Skip to content
Hire an AI employee
AI EmployeeSuper-AgentsAgent-to-AgentBenchmarksPricingBlog
Guides 08 Choosing a platform

CoreWeave and Wayve: A Multi-Year Deal for the AI Driver

Nitish Garg · Founder & CEO, CellCog · 09 November 2026 · 4 min read · 876 words
At a glanceQuick answers
What was announced?
A multi-year deal for Wayve to train its AI Driver foundation models on CoreWeave Cloud with Nvidia Blackwell GPUs, covering training, evaluation and simulation.
How big is it?
Not disclosed. CoreWeave gave no value, no GPU count and no exact term.
Why does it matter?
Physical AI joins language models as a buyer of the newest GPUs, and driving models need constant evaluation before anything ships.
Editorial illustration on a near-white ground: a teal self-driving car on a looping navy road around a large navy GPU chip, under the words Wayve trains on CoreWeave and a huge 70+ for countries of driving data.
Fig 0Wayve's AI Driver learns from driving data in 70+ countries, now trained on CoreWeave. Made by CellCog's image agent, running GPT Image 2.5.

CoreWeave said on November 9, 2026 that Wayve, the London company building an AI Driver for cars and robotaxis, will train its foundation models on CoreWeave Cloud with Nvidia Blackwell GPUs under a multi-year agreement. The release went out on Business Wire at 8:00 a.m. ET and covers training, evaluation and simulation.

Neither company gave a price, a GPU count or a contract length. Below: what the deal covers, why a driving model is a different kind of training job, and what is still missing.

On this page · 7 sectionsOpen
  1. What the deal covers
  2. Why a driving model is a different training job
  3. Where Wayve stands
  4. What it means for CoreWeave and Nvidia
  5. What this means if you build on AI
  6. What we are watching
  7. Sources
Key points5 · 4 min full read
  1. A self-driving car on top of a computer chip.
    CoreWeave announced a multi-year agreement on November 9, 2026: Wayve will train its AI Driver foundation models on CoreWeave Cloud with Nvidia Blackwell GPUs.
  2. Three cards for training, evaluation and driving.
    The deal covers training, evaluation and simulation, not just one big training run.
  3. A contract with blanked-out lines and a question mark.
    Neither company disclosed a price, a GPU count or the exact length of the contract.
  4. A loop around a road with a check mark.
    Wayve deploys a new model only when it beats the one already on the road, so evaluation compute grows with training compute.
  5. A map pin with a car on a city street.
    Wayve is backed by SoftBank, Nvidia and Uber, and Uber rides powered by Wayve run in London with a safety driver on board.

§ 01What the deal covers

Wayve was already a customer of Weights & Biases, the model tracking tool CoreWeave bought in 2025. Per the release, Wayve “has now scaled into leveraging CoreWeave Cloud”, so the same company now holds both its experiment records and its training compute.

Animated diagram: four stations appear around a loop, driving data from more than 70 countries, training on Nvidia Blackwell GPUs in CoreWeave Cloud, an evaluation where a new model's bar grows past the current model's bar, then a deploy gate that opens only for the better model
Fig 1Wayve's data flywheel: collect, train, evaluate, and deploy only what beats the road
Item Per the release
Customer Wayve, building an AI Driver for automakers and fleets
Hardware Nvidia Blackwell GPUs on CoreWeave Cloud
Workloads Training, evaluation and simulation
Training data Real-world driving collected in more than 70 countries
Term Multi-year, no exact length
Value Not disclosed
GPU count Not disclosed
Table 1What the November 9 release says, and what it leaves out

§ 02Why a driving model is a different training job

A chatbot that gets worse costs an awkward answer. A driving model that gets worse costs a crash, so every new version has to prove it beats the one already on the road. Wayve’s SVP of Engineering, Silvius Rus, describes the loop in the release: “gathering real-world driving experience, training on it, evaluating it, and deploying only what outperforms what is already on the road.”

CoreWeave’s own post on X puts it more bluntly: “None reaches the road until it beats the one already there.” That is why the deal names evaluation and simulation beside training. The compute bill is not one big training run; it is a standing cycle of train, test, compare.

Gavin Jackson, CoreWeave’s managing director for EMEA, called it “one of the most demanding physical AI programs in the world” and framed it as part of a larger move: “As AI moves from the screen into the physical world”, companies need training, evaluation and simulation built to work together.

§ 03Where Wayve stands

The release names SoftBank Group, Nvidia, Uber and Eclipse Ventures as Wayve’s backers. Its AI Driver is mapless and built to run on any vehicle, sold to automakers for assisted driving and to fleets for robotaxis. Wayve’s site says Uber rides powered by Wayve are available in London; reporting describes them as supervised, with a safety driver on board.

§ 04What it means for CoreWeave and Nvidia

CoreWeave says it is used by 9 of the 10 leading foundation model providers. Wayve adds a different kind of buyer: physical AI, where the data is video and sensor logs rather than text, and the evaluation never stops.

The news landed on a rough morning for both stocks. When we read Yahoo Finance’s page at 8:34 a.m. ET, before the market opened, CoreWeave was down 2.9% and Nvidia down 5.5%, the latter amid its board’s investigation into Jensen Huang. The deal is too small a line in either company’s books to explain either move, and neither company gave it a size.

§ 05What this means if you build on AI

  • Physical AI is now a cloud buyer. Driving and robotics models join language models in the queue for the newest GPUs, as at Nvidia’s AI Day Seoul.
  • Evaluation is part of the product. A team that cannot prove a new model beats the old one cannot ship it, so test and simulation compute grows with training compute.
  • Clouds sell the tooling, then the chips. Wayve started on a tracking tool and grew into training on the same provider, the path CoreWeave bought Weights & Biases for.

§ 06What we are watching

  • A size. Any value or GPU count in CoreWeave’s quarterly reporting.
  • Wayve’s next city. Where its robotaxis go after London, and when the safety driver leaves.
  • Rivals. Whether other self-driving companies move training to neoclouds, or to chips like Google’s new TPUs.

§ 07Sources

Frequently asked5 questions

Q1What is Wayve?

A London company founded in 2017 that builds a mapless, hardware-agnostic AI Driver for automakers and fleet owners.

Q2What GPUs will Wayve use?

Nvidia Blackwell GPUs on CoreWeave Cloud, per the release.

Q3Was Wayve already a CoreWeave customer?

It used Weights & Biases, the model tracking tool CoreWeave owns, and has now moved its training onto CoreWeave Cloud too.

Q4Who backs Wayve?

The release names SoftBank Group, Nvidia, Uber and Eclipse Ventures.

Q5Did the deal move CoreWeave stock?

We cannot say. Shares were down before the market opened on November 9, and the release gave no deal size.

Published 09 November 2026 All Choosing a platform →