Nebius (Nasdaq: NBIS) announced public live spot pricing for GPU compute and expanded its AI Builder Program across Europe with an additional $10 million in compute grants, according to a company press release issued at Nvidia's GTC conference in Berlin.

The new pricing structure includes preemptible spot pricing, reserve auctions for short-term capacity blocks, and pay-as-you-go access. Nebius is also piloting a capacity-rebate program that allows customers to recover value by releasing idle GPUs back into the spot capacity pool. Live pricing is available through third-party platforms including Shadeform, Silicon Data, and GPU Finder.

Nebius operates compute capacity across Finland, France, the UK, Spain, Estonia, and Iceland. The company said it is evaluating additional European deployments.

Several European companies are cited as Nebius customers. Prosus reported doubling throughput using Nebius Token Factory for open-model workloads. Neobank bunq said deploying open weight models on Nebius infrastructure costs 60% less per answer compared to leading closed models. Black Forest Labs used Nebius Nvidia B200 infrastructure for research underlying its FLUX 3 models.

Nebius also announced the addition of agentic workflows to its Physical AI Workbench, with customers reporting reductions in engineering time of up to 40%.

The company said it is serving as a compute supplier for the Next Frontier AI Challenge run by SPRIND, Germany's Federal Agency for Breakthrough Innovation.

Nebius Academy is collaborating with Nvidia's Deep Learning Institute to expand hands-on AI training for developers. The AI Builder Program expansion includes founder funding opportunities and community programs.

Nebius claims to be the first AI cloud to adopt Nvidia Groq 3 LPX, an inference accelerator designed for agentic systems, and says it was among the first to validate Nvidia Vera Rubin NVL72 rack-scale systems. The company holds Nvidia Exemplar Cloud recognition for training workloads on Nvidia HGX B300, GB300 NVL72, and HGX H200 hardware.