CoreWeave signs multiyear deal with Wayve to train autonomous-driving AI on Blackwell GPUs

CoreWeave announced on November 9, 2026, that it had signed a multiyear deal with Wayve. Wayve will train foundation models on CoreWeave Cloud equipped with NVIDIA Blackwell GPUs and expand its AI infrastructure for training, evaluation and simulation.

Wayve says it uses real-world driving data from more than 70 countries and deploys models that, after training and evaluation, outperform existing driving performance. The deal will bring a cloud equipped with Blackwell GPUs into that process for training, evaluation and simulation, giving model developers an option to use computing infrastructure and validation processes together.

The focus of the deal is not only training foundation models on Blackwell GPUs, but also including training, evaluation and simulation within the scope of the same cloud infrastructure. Wayve’s stated process is to collect real-world driving data from more than 70 countries, train and evaluate models, and deploy models that outperform existing driving performance. In practice, the significance of the deal will depend on whether it enables the company not only to secure computing resources but also to continually compare model versions and validate them before deployment. Related coverage by this newspaper on computing infrastructure for physical AI includes a report that NEXCOM and NexCOBOT showcased RoboWIZ and four technology partners at GTC Berlin, which covered a robotics platform based on NVIDIA IGX Thor, and Seeed Studio lists a carrier board for Jetson AGX Thor, which covered an edge-side product listing. These do not indicate a connection to the deal with Wayve, but help distinguish the cloud-based model training covered in this announcement from a different layer of computing infrastructure: robotics edge platforms. In autonomous driving, how cloud-based training and evaluation connect with execution in vehicles is an issue, but the announcement does not describe in-vehicle hardware or how it would connect. The competitive implications lie not simply in providing GPUs, but in securing an operational platform that links real-world driving data with model improvement and evaluation. Wayve’s expansion from using Weights & Biases to using CoreWeave Cloud indicates an expansion of its development process, including experiment management and computing infrastructure. However, the specific system configuration and data-transfer methods are not known. Points to watch include the number of GPUs used, the scale of training, evaluation metrics, the scope of simulation, and the vehicles or regions targeted for deployment. The announcement describes the agreement as multiyear but does not specify its duration. It also does not disclose the volume of real-world driving data or the standards for safety evaluation. The deal’s effectiveness will therefore need to be judged by how its approach to ongoing model improvement connects to actual evaluation and deployment.

Wayve’s AI Driver learns from real-world driving data collected in more than 70 countries. Wayve had begun using Weights & Biases and is now expanding its use of CoreWeave Cloud. Wayve says it repeatedly collects real-world driving experience, trains and evaluates models, and deploys only models that outperform existing driving performance. The agreement is described as multiyear, but the announcement does not specify its duration or the scale of computing resources.