CoreWeave Announces Multi-Year AI Training Agreement With Wayve

CoreWeave (CRWV ) announced a multi-year agreement with Wayve on November 9, 2026, under which the autonomous-driving software developer will train its AI Driver foundation models on CoreWeave Cloud using NVIDIA Blackwell GPUs. The infrastructure will support training, evaluation and simulation as Wayve works toward commercial deployment. The announcement did not disclose a contract value, GPU quantity or computing capacity.

The agreement expands an existing relationship rather than introducing Wayve to CoreWeave’s platform. Wayve began by using Weights & Biases and has now scaled into using CoreWeave Cloud. The duration is described only as multi-year, with no deployment locations or detailed rollout schedule provided.

CoreWeave said Wayve’s AI Driver learns from real-world driving data collected in more than 70 countries. The system is intended for applications ranging from personally owned cars to robotaxis, giving the training program a broad base of driving environments and potential vehicle uses.

CoreWeave characterized the work as frontier model training in which physical behavior and safety are central concerns. The infrastructure must sustain long training runs and provide enough observability to validate safety at scale. That requirement explains why the agreement covers evaluation and simulation alongside training, rather than focusing exclusively on model-training compute.

“Our AI Driver gets better by running this data flywheel: gathering real-world driving experience, training on it, evaluating it, and deploying only what outperforms what is already on the road,” said Silvius Rus, Wayve’s senior vice president of engineering. “Being able to access CoreWeave’s AI cloud, purpose-built for workloads like ours, helps accelerate the commercial deployment of our AI Driver.”

Simulation And Validation Shape The Infrastructure

CoreWeave’s August 24, 2026 technical overview of physical-AI workloads provides context for the agreement’s structure. It identifies autonomous-vehicle uses including world-model training, vision-language-action model fine-tuning, sensor simulation and closed-loop evaluation. These programs can generate terabytes of sensor data per vehicle per day, requiring that data to be moved, stored and used efficiently for training.

The overview also explains why training is not a one-off exercise. Edge cases accumulate after deployment, making continuous retraining, checkpointing and fast data pipelines operational requirements. CoreWeave says most physical-AI work runs through simulation and staged validation before a model operates in the real world, rather than using live deployment as the main test environment.

CoreWeave’s platform description connects those requirements to CoreWeave Kubernetes Service for orchestration, distributed and object storage for petabyte-scale sensor and simulation data, Weights & Biases Models for experiment tracking and Registry for tracing checkpoints to their data and training runs. Those are broader platform capabilities; the agreement announcement does not say which additional services Wayve will use.

The agreement gives CoreWeave a multi-year autonomous-driving workload and extends Wayve’s relationship from experiment tooling into cloud infrastructure. It also ties that expanded infrastructure to Wayve’s commercial-deployment objective. How quickly the combined training, evaluation and simulation environment produces AI Driver versions ready for deployment remains the central operational question, with no commercial-deployment timetable disclosed.