Godela: Physics-informed ML model runs CFD simulations 4,500x faster than GPU-accelerated solvers

Summary

Godela is building a physics-informed ML model to replace computational fluid dynamics (CFD) software for mechanical engineers. Co-founder Cinnamon is a mechanical engineer who built hardware at Apple, Google, and the Stanford Linear Accelerator Center. Co-founder Abujet is a researcher from Stanford, Harvard, and Intel.

The model runs 4,500x faster than GPU-accelerated CFD solvers. Instead of reinforcement learning, the system encodes fluid meshes into a low-dimensional latent space and applies symbolic regression to extract generalizable physics. The founders argue this outperforms standard ML models on novel geometries and conditions.

The company targets two problems. Speed is the first: a drop simulation on a 14-inch MacBook Pro takes two weeks using conventional software and achieves around 70% accuracy. Second is coverage: physics-informed ML handles multi-physics and multi-scale problems where traditional equations break down, filling gaps that conventional simulators cannot address.

Two weeks after launch, Godela signed a $25,000 annual contract with an engineering firm to replace ANSYS, the $30 billion incumbent. The stronger near-term opportunity, according to the founders, lies with enterprise customers facing high-value simulation problems without adequate existing solutions. Aerospace, consumer electronics, and defense are the stated target sectors.

The team consists of three people plus one advisor. A fundraise was in progress at the time of the interview, with terms not yet disclosed.