Nebius AI Cloud Goes Serverless

Nebius is injecting serverless capabilities into its AI Cloud platform with the release of version 3.5, codenamed "Aether." The move aims to eliminate infrastructure management headaches for developers, allowing them to launch AI workloads almost instantly. This update targets real-world AI applications, promising faster prototyping and deployment.

The introduction of serverless AI compute means developers no longer need to configure infrastructure before running experiments, training models, or serving them in production. Nebius will handle the underlying infrastructure and runtime management, freeing up AI teams to concentrate on their core development tasks.

Alongside the serverless push, Nebius is bolstering its hardware offerings with the NVIDIA RTX PRO 6000 Blackwell Server Edition GPU. This addition is geared towards demanding workloads like AI inference, industrial robotics, and complex simulations.

Version 3.5 also brings Nebius's Data Transfer Service, designed to simplify data migration and replication between S3-compatible storage and Nebius cloud regions. This addresses a common pain point for teams working across diverse environments.

Further enhancements include an overhauled configuration wizard for Nebius's managed Slurm-on-Kubernetes solution, Managed Soperator, offering more granular control. Kubernetes observability has also been updated for better cluster management.

The AI application marketplace has been redesigned for faster access to tools and models. Nebius AI Cloud platform updates also include improved user administration, role-based permissions, and new public APIs for billing data export.

All features from the "Aether" 3.5 release are available now, with the serverless service in public preview. The NVIDIA hardware is available immediately.