CoreWeave Embeds Engineers to Build Physical AI from Customer Data

CoreWeave brings AI expertise to the factory floor

CoreWeave has launched a field engineering service that places its engineers directly alongside customer teams to build and deploy AI models for physical systems. The Physical AI Field Engineering offering targets industries where domain expertise and machine learning capability rarely overlap—automotive, aerospace, and robotics—and where models must meet strict requirements for explainability, accuracy, and safety.

The service builds on capabilities CoreWeave acquired through Monolith AI and has already been applied across more than 100 engineering projects. Customers retain full ownership of their proprietary data and the resulting models.

From race weekends to production systems

CoreWeave engineers work on-site with customer teams, building models from existing data sources including test bench results, simulation output, production sensors, and live telemetry. Each model is validated against the actual physics of the customer's systems before deployment.

For the Aston Martin Aramco Formula One Team, CoreWeave engineers embedded during live race weekends built a transcription model trained on seven hours of hand-annotated race audio. After 75 iterations, the system now processes 40 radio channels simultaneously, delivering answers to tire strategy questions within pit windows that close in under 30 seconds.

Emma Deutsch, Director of Engineering & Test Operations at Nissan Technical Centre Europe, said the approach helps unlock value from engineering and test data, allowing engineers to focus on vehicle quality, safety, and reliability.

Closing the expertise gap

The service addresses a persistent challenge in industrial AI: the people who understand combustion dynamics or aerospace loads typically cannot build machine learning models, while AI specialists lack the domain knowledge to validate whether their models reflect real-world physics.

Engagements begin with an on-site scoping workshop to map engineering workflows and identify priorities. CoreWeave engineers then prototype solutions end-to-end and remain involved through production deployment across four areas: strategy, simulation infrastructure, real-world data modeling, and agentic learning systems.

The work runs on CoreWeave's integrated engineering AI stack, which includes Weights & Biases for experiment tracking, marimo for data exploration, and CoreWeave ARIA for continuous model improvement. Domain-specific libraries handle anomaly detection, test reduction, and system optimization.

Why it matters

Physical AI applications in manufacturing and engineering fail more often on data quality than on compute or architecture. By embedding engineers who speak both the language of machine learning and the specific physics of customer systems, CoreWeave is attempting to compress the timeline from prototype to production deployment. The approach also keeps model operation in-house—customer engineers learn to retrain and modify models themselves rather than depending on external consultants for every iteration.

Richard Ahlfeld, Senior Vice President of Physical AI at CoreWeave, emphasized that engineering teams adopt new methods only after seeing them work on their own systems. The company delivers working applications and dashboards deployed directly into existing workflows, not static reports requiring separate implementation.

Details of the launch were first reported by CoreWeave in a September 10, 2026 announcement.