Nebius AI Cloud 3.1 Delivers Next-Gen Nvidia Blackwell Ultra Compute

Nebius B.V. announced Nebius AI Cloud 3.1, bringing next-gen Nvidia Blackwell Ultra compute and enhanced operational capabilities to the latest release of its full-stack AI cloud platform. Version 3.1 builds on the foundations of Nebius AI Cloud ‘Aether’, adding transparent capacity management and expanded infrastructure to deliver the operational visibility and resource-planning capabilities that customers need as they scale AI in production.

As customers move beyond experiments to large-scale adoption, they are identifying clear operational priorities, including real-time visibility into GPU capacity across regions, transparent resource allocation for multi-team environments, and access to next-gen compute as models grow in complexity. The firm’s AI Cloud 3.1 addresses these needs across three dimensions.

Server racks in Nebius’ data center in Finland

The company is deploying Nvidia Blackwell Ultra infrastructure globally, with Nvidia GB300 NVL72 systems and Nvidia HGX B300 systems already being used by customers. Nebius is now the first cloud in Europe to operate both platforms in production. The firm’s AI Cloud is also the first provider globally to run production GB300 NVL72 systems on 800Gb Nvidia Quantum-X800 InfiniBand interconnect – doubling throughput for distributed workloads. This, alongside hardware-accelerated networking and enhanced storage caching that eliminate infrastructure bottlenecks, further complements leading results in MLPerf Training v5.1 benchmarks.

To meet enterprise demands for operational transparency, version 3.1 introduces Capacity Blocks and a real-time Capacity Dashboard that give customers complete visibility into their reserved GPU capacity and availability across all data center regions. Project-level quotas and new lifecycle object storage rules extend granular control over resource allocation and costs.

The Nebius AI Cloud platform ecosystem is also continuing to expand. Beyond a new native integration with Dstack, version 3.1 simplifies deployment and consumption of Nvidia BioNeMo NIM microservices, including Boltz2, Evo-2, GenMol, and MolMIM, without requiring NGC keys or Nvidia AI Enterprise licenses. Additional features focused on developer usability include enhanced Slurm-based orchestration with Manager Soperator, FOCUS-compliant billing exports and various console UX improvements.

The latest release also builds on Aether’s enterprise-ready security foundation with object storage data-plane audit logs for HIPAA compliance, per-object access controls, and VPC security groups. Enhanced IAM with Microsoft Entra ID integration and granular service roles provide stronger governance with simplified administration, advancing the compliance capabilities that enable customers in highly regulated sectors and government to deploy production AI workloads.

Nvidia Blackwell systems deployment in Nebius’ data center in Finland

Nebius AI Cloud 3.1 at glance:

Next-gen compute

Nvidia HGX B300 and GB300 NVL72 systems with 800Gb Nvidia Quantum-X800 InfiniBand
Hardware-accelerated networking, storage write-back caching, and expanded filesystem performance scaling
Leading performance in MLPerf Training v5.1 benchmarks

Operational visibility and efficiency

Capacity Blocks for reserved GPU planning and tracking
Capacity Dashboard with real-time GPU availability across all regions
Project-level quotas for granular resource control
New lifecycle object storage rules for cost optimization

Developer productivity

Nvidia BioNemo NIM microservices for Healthcare and Life Sciences workflows
Slurm-based scaling and Dstack native support FOCUS-compliant billing exports for FinOps integration

Security and governance

Audit logs with object storage data-plane events for HIPAA-compliant configurations and per-object access controls
VPC Security Groups for network-level protection
Enhanced IAM with Microsoft Entra ID login, visual SSO management, and granular service roles

Resources:
Blog: Nebius AI Cloud “Aether 3.1” release: Next-gen compute for AI operations at scale
Blog: Nebius delivers Europe’s first live Nvidia GB300 NVL72 deployment
Blog: Running Nvidia NIM and Nvidia Blueprint in Nebius AI Cloud
Blog: Visibility that drives elasticity: Introducing Capacity Blocks and Capacity Dashboard