Nebius announced general availability of Nvidia GB200 Grace Blackwell Superchip capacity for customers in Europe, positioning the Amsterdam‑headquartered company as an early provider of Blackwell infrastructure in the region. Founder and CEO Arkady Volozh framed the launch as part of a broader push to deliver full‑stack Artificial Intelligence infrastructure that helps startups, enterprises, developers, and researchers build, scale, and deploy next‑generation Artificial Intelligence applications. The company also said Nvidia B200 capacity will be available on demand through Nebius’s self‑service platform and via Nvidia DGX Cloud Lepton, underscoring an emphasis on flexible access models for advanced accelerators.
Beyond hardware, Nebius is expanding its Artificial Intelligence cloud software capabilities. The platform is integrating the Nvidia Artificial Intelligence Enterprise stack, including Nvidia NIM microservices and Nvidia NeMo, to streamline model development and deployment for enterprise use cases and national projects. Nebius highlighted additional first‑to‑market and regional access milestones, including early access to Nvidia Blackwell via Nvidia DGX Cloud Lepton, first‑in‑Europe access to Nvidia Hopper GPUs through the Nvidia Brev platform, and plans to offer instances with the RTX Pro 6000 Blackwell Server Edition later this year. Nvidia’s Dave Salvator said Nebius’s focus on performance, scalability, and compliance aims to speed delivery of reasoning, agentic, and physical Artificial Intelligence.
Nebius pointed to early Blackwell work with research group LMArena, which used GB200 NVL72 infrastructure to bring its Prompt‑to‑Leaderboard system into production. Trained in four days via Nvidia DGX Cloud on Nebius, the system routes real‑time user queries to the most accurate and cost‑efficient models and adapts using live feedback. LMArena’s Wei‑Lin Chiang said the setup improved flexibility, experimentation speed, and accuracy. Separately, biotech firm Prima Mente is using Nebius’s GPU cloud to train and fine‑tune epigenetic foundation models ranging from 90 million to 7 billion parameters, with CEO Ravi Solanki citing elastic, cost‑efficient high performance computing and quick cluster provisioning that compresses model‑development cycles.
Nebius emphasized compliance and sovereignty as differentiators. As the first European‑headquartered Reference Platform Nvidia Cloud Partner, the company says it aligns with European Union regulatory standards and provides the transparency needed for national and sovereign Artificial Intelligence initiatives. One example is Milestone Systems’ Project Hafnia, developed with Nvidia and Nebius, which creates a compliant, anonymized video data platform for training vision‑language models used in smart city Artificial Intelligence solutions. All data for the project is stored and processed within the European Union. Milestone CEO Thomas Jensen said the effort reflects a commitment to transparency, fairness, and regulatory integrity in Artificial Intelligence and positions Nebius as a European cloud provider for compliant visual model training.
