Nebius Is Raising the Price of Its AI Compute on Oct. 1. Here's What That Says About the Shortage.
Nebius Group is raising prices for its AI computing capacity on October 1st. On that date, the company's on-demand rate for an Nvidia B300 graphics processing unit climbs 21%, from $7.85 per hour to $9.50. This move is striking given Nebius's scale: the company expects to spend $20 billion to $25 billion on capital expenditures this year against guided revenue of $3 billion to $3.4 billion. A company building capacity that fast typically needs to work to fill it, not raise prices on it. But the price increase suggests something notable about AI computing supply constraints.
This is not the first price increase this year. In early May, Nebius' pricing page showed the older H100 chip at $2.95 per hour and the B300 at $6.10. After this October increase, those same chips will cost $4.50 and $9.50 respectively—representing increases of around 53% and 56% over five months. The increase affects four Nvidia GPUs: the H100, H200, B200, and B300, with hikes ranging from 17% to 21%, plus Nebius' CPU-only servers. The H100, a generation behind Nvidia's Blackwell chips, is particularly telling. Rising prices for older hardware arguably signal a shortage more than a premium on the latest technology does. Nebius' second-quarter shareholder letter already reported over 30% higher pricing for older-generation GPUs versus the first quarter.
The company's executives have made clear they cannot add capacity fast enough to meet demand. CFO Dado Alonso stated on the second-quarter earnings call that the company "sold out of capacity because, as fast as we bring capacity online, we can sell it." CEO Arkady Volozh went further in the shareholder letter, saying Nebius could sell all its 2027 capacity today on the terms of its newest deals—it is preserving some for customers with immediate needs. A recent capacity auction cleared at the highest price Nebius has seen for Nvidia's Blackwell chips, 15% above prior rates. The company's partnership with Palantir Technologies, announced September 8th as Nebius's preferred sovereign AI infrastructure provider, reinforced this constraint; both companies committed to accelerating capacity deployment.
The revenue impact is more nuanced than the headline rate increases might suggest. Nebius sells computing power both hourly and through reserved-capacity contracts, but doesn't disclose the split. However, at the end of June, the company had approximately $37.5 billion in remaining performance obligations—signed contract revenue not yet recognized—representing over 60 times its second-quarter revenue of $582 million. These contracts are priced when signed, so the October 1st increase does not reprice existing commitments. The new rates matter more as a benchmark for future deals.
The trajectory of Nebius' pricing power is steep. Annual contract value per megawatt rose from approximately $12 million in 2026 to over $20 million on deals closed in the second quarter. Management is seeing short-term deals at $40 million to $50 million per megawatt, with the first one signed in the third quarter. This pricing power has translated to rapidly improving margins: the AI cloud business's non-GAAP EBITDA margin nearly doubled from the fourth quarter of 2025 to 45% in the first quarter, then widened further to around 50% in the second. Nebius estimates that the capital behind its second-quarter deals pays back in roughly one year and ten months, down from two to three years earlier—a significant improvement that reduces reliance on outside funding for expansion.
Shortage pricing carries risks. If supply catches up to demand while Nebius and rivals continue building, hourly rates could fall as quickly as they rose. The expansion is being funded partly by $5.75 billion in convertible notes sold in August and around $2.8 billion in new shares sold through June. Still, the October 1st increase serves as strong evidence of how constrained AI computing supply is today. With shares valued at over $65 billion—more than 20 times sales at the midpoint of management's 2026 revenue guidance—the market appears to be pricing in the assumption that current pricing will hold well into 2027.