Nebius

Nebius operates an AI-focused cloud platform spanning the full model lifecycle from training through inference. Its infrastructure is built on custom hardware with non-virtualized GPUs and InfiniBand networking, and the company emphasizes cluster provisioning in minutes, self-service access, repeatability and MTBF/MTTR reliability metrics. The platform includes built-in MLOps tooling, serverless and managed inference, and a product marketed as Nebius Token Factory for production inference workloads.

Workloads supported include large-scale distributed training with Slurm clusters, simulation and evaluation pipelines for robotics and physical AI, foundation model pretraining, and high-throughput inference. Nebius provides NVIDIA H100 and Blackwell-generation infrastructure together with scalable storage, and offers flexible consumption options intended to cover both small experiments and global-scale environments.

Business model

Nebius sells cloud compute, storage and inference capacity to AI developers and enterprises, with flexible consumption options and accompanying engineering and support services, including a free white-glove proof of concept.

Traction

Customer deployments cited on the company site include Revolut running FinCrime agents, chat orchestration and its PRAGMA transaction model on 200+ NVIDIA H100 GPUs and Nebius Token Factory while handling up to 1.2 million support tickets per month with 80% of chats completed without human intervention; Sword Health serving a 200B+ parameter model behind 10 million AI-guided sessions in 2025; RoboForce cutting AI setup pipeline time 70%; Recraft training a 20B-parameter image model; Photoroom storing 300 TB of data and executing multi-month training runs; and Shopify training foundation models on the platform.