How they use Nebius

DataRobot says its Single-Tenant SaaS runs on Nebius Managed Kubernetes. Its technical guide describes NVIDIA NIM inference on Nebius GPUs. The partner solution describes GPU-aware scheduling, dedicated GPU worker pools for hardware-accelerated components, and CPU pools for cost efficiency. Containerized models, agents, and applications receive governed endpoints with autoscaling and monitoring. The guide describes Nebius storage supporting model artifacts, embeddings, and telemetry for continuous evaluation and logging. It describes dedicated cluster configurations supporting tensor parallelism and KV-cache-intensive inference, with InfiniBand RDMA for cross-node scaling. Dedicated tenant boundaries use Nebius IAM and network policies. DataRobot CEO Debanjan Saha says the joint offering includes dedicated NVIDIA Hopper and Blackwell GPU infrastructure for sustained agent workloads. DataRobot says the joint offering addresses latency spikes and unpredictable costs and reduces deployment delays to days instead of months.

DataRobot’s Agent Workforce Platform integrates with Nebius Token Factory for model experimentation on a pay-per-token basis, including NVIDIA Nemotron. Its guide describes a path from Token Factory experimentation to dedicated inference deployment, such as NVIDIA NIM. The joint offering includes OpenTelemetry monitoring and tracing for agent health, resource utilization, and token quotas. Nebius’s stack description also includes NVIDIA NeMo Guardrails for policy enforcement. DataRobot CPO Venky Veeraraghavan says GEICO worked with DataRobot and Nebius to shape and validate the offering.

Nebius reports benchmarking the validated stack on B200 systems. That benchmark describes the joint platform configuration and does not establish DataRobot’s current GPU allocation.