Wayve Taps CoreWeave Blackwell GPUs for AI Driver Training

By Amara Osei

CoreWeave announced a multi-year agreement on November 9, 2026, for Wayve to train its AI Driver foundation models on CoreWeave Cloud with NVIDIA Blackwell GPUs. The work covers training, evaluation and simulation using real-world driving data collected in more than 70 countries, CoreWeave said.

Wayve is developing the AI Driver for uses ranging from personally owned cars to robotaxis, according to CoreWeave’s announcement. The company described the agreement as an expansion of the infrastructure supporting model development.

How does Wayve’s model-development cycle work?

CoreWeave said Wayve began with Weights & Biases before scaling into CoreWeave Cloud. CoreWeave said the expansion adds cloud compute to a development process that must show progress between software versions and sustain long training runs. CoreWeave also said the process needs enough observability to compare models and validate safety at scale.

Silvius Rus, Wayve’s senior vice president of engineering, described the operating model:

“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.”

Rus expects access to CoreWeave’s AI cloud to help accelerate commercial deployment of the AI Driver.

Why does simulation matter for autonomous driving?

CoreWeave’s technical overview identifies world-model training, sensor simulation and closed-loop evaluation as autonomous-vehicle workloads. The overview says autonomous vehicles generate terabytes of sensor data per vehicle per day.

Handling that flow requires storage, networking and training pipelines to operate efficiently together, CoreWeave said. Physical AI development continues after deployment as new edge cases appear, the company said.

Repeated retraining, checkpointing and fast data pipelines are therefore operational requirements, according to CoreWeave. Because failures can have physical safety consequences, most model development runs through simulation and staged validation before real-world use, the company said.

In a reference demonstration outside the Wayve agreement, CoreWeave reported running 1.25 million CARLA autonomous-vehicle simulations in approximately 12 hours. The company used independent shards with one GPU each.

The demonstration varied weather, traffic density, maps, intersections and seeds. CoreWeave said it illustrated the range of simulated conditions that can be generated for model development.

How does the stack support long training runs?

NVIDIA says Blackwell’s Reliability, Availability and Serviceability Engine continuously monitors thousands of hardware and software health data points to predict and intercept sources of downtime. The engine is designed to identify potential faults early, localize problems and aid maintenance, NVIDIA said.

CoreWeave said those capabilities are relevant to the reliability requirements for long training runs described in its Wayve announcement.

CoreWeave’s broader physical-AI platform provides the orchestration, storage and networking beneath those workloads, according to the company.

• CoreWeave says its Kubernetes Service orchestrates training and simulation jobs, while distributed and object storage hold petabyte-scale sensor and simulation data.
• According to CoreWeave, InfiniBand and RoCE networking move data directly between GPUs across servers, bypassing CPUs to reduce idle waiting. The networks carry sensor data, simulation state and gradients during distributed jobs.
• CoreWeave says Weights & Biases Models tracks experiments, while Registry connects a checkpoint to the exact dataset and training run that produced it. Weave supports traces, evaluations and monitoring.

CoreWeave said these experiment tools add version lineage. The records give engineers a route from a problem found in testing back to the model checkpoint, dataset and training run involved, the company said.

For a safety-critical driving system, CoreWeave said, that traceability supports model comparison and validation alongside training capacity.

What will determine faster deployment?

The agreement makes the entire iteration cycle part of Wayve’s infrastructure expansion, according to CoreWeave. The company tied faster commercial deployment to how efficiently training, evaluation and simulation turn real-world driving experience into validated improvements.

Reliability and traceability remain central requirements for safety-critical driving software, CoreWeave said. Over the multi-year term, the commercial test is whether the integrated loop can accelerate deployment without weakening staged validation or traceability.