Predictive modelling
Trained models that predict performance instantly and optimise geometry for thermals, flow and acoustics.
Onyx is an AI-native engineering orchestration platform. A designer moves a surface; every affected discipline gets its answer in seconds, comments land directly on the geometry, and the decision and version stay attached to the project.
—
Illustrative view of review inside OnyxHub.
Core team of three, first aerodynamic backend running, interface in active development.
More than a decade in automotive engineering, most of it at Audi. We have lived this problem.
NVIDIA Inception member, EXIST-funded through THI, built on patent-published research.
Not more development — the right industrial validation, in a real engineering environment.
The problem
Developing a car — or almost any complex physical product — is relentlessly iterative. A single design change touches everything, and every discipline has its own engineers, its own tools and its own simulation process.
Simulation speed is not the real problem. Engineering knowledge arrives too late, and it is scattered across departments.
What Onyx does
Physics AI can now predict physical behaviour in seconds instead of solving every high-fidelity simulation from scratch. That makes a different way of working possible for the first time.
A designer edits the geometry. Nothing has to be scheduled, requested or waited for.
Aero, cooling, crash, NVH, manufacturing and regulations return their first verdict in seconds.
Engineers and designers comment on the model itself, in place, instead of in a meeting three weeks later.
High-fidelity simulation is spent only on the questions that genuinely need it.
The reasoning, the result and the new design version stay attached to the project history.
We are not replacing simulation tools. We orchestrate them. Onyx connects design changes, engineering models, simulations, people and project history into one workflow — the layer that decides what runs, when, and what it meant.
From the field
Conversations in spring 2026 with exterior designers, studio engineers, programme managers, design directors and design researchers working across German and Chinese manufacturers. What follows is what they told us, in their words.
The problems they described
feasibility
Technical feasibility checks are the headache, that's really frustrating, you burn the money in this field.Academic lead, vehicle design programme, German university
manual work
I have to open the model and check manually, and I often miss something because the requirement changed.Studio engineer, Chinese EV OEM
experience gap
Most designers need years of experience to know something's going to be an issue later.Programme manager, Chinese EV OEM
waiting
He can kill some direction immediately instead of waiting weeks for a full simulation.Studio engineer, Chinese EV OEM
ping-pong
If there was a base where everything got checked first before coming to us, that would be wonderful.Interior designer, ex-Hyundai and Mercedes-Benz
no memory
You could check for similarities to earlier ones, comparing to former vehicles.Research group leader, vehicle design methodology, German technical university
If you don't develop it, somebody else will. Automotive designer, career across five manufacturers
“If you can cut two weeks off the aerodynamic process, that's definitely worth more than 20, 30, 40, 50 thousand euros.”
Academic lead, vehicle design programme, German university
“It would be kind of a dream land if I could tell the designer, as he's pushing the volume in or out here, what the impact on drag is right there.”
Studio engineer, Chinese EV OEM
What they demanded
I don't want to be limited as a designer in my creativity.Automotive designer, career across five manufacturers
Suggest an improvement without harming the main design theme.Show-car designer and modeller, 25 years
Early on we're off by 10 to 15 millimetres and that's fine. Later I need to trust the data.Studio engineer, Chinese EV OEM
Relative comparison between options rather than absolute accuracy.Design director, 25+ years across three manufacturers
Change management, to highlight what changed between models tested weeks apart.Programme manager, Chinese EV OEM
Creative solutions, not just surface changes.Automotive designer, career across five manufacturers
Make it work interactively and quick, with a very simple UI.Automotive designer, career across five manufacturers
These were research interviews, not marketing testimonials. Quotes are verbatim and lightly trimmed for length; speakers are described by role rather than named, because a name on a website needs its own separate permission. Not everyone we spoke to was enthusiastic, and the caveats above are theirs, kept in rather than trimmed out.
Entry point
The long-term platform is multidisciplinary. The first workflow is automotive design and aerodynamics, where our Physics-AI backend already exists. From there we expand across the development process.
Available today
Focused, paid projects where predictive modelling pays for itself inside one development cycle. Same team, same technology, delivered as a service.
Three to five geometries, chosen on intuition.
Labour-intensive preparation before anything solves.
Hours or days per variant.
Expensive, and late.
If it fails, the loop starts again.
Past simulation and test data becomes a trained predictive model.
Thousands of variations evaluated in minutes.
Selected for weight, thermals, flow or acoustics.
High-fidelity simulation to sign off the chosen design.
A measurably better product, sooner.
Geometric variations evaluated per R&D sprint.
Illustrative. Traditional CFD limits variations through meshing and compute time; a trained model infers performance across a far larger space.
Time spent iterating before a prototype is built.
Illustrative. The predictive workflow costs an upfront investment in data and training, then removes most of the trial-and-error phase.
Trained models that predict performance instantly and optimise geometry for thermals, flow and acoustics.
Dormant simulation and test archives turned into datasets that train your own models.
High-fidelity meshing and analysis to validate a final design before tooling.
Where the return actually is, and a roadmap for putting predictive engineering into your pipeline.
Drag reduction, battery thermal management, brake cooling ducts.
Fan blade optimisation, acoustic noise, heat exchanger efficiency.
Internal flow paths, pressure drop, heavy-duty thermal enclosures.
Hull hydrodynamics, propeller design, turbine blade efficiency.
Investors, industrial partners and engineering teams reach us at the same address. If you want to see the workflow and the first model, book a slot or write to us.
Angaben gemäß § 5 DDG (Digitale-Dienste-Gesetz).
Onyx Dynamics ist derzeit nicht im Handelsregister eingetragen. Angaben zu Registergericht, Registernummer und Umsatzsteuer-Identifikationsnummer entfallen daher; sie werden gemeinsam mit den Angaben zur vertretungsberechtigten Person ergänzt, sobald die Gründung abgeschlossen ist.
Verbraucherstreitbeilegung. Wir sind nicht bereit oder verpflichtet, an Streitbeilegungsverfahren vor einer Verbraucherschlichtungsstelle teilzunehmen.
Haftung für Links. Unser Angebot enthält Links zu externen Websites Dritter, auf deren Inhalte wir keinen Einfluss haben. Für die Inhalte der verlinkten Seiten ist stets der jeweilige Anbieter verantwortlich.
Noch offen: Eine Datenschutzerklärung nach Art. 13 DSGVO fehlt und ist unabhängig von einer Registereintragung verpflichtend, sobald personenbezogene Daten verarbeitet werden — hier bereits durch Server-Logfiles und die Kontaktaufnahme per E-Mail. Diese Seite lädt zudem Schriftarten von Google Fonts und Skripte von jsDelivr; dabei wird die IP-Adresse der Besucher an Dritte übertragen. Beides sollte lokal eingebunden werden.