Key Takeaways:
Nebius plans to deploy over 1 GW of AI compute annually from 2027, backed by $40 billion in customer commitments and a fivefold revenue surge.
Nebius plans to deploy more than 1 gigawatt of AI computing capacity annually starting in 2027, enough to power roughly 750,000 US homes and position the Amsterdam-based cloud provider among the largest AI infrastructure builders.
"We are converting growing demand into contracted, profitable growth," Arkady Volozh, founder and chief executive officer of Nebius, said in a shareholder letter.
Revenue jumped 454 percent year-over-year to $582.3 million, beating the $572.75 million consensus compiled by LSEG, as the AI cloud unit grew nearly sixfold to $574.9 million. Adjusted EBITDA reached $236.2 million versus a $21 million loss a year earlier. The company raised its year-end contracted power target to 5 gigawatts from more than 4 gigawatts.
The expansion carries a heavy price tag — $5.7 billion in quarterly capital expenditures, above the $4.7 billion analysts expected per Visible Alpha — but customer prepayments of more than $9 billion expected this year and $40 billion in total commitments provide funding support. Nebius expects to sell all of its planned 2027 capacity at current contract terms.
Deals average $1 billion as pricing power strengthens
Nebius signed four large AI cloud contracts during the quarter with an average total contract value exceeding $1 billion each. Named customers include AI developers Reflection and Cohere, plus a US AI laboratory and a quantitative-trading firm. Annual contract value on the new deals runs $20 million to $25 million per megawatt, compared with a base of about $12 million across its 2026 capacity. Total contract value nearly quadrupled from the first quarter, while new-customer contract value jumped more than ninefold. Most of the capacity supporting these deals arrives late this year and will contribute primarily to 2027 revenue, according to Volozh's letter.
The demand picture extends beyond Nebius. Larger rival CoreWeave raised its annual forecasts a day earlier, reporting $2.58 billion in second-quarter revenue, and shares of AI infrastructure companies broadly advanced as investors concluded that computing capacity remains scarce relative to demand. When asked about new entrants such as xAI, Volozh said demand for AI computing remains well ahead of supply. Emarketer analyst Jacob Bourne agreed that demand for AI cloud capacity keeps soaring despite intensifying competition, but added that "the bigger question is how diversified and durable that demand proves to be beyond the AI industry itself."
Customer prepayments are becoming a critical funding source. About 70 percent of the quarter's new deals included upfront payments covering an estimated 50 to 60 percent of related capital expenditures. Those payments helped produce $2.25 billion of operating cash flow during the quarter despite a GAAP net loss of $190.4 million. Nebius ended June with $8.04 billion in cash, up from $3.68 billion at the end of 2025. It raised about $2.8 billion through an at-the-market equity program, selling 12.7 million Class A shares at an average price of $223.60, and secured a further $775 million in July through an asset-backed facility priced at SOFR plus 2.5 percentage points.
The company is also introducing a partnership model under which outside investors finance and own data-center assets while Nebius supplies the cloud software, systems design and customers. The structure could reduce the amount of capital Nebius must provide directly, though growth remains dependent on securing power, procuring advanced chips and completing facilities on schedule.
The economics of GPU longevity also support the model. A recently disclosed CoreWeave contract shows Nvidia's A100, launched in 2020, still winning new customer contracts in 2026 with terms extending through 2029, suggesting older GPUs can shift toward inference workloads and keep generating revenue well beyond their initial deployment cycle.
Nebius shares jumped 17 percent following the results. The stock has traded between roughly $50 and $300 over the past year as investors weighed surging AI-infrastructure demand against execution risks. Based on Goldman Sachs estimates, the stock trades at about 18.1 times 2026E enterprise value-to-sales, a multiple that would compress to roughly 6.1 times if 2027 revenue reaches the approximately $13.7 billion that current orders and capacity plans support.
This article is for informational purposes only and does not constitute investment advice.
