Nebius Is Raising the Price of Its AI Compute on Oct. 1. Here's What That Says About the Shortage.
Key Points
Nebius' on-demand hourly rates for four Nvidia GPU models climb between 17% and 21% on Oct. 1.
Management says it could sell all its 2027 capacity today on the terms of its newest contracts.
The company expects to spend $20 billion to $25 billion on capital expenditures this year.
Nebius Group (NASDAQ:NBIS) is about to charge more for the computing power it rents out by the hour. On Thursday, Oct. 1, the artificial intelligence (AI) cloud company's on-demand rate for an Nvidia B300 graphics processing unit (GPU) climbs 21%, from $7.85 per hour to $9.50.
Nebius expects to spend $20 billion to $25 billion on capital expenditures this year, versus guided revenue of $3 billion to $3.4 billion. A company building capacity so fast normally needs to work to fill it, not raise prices on it.
Missed AI’s "Act 1"? Act 2 Could Be 15x Bigger. Most investors think they missed the AI boat because they didn't buy Nvidia in 2005. But according to our analysts, we’re only at the end of "Act 1"—the R&D phase. "Act 2" is the global rollout. Continue »
But what does a price hike now say about how tight AI computing supply is? I think it says plenty. Still, the increase itself touches a smaller share of Nebius' business than the new rates imply.
Image source: Getty Images.
Even older chips cost more
The Oct. 1 increase isn't the first 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 round, the same chips will cost $4.50 and $9.50 -- around 53% and 56% more than five months before. This round includes four Nvidia GPUs (the H100, H200, B200, and B300), with hikes of 17% to 21%, plus Nebius' CPU-only servers.
What strikes me is the H100, a generation behind Nvidia's Blackwell chips (the B200 and B300). Rising prices for older hardware arguably say more about a shortage than a premium on the latest chip does. After all, Nebius' second-quarter shareholder letter already reported over 30% higher pricing for older-generation GPUs versus the first quarter.
Nebius is selling what it builds
"[W]e sold out of capacity because, as fast as we bring capacity online, we can sell it," CFO Dado Alonso said on the company's second-quarter earnings call in August.
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's saving some for customers with immediate needs instead. What's more, its first capacity auction cleared at the highest price Nebius has seen for Nvidia's Blackwell chips, 15% over anything it had charged earlier.
Its latest partnership tells the same story. When Palantir Technologies made Nebius its preferred sovereign AI infrastructure partner on Sept. 8, both also said they'd work to bring new capacity online quicker.
In short, I think the hike is what it looks like: a provider that can't add capacity as fast as customers want it.
What does it do to revenue per megawatt?
Less than the new rates could suggest. Nebius sells computing power both by the hour and under reserved-capacity contracts, but it doesn't break out how much revenue comes from each.
Still, at the end of June, Nebius had around $37.5 billion in remaining performance obligations (signed contract revenue it hasn't recognized yet), over 60 times its second-quarter revenue of $582 million. And these contracts are priced when they're signed, so an Oct. 1 increase doesn't reprice them.
The increase matters more as a benchmark for capacity Nebius hasn't sold yet. Showing how fast that price is rising, Nebius' annual contract value per megawatt went from a 2026 base of around $12 million to over $20 million on deals closed in the second quarter. Management also sees short-term deals at $40 million to $50 million per megawatt, and it signed the first one in the third quarter. And profitability has risen with those prices. The AI cloud business's non-GAAP (adjusted) margin on earnings before interest, taxes, depreciation, and amortization (EBITDA) almost doubled from the fourth quarter of 2025 to 45% in the first quarter. It widened again to around 50% in the second.
Nebius also estimates the capital behind its second-quarter deals pays back in about one year and 10 months, down from two to three years earlier. For a business spending this much, a quicker payback means every new megawatt can rely less on outside funding.
Of course, shortage pricing can reverse. If supply catches up to demand while Nebius and its rivals keep building, hourly rates might drop as fast as they rose. And the build-out 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.
Ultimately, the Oct. 1 increase is strong evidence of how tight AI computing supply is right now. But it's a sign more than a revenue driver.
With shares around $243 as of this writing, Nebius is valued at over $65 billion, or more than 20 times sales at the midpoint of management's 2026 revenue guidance. I think that price already assumes today's pricing holds well into 2027.
Daniel Sparks and his clients do not have positions in any of the stocks mentioned. The Motley Fool has positions in and recommends Nvidia and Palantir Technologies. The Motley Fool has a disclosure policy.