ANSYS Charges $50,000 a Seat for Industrial Simulation. PhysicsX Just Raised $300M to Run It in Seconds.

Industrial simulation software is the most unglamorous $10 billion category in enterprise technology. ANSYS charges up to $50,000 per seat for computational fluid dynamics and finite element analysis tools — software that takes hours or days to run a single design iteration, requires graduate-level training to operate, and produces results that arrive too late in the product development cycle to change fundamental decisions. Boeing, Airbus, and every major automotive OEM use it. The market is sticky, high-margin, and essentially unchanged for thirty years. Until now.

PhysicsX announced a $300 million Series C on June 8 — Temasek-led, at a $2.4 billion valuation. The cap table includes NVIDIA, Siemens, General Catalyst, Applied Materials, and Atomico. That composition isn't incidental: NVIDIA sells the GPUs that run AI inference; Siemens sells Simcenter, a direct legacy-simulation competitor; General Catalyst has a consistent record of backing vertical AI companies attacking defensible enterprise categories. When both the vendor building the replacement and the company selling the legacy product write the check, the disruption thesis has a specific kind of validation.

The capability claim is precise. PhysicsX's AI models predict physical behavior in seconds rather than hours or days, enabling engineering teams to evaluate orders of magnitude more design variants in the same product development cycle. The company has doubled year-over-year revenue, tripled booked revenue, and doubled its customer count in the past twelve months. Revenue growth at that rate, in a market this historically resistant to change, indicates models working in production — not in controlled demonstrations.

The structural argument is the same one that plays out in every vertical AI category where AI-native tools attack domain-specific software built for the pre-GPU era. ANSYS, Siemens Simcenter, and Dassault Simulia were designed to run on hardware from 2005 in workflows built for sequential iteration. PhysicsX builds pre-trained Large Physics Models — a new class of foundation model trained on physical simulation data across aerodynamics, thermal management, structural stress, and fluid flow. The same architectural shift that made language models dramatically more capable than the NLP tools of ten years ago is now running in industrial simulation. The vertical just took longer to notice.

For the application-layer investor thesis, PhysicsX is a clean data point. Vertical AI companies that own the training data for their domain — here, years of physical simulation outputs — build moats that neither general-purpose frontier models nor open-source alternatives easily replicate. NVIDIA designed a chip for AI. Siemens built a competitor into the cap table rather than fight it directly. Both moves carry the same signal: the physics simulation market is being repriced, the incumbents know it, and the first company to build a genuinely scalable AI-native platform holds a structural lead that compounds.