Flexcompute secures strategic investments for the global physics model – Unite.AI

Boston-based Flexcompute announced on October 12, 2026 that it has received a strategic investment from Samsung Ventures and Applied Ventures, LLC, the venture capital arm of Applied Materials, Inc., among other investors. The company said the funding will accelerate its work on the Physics World Model, an artificial intelligence system that predicts how any design will behave in the real world by learning the laws of physics itself. Details of the investment were not disclosed.

History of investments and strategic financing

Flexcompute said it will use the investment to grow its research team, scale GPU computing and extend the world’s model of physics to fluids, electromagnetism, structures, heat and multiphysics couplings. The company described the support as coming from leaders in the semiconductor and manufacturing industries and said it will extend its research toward artificial intelligence that can understand the laws of physics rather than patterns found only in data.

The new investment follows a $22 million Series B round announced by the company on November 29, 2021, led by Coatue Management with participation from GGV Capital, 3W Investment, WI Harper, River Delta and other investors. That release described the company’s computing technologies as accelerating the design process of industrial and consumer products by solving massive differential equations to simulate real-world physics, allowing engineers to evaluate product performance before building physical prototypes. Flexcompute was founded in 2016.

A representative from Samsung Ventures said that AI is changing the way products are designed while those products still have to obey the laws of physics, and described Flexcompute as an AI system that can understand those laws rather than just data. The representative said that the company believes that the Physics World Model will determine how the next generation of hardware will be developed.

Yu Cao, chief AI officer at Applied Materials, said the limitation in putting AI to work on real engineering problems is how quickly each can be validated against the laws of physics. “Flexcompute has built something unique: a simulation fast enough to keep up with artificial intelligence and rigorous enough for manufacturing decisions,” Cao said, adding that in advanced manufacturing this means much of the design cycle happens digitally, saving time and costs.

The model of the physical world and tokenized physics

Flexcompute describes the Physics World Model as an artificial intelligence that learns the laws of physics itself, citing examples ranging from a chip package deforming under heat to a spacecraft re-entering the atmosphere. The company contrasts this approach with current AI systems, which they say learn from the past: text, images and simulation results.

According to the company, its team of more than 70 PhD students has built computational fluid dynamics, electromagnetic, structural and thermal solvers from scratch on GPUs. Through what it calls Tokenized Physics, Flexcompute claims to build both neural networks and first-principles physics directly into the model, so that each prediction obeys the same conservation laws as the physical world and remains accurate even on projects the model has never seen.

Vera Yang, co-founder and president of Flexcompute, said that language models learn the world from text, while hardware is governed by laws rather than text, and no amount of data replaces them. “We are building the verification layer between bits and atoms: a physical model of the world that any engineer or AI agent can ask if a design will work, before it is built,” Yang said. He added that Samsung and Applied see what the company sees: that the next breakthrough in artificial intelligence comes from understanding the physical world.

Demonstration reported and access to physical twins

Flexcompute says the approach is already solving problems that traditional simulation can’t solve. For a complete package of advanced semiconductors with more than 800,000 solder joints, an ASIC, and 32 optical engines, the company reports solving a 5 billion degrees of freedom deformation model within 2% of commercial finite element results in minutes.

Customers access the Physics World Model through Physics Twins, which the company describes as models of its products that engineers and AI agents can question, test, and optimize before anything is built. As part of that workflow, engineers set the goal and approve the outcome while agents iterate, with physics governing what works.