Nebius Group (NASDAQ: NBIS) released Aether 3.5, an update to its AI cloud platform that introduces serverless computing capabilities and adds the NVIDIA RTX PRO 6000 Blackwell Server Edition GPU to its infrastructure offerings.

The serverless feature set includes three services: DevPods for development environments, Jobs for batch processing workloads, and Endpoints for model inference testing. Jobs and Endpoints are available in public preview through the web console and command-line interface, while DevPods remain in private preview with general availability expected within three months.

The NVIDIA RTX PRO 6000 Blackwell Server Edition GPU features 96 GB of GDDR7 memory and targets applied AI workloads in robotics, physical simulations, and drug discovery. The GPU is available through account managers, with self-service access planned for later.

Nebius also introduced a Data Transfer Service for moving datasets between S3-compatible storage systems and across regions within its cloud platform. The service is in private preview with general availability expected next month.

The update includes improvements to Managed Soperator, the company's Slurm-on-Kubernetes orchestration system, with a new configuration wizard and expanded cluster architecture controls. The platform now supports enterprise single sign-on integration.

Additional platform enhancements include support for Kubernetes 1.33, expanded encryption capabilities, generally available Kubernetes Audit Logs, and a new 1-GPU preset for NVIDIA HGX B200. The company also introduced a public API for billing export that provides FOCUS-compliant data.

The Applications section of the platform console received interface updates to improve navigation and tool discovery, including support for deploying models from the NVIDIA NIM Microservices catalog with pay-as-you-go pricing.