Nebius raises GPU rental prices 30%

Nebius has raised the published on-demand prices for several of its flagship GPU instances, according to the company’s pricing page as of Saturday. The changes affect the cloud provider’s Nvidia Hopper and Blackwell offerings, including HGX H100 and HGX B200 systems. Nebius’ site now shows NVIDIA HGX H100 at $2.95 per GPU-hour, NVIDIA HGX H200 at $3.50, and NVIDIA HGX B200 at $5.50. (nebius.com) The move lands as Nebius continues to market itself as an AI-focused cloud with access to Blackwell and Hopper systems in Europe and the United States. (nebius.com) The company says its AI cloud offers “thousand-GPU clusters” and immediate web-console access to as many as 32 Nvidia GPUs, while also steering larger customers toward reserved capacity and direct sales arrangements. ### Which prices changed on Nebius’ public page? Nebius’ current pricing page lists on-demand rates of $2.95 for NVIDIA HGX H100, $3.50 for NVIDIA HGX H200, $5.50 for NVIDIA HGX B200 and $6.10 for NVIDIA HGX B300. (nebius.com) The same page lists preemptible rates of $1.25 for H100, $1.45 for H200, $2.90 for B200 and $3.40 for B300. Nebius’ compute documentation says GPU virtual machines are billed by the second, with prices displayed per hour. The company also says the B200 platform is available in its us-central1 region, while B300 is listed for a private region in the United Kingdom. (nebius.com) ### Why does the H100 number matter most? The H100 figure is the clearest marker because it remains one of the most widely used chips for both model training and inference. (nebius.com) Nebius now lists HGX H100 on-demand at $2.95 per GPU-hour, while its H200 product page separately advertises H200 on-demand access “from $3.50 per GPU/h.” Nebius describes H200 as a step up from H100 for large-language-model and multimodal workloads, citing 141 GB of HBM3e memory on H200 versus 80 GB on H100-class systems. (docs.nebius.com) On its broader AI cloud page, the company positions H100 as a lower-cost platform for building and serving foundation models at scale. ### Is Nebius pushing customers away from pure on-demand usage? Nebius’ own sales language points customers toward longer commitments. The pricing page says users can save “up to 35%” on on-demand rates with multi-month commitments, and the H200 page says three-month-plus reservations can go as low as $2.30 per GPU-hour. (nebius.com) The company’s Blackwell reservation page uses the same approach. Nebius says customers can reserve GB200 NVL72 and HGX B200 capacity ahead of deployment and receive “up to 20% discount off public pricing” for early reservation. (nebius.com) ### How does this fit Nebius’ broader capacity story? Nebius has been expanding AI infrastructure aggressively. On its AI cloud page, the company says it offers Blackwell and Hopper systems across data centers in Europe and the U.S., with non-blocking Nvidia InfiniBand networking and large multi-node clusters. (nebius.com) Recent market coverage of the company’s first-quarter results said Nebius was benefiting from demand that exceeded supply and was increasing capital spending to add capacity. (nebius.com) Those reports are secondary accounts, but they align with the company’s current emphasis on reservations, direct sales and commitment pricing rather than spot-style access alone. ### What can customers watch next? (nebius.com) Nebius’ pricing page remains the main public source for further changes, and the company’s product pages continue to route larger buyers to sales for reserved capacity. The next visible signals are likely to appear on the Nebius pricing page, its H200 and Blackwell reservation pages, or in future updates to its compute pricing documentation. (nebius.com) (marketbeat.com)