Self-driving software doesn’t get released because it is new. At Wayve, it gets released because it is better. The UK company says its AI Driver improves through a repeating cycle: collect real driving experience, train on it, evaluate the result, and deploy only what outperforms the version already in use. That cycle now has a new home. Wayve has signed a multi-year agreement to train its foundation models on CoreWeave Cloud, running on Nvidia Blackwell GPUs.
Count how many steps in that loop need heavy computing. Training is the obvious one. Evaluation and simulation also consume large amounts of capacity, because a new model must prove itself before it reaches a public road. Gavin Jackson, CoreWeave’s vice president of international, argues that physical AI programs need training, evaluation and simulation infrastructure built to work together from the start. A vendor would say that, of course. Still, the argument fits the loop, since a delay at any step slows the next round.
The route to this deal was short. Wayve already used Weights & Biases, the developer platform CoreWeave acquired in 2025, so the two companies had a working relationship. The new agreement widens it from tracking experiments to running them. Silvius Rus, Wayve’s engineering chief, says access to a cloud built for workloads like theirs speeds up commercial deployment of the driver.
Those models aim to become the intelligence layer behind driverless taxis, and eventually behind self-driving personal cars. That is a long road. Money and hardware alone won’t shorten it, but faster iteration might.
Wayve keeps other options open. It also buys cloud services from Microsoft Azure, and Microsoft joined its Series B round. Amar Shah and Alex Kendall founded the company at the University of Cambridge in 2017. Neither side has revealed the deal’s value, start date or GPU count.
CoreWeave, meanwhile, runs two UK data centers that opened in January 2025. In September 2025 it announced another £1.5 billion ($2 billion) for UK AI capacity, on top of £1 billion ($1.4 billion) it had already committed, including a deployment at a DataVita facility in Scotland. The announcement doesn’t say where Wayve’s training will run.

