Nebius will raise pay-as-you-go prices for leasing selected Nvidia graphics processing units by 17% to 21% starting October 1, the Amsterdam-based cloud provider said Thursday. It is the company's second price increase in three months, as demand for computing capacity to train and run AI models continues to outstrip supply.
The hikes extend beyond GPUs. On-demand rates for some central processing units made by AMD and Intel will rise by 25%, while memory offerings jump roughly 41%. Nebius said customers who reserve large-scale clusters for multiple months can receive commitment discounts, softening the impact for long-term users.
A market straining to meet AI demand
The price increases follow a quarter of rapid growth in Nebius' AI cloud business. The company signed four customer contracts averaging more than $1 billion each, underscoring the scale of investment flowing into AI infrastructure. Rival CoreWeave said earlier Thursday it is also signing contracts at higher prices.
Neocloud firms like Nebius rent out high-powered chips and data center infrastructure that AI developers need to train and run large models. The business model has drawn intense interest as enterprises race to secure compute capacity, pushing providers to expand aggressively while raising rates.
What the price hikes cover
The new rates apply to selected Nvidia GPUs, certain AMD and Intel CPU-only instances, and memory offerings. Nebius did not specify which exact GPU models are affected, but Nvidia's H100 and H200 chips remain the workhorses of AI training workloads and command premium pricing across the market.
The commitment discounts for multi-month, large-scale reservations signal that Nebius wants to lock in steady revenue from its biggest customers while keeping spot-market buyers on higher variable rates. That structure mirrors pricing strategies seen across other infrastructure-as-a-service providers.
Why this matters for executives and strategy
Rising cloud compute costs directly affect budgeting for AI initiatives. Companies building or fine-tuning models face a market where capacity is tight and prices are climbing in double-digit increments, with no sign of easing. Locking in reserved capacity now, before further hikes, is becoming a tactical financial decision rather than a purely technical one. For finance leaders, the 17% to 41% jumps across compute and memory mean AI project forecasts built on last quarter's pricing are already outdated. The trend also benefits chip suppliers and data center operators, whose pricing power grows as neocloud firms pass costs downstream.
Your membership also unlocks:
