Nebius raises on demand GPU prices October 2026

Nebius plans to raise prices for several of its most in-demand Nvidia GPU cloud instances by roughly 17% to 21% from Oct. 1, 2026, according to a customer notice. The changes would lift the hourly cost of renting H100, H200, B200 and B300 accelerators on an on-demand basis, extending evidence that scarce high-performance computing capacity retains substantial pricing power.

The largest listed increase applies to Nvidia’s B300 GPU, whose on-demand price would rise to $9.50 per GPU-hour from $7.85, an increase of about 21%. The H200 rate would move 20% higher to $5.40 per GPU-hour from $4.50.

H100 instances would increase to $4.50 per GPU-hour from $3.85, a rise of approximately 16.9%, while B200 capacity would cost $8.50 per GPU-hour rather than $7.15, an increase of about 18.9%.

The planned rates had not yet been reflected on Nebius’ public pricing page when the notice was circulated. The increase applies to on-demand capacity, leaving previously signed long-term agreements outside the stated changes. That distinction gives large customers with committed capacity greater protection from near-term spot-market price moves, while companies needing flexible or short-notice compute face higher costs.

Demand visibility extends into 2028

Nebius has pointed to a lengthening order book as a reason for confidence in its infrastructure buildout. At a Goldman Sachs technology conference, company management said its visibility into customer demand had extended beyond 24 months, from roughly 18 months previously. Some customers have reserved capacity through the first half of 2028.

Longer reservations give a cloud provider more certainty when committing capital to data centers, power infrastructure and expensive GPU deployments. They also reduce the amount of unallocated capacity available for customers seeking immediate access, particularly for newer chips used in training and running large AI models.

The company also described an auction-based test for Blackwell-generation capacity in which the clearing price was about 15% above its prior peak price. According to management, that result was about 20% above prices quoted through its standard sales pipeline.

An auction result does not establish a universal market price for GPU computing, since contracts differ in duration, hardware configuration, networking, location and support terms. Yet the gap between auction demand and ordinary sales quotes suggests that some customers are willing to pay a premium when capacity is limited or delivery timelines matter.

Rapid revenue growth accompanies a heavy buildout

Nebius reported group revenue of about $582 million in the second quarter of 2026, up 454% from a year earlier. Its AI Cloud unit generated approximately $575 million, a 514% annual increase, according to the company’s quarterly results.

Adjusted EBITDA reached about $236 million during the quarter, while the AI Cloud segment recorded an adjusted EBITDA margin of about 50%. By the end of the quarter, Nebius said AI Cloud had reached an annualized revenue run-rate of roughly $3 billion.

Those operating figures provide a clearer backdrop for the price changes than the headline percentages alone. Nebius is expanding a business that has shown strong growth and reported positive adjusted EBITDA, but it is doing so with an infrastructure budget far larger than its near-term revenue base.

The company maintained 2026 revenue guidance of $3 billion to $3.4 billion and reiterated a year-end annual recurring revenue target of $7 billion to $9 billion. It also left its 2026 capital expenditure forecast unchanged at $20 billion to $25 billion.

That spending range is many times Nebius’ forecast annual revenue. The mismatch reflects the economics of building AI cloud capacity: providers must pay for chips, data-center equipment, networking and power before much of the related revenue is recognized. Higher hourly rates can help recover those costs, especially on hardware that commands the highest customer demand.

Contracts and financing support the expansion

Nebius has disclosed a Meta contract valued at up to $27 billion, along with a $2 billion investment from Nvidia. The company has also said that customer prepayments cover around 50% to 60% of capital expenditures tied to related projects.

Customer prepayments can reduce the amount of capital a provider needs to fund before new capacity begins producing revenue. They also indicate that at least some customers are prepared to make financial commitments well ahead of delivery, rather than relying entirely on hourly on-demand rentals.

In August, Nebius completed about $5.75 billion in convertible bond financing. Convertible bonds allow companies to borrow at issuance while giving bondholders the option to convert their debt into shares under specified conditions. The financing adds to the resources available for the company’s planned infrastructure deployment, though the scale of its capital expenditure outlook means execution will remain dependent on customer demand, financing access and data-center delivery schedules.

On-demand users face the immediate impact

The Oct. 1 increases place the largest incremental burden on customers that value flexibility over long-term price certainty. An H100 customer running 1,000 GPUs continuously for a 30-day month, for example, would see its gross on-demand compute bill rise from about $2.77 million to about $3.24 million before accounting for storage, networking or other services. The same usage pattern on B300 GPUs would rise from roughly $5.65 million to $6.84 million.

For AI developers, the effect will vary widely. Teams running occasional experiments may absorb higher rental fees more easily than companies operating large inference services or conducting extended training runs. Customers with long-term reservations may be shielded from the stated adjustment, while new buyers could face a choice between higher on-demand prices, longer commitments or alternative hardware configurations.

The price changes also offer a practical measure of the strain created by the AI infrastructure race. Nebius is committing up to $25 billion in 2026 capital expenditure while demand visibility reaches into 2028, and its revised rates indicate that the cost of immediate access to leading GPUs is rising alongside that buildout.