a customer needed a large, contiguous GB300 cluster—Nvidia's latest Blackwell-generation hardware. "When we closed with them in May, they cited our responsiveness, our transparency, the white glove support they received as key differentiators," Mark said. That customer is now discussing capacity expansion and evaluating Nebius's inference product, Token Factory.
During Q2, Nebius ran its first capacity auction. The auction cleared 15 percent above the company's highest-ever Blackwell pricing and 20 percent above the pipeline for current-generation hardware. Mark described the mechanism as "all about clear price discovery" in a market where rates shift within a single sales cycle.
Nebius AI now represents 98 percent of group revenue and posted an adjusted EBITDA margin of 50 percent for the segment. The annualized revenue run rate, calculated from the last month of Q2 cloud revenue multiplied by 12, tripled to $3 billion. Group revenue rose 454 percent year over year. CFO Dado said Nebius AI margins will continue to represent the large majority of group adjusted EBITDA.
The pricing architecture Nebius is deploying—a tiered system with a long-term floor, a short-term premium layer, and an auction mechanism—mirrors commodity power markets more than enterprise software. At 50 percent EBITDA margins on the AI segment, hardware and operational costs per megawatt are covered well within the long-term contract rate, making the short-term premium almost entirely incremental margin.
Volozh told analysts Nebius has exceeded its own targets across metrics set entering 2026. The four contracts above $1 billion each represent the company's first at that scale. With 2027 capacity withheld from long-term sale and auctions generating prices above prior Blackwell highs, Nebius is betting short-term compute scarcity will sustain the premium pricing layer through at least the next several quarters.
