Physical AI and Robotics on Nebius

“Manual handoffs between data generation, simulation, and training means our GPUs can sit idle — costing us both time and money. With the Nebius and NVIDIA joint solution, including OSMO orchestration, our engineers can push a single configuration file and run the entire pipeline end-to-end.

We’re generating thousands of scenario variations with NVIDIA Cosmos on Nebius AI Cloud, powering our AI data flywheel and accelerating the development of our robot foundation model. This allows us to push hardened robot models straight to the edge and cut our iteration cycles from weeks to days.”

“We evaluated several cloud providers, and Nebius offered the best combination of GPU availability, price-performance, and hands-on engineering support for our physical AI and VLM training workloads.

We chose Nebius not just for their tech, but because they share our commitment to data sovereignty — guaranteeing that our European customers' data remains securely within Europe.”

“Data is the biggest determinant of computer vision success. As vision AI systems become more capable, the limiting factor is no longer algorithmic innovation but the quality, coverage, and observability of the data used to train models. Nebius gives our users the compute infrastructure for running complex data tasks such as auto-labeling and generating novel scenes at the speed and scale needed by physical AI systems.”

“With Nebius infrastructure we at Avride built a scalable backbone for autonomous driving — handling petabyte-scale sensor data, millions of simulation scenes, and large-scale model training for perception and planning.

This foundation enabled robust safety validation across both simulated and real environments.

At the same time, Nebius helped us streamline complex workloads into efficient, production-ready pipelines — accelerating iteration and keeping compute costs under control.”