Nebius
Nebius is an AI-native GPU cloud platform that rents NVIDIA H100 through GB200 clusters with managed Slurm, Kubernetes and an inference API.
Overview
Nebius
Nebius is an AI-native GPU cloud, traded on the Nasdaq as NBIS, that builds and operates large NVIDIA clusters for training and inference. Spun out of Yandex's non-Russian assets and led by Yandex co-founder Arkady Volozh, Nebius designs its own racks and data centers rather than reselling another provider's capacity. The platform gives AI teams bare-metal and virtualized access to H100, H200, B200 and GB200 GPUs, wired with InfiniBand and orchestrated through managed Slurm or Kubernetes. Alongside raw compute, Nebius runs a Token Factory inference API and managed data services. For labs and startups that want owned-hardware scale with a public-company balance sheet behind it, Nebius is a serious option.
Production credibility: Nebius Group N.V. is headquartered in Amsterdam and trades on Nasdaq under NBIS; its market capitalization was roughly $58 billion in late May 2026. The company was created in July 2024 when Yandex N.V. sold its Russian assets for about $5.2 billion (the largest corporate exit from Russia) and renamed the remaining non-Russian businesses Nebius Group, with Yandex co-founder Arkady Volozh (who first co-founded Yandex in 1997) as CEO; Nasdaq trading resumed in October 2024. NVIDIA made a strategic investment of approximately $2 billion. In Q1 2026 Nebius reported revenue of about $399 million, up roughly 684% year-over-year, and guided to $3.0-3.4 billion in 2026 revenue. Its contracted backlog approaches $50 billion, anchored by a reported ~$27 billion multi-year Meta deal and a commitment of up to $19.4 billion from Microsoft. Named customers include Brave, Recraft, CentML and the open-source vLLM project. Nebius is an NVIDIA Reference Platform Cloud Partner and is building an owned AI data center campus in Pennsylvania.
Key Features
- Bare-metal and virtualized NVIDIA GPUs: H100, H200, B200, B300, GB200 NVL72 and GB300 NVL72
- High-bandwidth InfiniBand networking for multi-node distributed training
- Managed Slurm and managed Kubernetes orchestration for cluster scheduling
- Token Factory: a hosted LLM inference API for serving open-weight and custom models
- Managed data services including MLflow, PostgreSQL and Apache Spark
- Terraform provider, public API and CLI for infrastructure-as-code provisioning
- Owned data centers and racks rather than rented third-party capacity
- 24/7 support with solution architects for large training deployments
Ideal Use Case
AI labs and infrastructure-heavy startups use Nebius to train and serve large models on reserved NVIDIA clusters when they need owned-hardware scale and predictable multi-year capacity rather than spot availability on a hyperscaler.
How Nebius differentiates
Versus CoreWeave, Lambda and Together AI, the main difference is that Nebius is a public company (NASDAQ: NBIS) that owns and operates its own data centers and racks, which can mean steadier long-term capacity for anchor customers. Like CoreWeave, it sells reserved GPU clusters more than casual on-demand instances, so it suits committed training workloads over quick experiments. Compared with Lambda, Nebius leans further into full-stack managed orchestration plus a Token Factory inference layer rather than just GPU rental. The trade-off is that Nebius is not the place for someone who wants a single cheap A100 for an afternoon; its center of gravity is large, planned deployments. Its Yandex heritage also means most of its operating history sits outside the US market.
FAQ
Q: Is Nebius publicly traded and what is its ticker? A: Yes. Nebius Group N.V. trades on the Nasdaq under the ticker NBIS. Its market capitalization was roughly $58 billion in late May 2026, and it guided to $3.0-3.4 billion in 2026 revenue.
Q: Who founded Nebius? A: Nebius Group was formed in July 2024 from the non-Russian assets of Yandex N.V. and is led by Yandex co-founder Arkady Volozh, who first co-founded Yandex in 1997. The company is headquartered in Amsterdam.
Q: How much has Nebius raised or what backing does it have? A: As a public company Nebius is funded through the public markets rather than venture rounds, but it also took a strategic investment of approximately $2 billion from NVIDIA and has a contracted backlog approaching $50 billion.
Q: Nebius vs CoreWeave: which should I choose? A: Both rent large NVIDIA GPU clusters for AI. Nebius is publicly traded (NBIS) and operates its own data centers with a managed Slurm/Kubernetes stack and a Token Factory inference API; CoreWeave is a larger pure-play GPU cloud. Choose based on region, GPU availability, pricing and contract terms.
Q: What GPUs and services does Nebius offer? A: Nebius offers NVIDIA H100, H200, B200, B300, GB200 NVL72 and GB300 NVL72 GPUs over InfiniBand, plus managed Kubernetes and Slurm, managed data services, and a Token Factory LLM inference API.
tl;dr
Nebius is an AI-native GPU cloud, publicly traded as NASDAQ: NBIS, that owns its data centers and rents NVIDIA H100 through GB200 clusters with managed Slurm, Kubernetes and an inference API. Spun out of Yandex and led by Arkady Volozh, it reported ~$399M revenue in Q1 2026 and a ~$50B backlog.
Related
Looking for more options? Browse the AI Infrastructure directory or read our best AI infrastructure tools listicle. Nebius is also tracked on Crunchbase.
Why Use Nebius
Editorial Review
Our take on Nebius.
Nebius is an AI infrastructure platform renting H100 and GB200 GPUs with managed cluster orchestration for training and inference workloads.
What works
- Managed Slurm and Kubernetes reduce cluster setup overhead
- Direct access to current NVIDIA GPU generations (H100, GB200)
- Inference API removes bare-metal management friction
What doesn't
- Narrow audience—requires familiarity with distributed ML infrastructure
- Competes with larger, more established GPU cloud providers
Nebius operates as a GPU cloud provider focused on AI workloads, offering NVIDIA H100 and GB200 clusters with built-in Slurm and Kubernetes orchestration. Rather than forcing you to wire up your own cluster management, the platform handles job scheduling and resource allocation out of the box. The inference API layer lets you deploy models without managing underlying infrastructure directly. As of 2026, this positions it in a crowded market alongside RunPod and other GPU rental services, but the managed orchestration angle reduces operational friction for teams running distributed training or serving at scale.
The 4.93 community rating suggests users find real value, though the platform remains a specialized play—not a household name like cloud commodities. You're essentially trading flexibility and bare-metal control for faster time to experiment. It works best if you're already comfortable with Slurm and Kubernetes and want those layers pre-wired rather than spending weeks building CI/CD plumbing. The H100 and GB200 availability matters if you're targeting specific NVIDIA generation capabilities for your models.
User Reviews
Similar Tools
