The category of replacing (or accelerating) slow first-principles physics solvers — CFD, FEA, multiphysics, plasma — with machine-learning surrogate models (often neural operators: DeepONet, FNO, PINN-class) trained against the solver, turning “days/months per simulation” into “seconds per inference.” The pitch is the same everywhere: compress the design/test iteration loop in physical engineering.
The competitive landscape (2026)
Horizontal platforms (breadth across aerospace/automotive/materials):
- PhysicsX — the category anchor; $300M Series C @ ~$2.4B (June 2026, Temasek), NVIDIA + Siemens backed. CFD-led.
- Luminary Cloud — “Physics AI,” $72M Series B (Sutter Hill / NVIDIA NVentures); Shift pre-trained models.
- Neural Concept — 3D deep-learning surrogates in CAD/sim; $100M Series C (Goldman Sachs Growth).
- Pasteur Labs — “simulation intelligence,” extreme-physics framing (nuclear/aerospace/defense).
- Navier — agentic ML-CFD; the “automate the engineer” wedge.
Semiconductor-vertical (where Zenithon actually competes):
- Vinci (chip design + sim), Cognichip (chip design), SixSense (fab defect/yield), SandBox Semiconductor (etch/dep recipe opt), Zenithon (plasma process surrogate).
Incumbents / displacement target: Ansys (now Synopsys), COMSOL, Lam Semiverse, Applied Materials AIx.
The structural question — the bet lives on Physics Ai Solver Displacement
Does a horizontal foundation physics model generalise across domains (the PhysicsX / Luminary thesis), or do the deepest-data verticals stay defensible because the moat is proprietary calibration data and domain chemistry rather than the neural-net architecture (the Zenithon thesis)? NVIDIA NVentures hedges by backing both sides (PhysicsX, Luminary, Vinci).
That question, the contractual data-rights argument that turns on it, and the screening tests it implies are now the theme page Physics Ai Solver Displacement — moved there 1 Sep 2026 so this page describes the category and the idea page carries the bet. See also Zenithon Seed To A Strategy for the vertical-depth argument as it applies to one company.