CoreWeave signs multi-year deal to train Wayve’s self-driving AI

Wayve will use CoreWeave’s cloud to train the AI behind its autonomous-driving software under a multi-year agreement announced November 9, 2026. The deal brings NVIDIA Blackwell GPUs into Wayve’s development process, adding computing infrastructure for training, evaluation and simulation as the company works toward broader commercial deployment of its AI Driver.

Computing for more than training

CoreWeave describes two requirements for that work: keeping long training runs reliable and providing enough visibility into results to compare successive model versions. Its announcement puts training, evaluation and simulation together as infrastructure needed to validate improvements and safety at scale.

The cloud deal follows an existing software relationship. Wayve began using Weights & Biases before extending its use to CoreWeave Cloud, according to the announcement. The newly announced commitment therefore adds cloud computing to a relationship that started with another part of the development workflow.

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

Silvius Rus, Wayve’s senior vice president of engineering

Driving data feeds the next version

Wayve’s AI Driver learns from real-world driving data collected in more than 70 countries, according to CoreWeave. That experience supplies the material for the improvement cycle Rus describes. His account makes evaluation a gate between training a candidate model and deploying it, rather than treating completion of training as sufficient.

Wayve describes its AI Driver as mapless and hardware-agnostic: designed to work without maps and without being tied to one vehicle hardware setup. It targets automakers and fleet owners, with applications ranging from personally owned cars to robotaxis.

Rus said access to CoreWeave’s purpose-built AI cloud would help accelerate commercial deployment. That remains an expectation: the announcement does not quantify a reduction in development time or provide a deployment date. It also leaves the agreement’s financial value, GPU count and exact duration beyond “multi-year” undisclosed.