Surrogate models
A surrogate model is a fast, learned approximation of an expensive simulation or experiment. Trained on solver results, it predicts quantities of interest, or full physical fields such as pressure, velocity and temperature, for new designs in milliseconds instead of hours. Surrogate models are used in engineering design, optimisation, uncertainty quantification and digital twins.
Latest news
- Vinci raises $250M at a $1.5B valuation for physics AI in chip design: A physics foundation model for hardware, starting with chip thermals and moving to full systems.
- CANTO: aerodynamic fields straight from parametric CAD: A transformer neural operator that predicts aerodynamic fields directly from parametric CAD, and gives gradients to improve the design.
- Physical Superintelligence launches with a $58M seed round: A new "AI-native physics lab" building virtual physicists, starting with data-centre design.
- PhysicsNeMo 26.08: GeoTransolver and FLARE leave experimental, AeroJEPA arrives: NVIDIA's open-source framework adds new surrogate architectures and GPU mesh tools.
- NVIDIA brings PhysicsNeMo into its Agent Toolkit: Physics AI as libraries that AI agents can call to train and deploy surrogates.
- PhysicsX raises $300M to scale Large Physics Models: A Series C at a reported $2.4B valuation for deep-learning surrogates of engineering simulation.
- Quanscient raises €10M for cloud multiphysics built to feed AI: A Finnish simulation platform designed to generate the multiphysics data surrogates learn from.
- Mistral AI acquires Emmi AI to build an industrial engineering AI stack: The team behind AB-UPT joins Mistral; Linz becomes a Mistral office.
- BeyondMath closes an $18.5M seed round for generative physics: A foundation model trained on first-principles physics, reported up to 1,000× faster than conventional simulation.
- Neural Concept closes a $100M Series C: Goldman Sachs Alternatives leads a round for a CAD-native AI engineering platform used by GM, Renault and F1 teams.
- Luminary Cloud raises $72M for Physics AI: GPU-native CFD that generates the training data and trains the surrogates in one place.
- Emmi AI raises a €15M seed round: Austria's largest seed round, for real-time industrial simulation from the team behind AB-UPT.
Highlighted papers
- CANTO: Tokenizes native NURBS CAD patches and predicts continuous surface and volume fields at any query point.
- AB-UPT: A few anchor points carry full attention; millions of query points read only the anchors, with separate surface and volume branches.
- AeroJEPA: Predicts the latent representation of the flow from geometry and conditions, and decodes a field only when one is needed.