Nebius introduced AI Cloud 3.1, expanding its full-stack AI cloud platform with next-generation NVIDIA Blackwell Ultra compute and new tools designed to improve operational transparency for large-scale AI deployments. The update targets customers moving from pilot projects to production, where visibility into GPU availability, predictable capacity planning, and access to advanced interconnects become critical.
At the infrastructure layer, Nebius is deploying NVIDIA HGX B300 and GB300 NVL72 systems, making it the first cloud provider in Europe to operate both platforms in production. The company also said it is the first globally to run production GB300 NVL72 systems on 800 Gbps NVIDIA Quantum-X800 InfiniBand, doubling interconnect throughput for distributed AI workloads. Nebius noted that the platform combines hardware-accelerated networking, enhanced storage write-back caching, and expanded filesystem performance to reduce bottlenecks, alongside results published in MLPerf Training v5.1.
Nebius AI Cloud 3.1 adds Capacity Blocks and a real-time Capacity Dashboard that expose reserved GPU capacity and live availability across regions, alongside project-level quotas and new object-storage lifecycle rules. The release also expands developer and enterprise features, including native Dstack integration, simplified access to NVIDIA BioNeMo NIM microservices, enhanced Slurm-based orchestration, FOCUS-compliant billing exports, and additional security and governance controls such as HIPAA-aligned audit logs, per-object access policies, VPC security groups, and tighter IAM integration with Microsoft Entra ID.
• Next-generation compute
• NVIDIA HGX B300 and GB300 NVL72 systems
• 800 Gbps NVIDIA Quantum-X800 InfiniBand interconnect
• Hardware-accelerated networking and enhanced storage caching
• Operational visibility
• Capacity Blocks for reserved GPU planning
• Real-time Capacity Dashboard across all regions
• Project-level quotas and object-storage lifecycle policies
• Developer and enterprise features
• Native Dstack integration and Slurm-based orchestration
• Simplified access to NVIDIA BioNeMo NIM microservices
• FOCUS-compliant billing exports for FinOps workflows
• Expanded IAM, audit logging, and network security controls
“AI teams scaling into production need predictable access to advanced compute and clear visibility into how resources are allocated,” a Nebius spokesperson said. “With AI Cloud 3.1, we’re combining Blackwell Ultra infrastructure with capacity and governance tools that let customers plan and operate AI at scale.”
🌐 Analysis
Nebius Group N.V. is a publicly traded AI infrastructure and cloud technology company headquartered in Amsterdam, formed from the restructuring of Yandex N.V. following the divestment of its Russian assets. Led by co-founder and CEO Arkady Volozh, Nebius focuses on building full-stack, AI-optimized cloud infrastructure, combining large-scale GPU compute, high-performance networking, and cloud software designed for training and inference at scale. Its core platform, Nebius AI Cloud (Aether), integrates advanced capacity management, developer tooling, and enterprise governance, with infrastructure deployments increasingly centered on NVIDIA Blackwell-class systems. The company is investing heavily in AI-ready data centers across Europe and other regions, supported by strategic partnerships with NVIDIA and commercial relationships with large AI customers, including hyperscalers and model developers. Alongside its core cloud business, Nebius maintains interests in adjacent technology ventures such as autonomous driving and data platforms, positioning the company as a focused, independent provider of sovereign and hyperscaler-grade AI infrastructure.
