Icon for Nebius

Nebius

Free Trial

Full-stack AI cloud with GPU infrastructure for training and inference

Nebius is an Amsterdam-headquartered AI cloud platform (Nasdaq: NBIS) that provides on-demand access to NVIDIA H100, H200, B200, and GB200 GPUs along with managed Kubernetes, Slurm orchestration, and storage. Spun out of Yandex in 2024, it operates data centers in Finland and France.

Pricing: On-demand GPU pricing from $0.05/hr (CPU) to $5.50/hr (HGX B200). Commitment discounts up to 35%.

Hosting Cloud
Pricing Usage Based, from $2.00/hr (H100)
HQ 🇳🇱 Netherlands
Founded 2024
License PROPRIETARY
GitHub 97 stars
Compliance SOC 2 · HIPAA · GDPR · SSO
Screenshot of Nebius webpage

What Nebius is

Nebius is a GPU cloud for AI training and inference, offering current NVIDIA hardware on a per-GPU-hour basis with on-demand, preemptible, and committed pricing tiers. It targets enterprise and research teams running sustained AI workloads.

GPUs and pricing

Per-GPU-hour on-demand rates (per their pricing page, June 2026) include HGX H100 at $3.85, HGX H200 at $4.50, HGX B200 at $7.15, HGX B300 at $7.85, RTX PRO 6000 at $1.80, and L40S from $1.55-$1.82 depending on CPU. Preemptible rates run roughly 40-45% lower (for example H100 at $2.15, H200 at $2.45). GB200 NVL72 and GB300 NVL72 systems require contacting sales. Committed reservations advertise up to 35% off on-demand for multi-month clusters.

Who it fits

Nebius fits teams that want a clear three-tier choice between on-demand flexibility, cheaper preemptible capacity for fault-tolerant jobs, and committed clusters for steady workloads. The preemptible tier in particular gives a meaningful discount for training runs that can checkpoint and resume. For raw marketplace bargains, Vast.ai may go lower; for predictable published rates with a preemptible option, Nebius is a closer match.

Is your product missing?

Add it here →