PhysicsX Raises $300 Million to Bring AI-Driven Physics Simulation to the World's Factories
London-based PhysicsX, founded by two Formula 1 engineers, raised $300M in a Series C led by Temasek, valuing the industrial AI startup at $2.4 billion—more than double its year-ago valuation. Its Large Physics Models compress multi-day engineering simulations into seconds, accelerating design cycles for aerospace, semiconductor, and clean energy industries.
When Jacomo Corbo and his co-founder left the hyper-optimized world of Formula 1 engineering to start PhysicsX, they carried with them a motorsport engineer’s obsession with iteration speed. In F1, simulating airflow over a revised front wing can determine whether a car gains or loses tenths of a second per lap—and hundredths of a second separate championship contenders from also-rans. In commercial manufacturing, the same simulation cycles that take days or weeks to run can cost companies millions in delayed products and missed market windows.
On June 8, 2026, PhysicsX announced it had raised $300 million in a Series C funding round led by Singapore’s Temasek, with new investors M&G and Intrepid Growth Partners joining existing backers Nvidia, Applied Materials, Atomico, General Catalyst, and Siemens. The round values the London-based company at $2.4 billion—more than double its roughly $1 billion valuation from just twelve months ago—and brings total capital raised to approximately $500 million since the company’s founding.
From the Pit Lane to the Production Line
PhysicsX occupies a specialized but rapidly expanding niche within the AI landscape: physics-based industrial simulation. Traditional computational fluid dynamics (CFD) and finite element analysis (FEA) simulations—the tools engineers use to model how heat, stress, fluid, and electromagnetic forces interact with physical components—are extraordinarily compute-intensive. Simulating the thermal behavior of a next-generation jet turbine blade can take days to run on the most powerful HPC clusters available. Running it thousands of times to explore a design space is effectively infeasible.
PhysicsX’s approach is to train deep learning models—which it calls Large Physics Models (LPMs)—on massive datasets representing the behavior of physical systems: thermodynamics, fluid dynamics, structural mechanics, and coupled multiphysics interactions. Once trained, an LPM can evaluate a physics scenario in milliseconds rather than hours, enabling engineers to explore orders of magnitude more design iterations in the same calendar time.
“We’re compressing complex designs and simulation processes that once took months into seconds,” CEO Jacomo Corbo said in a statement accompanying the raise. The implications extend beyond speed: certain optimization problems only become tractable when the cost of each evaluation drops from hours to fractions of a second. An LPM doesn’t just accelerate existing workflows—it makes previously impossible searches routine.
A Cap Table That Reads Like a Strategic Alliance
The composition of PhysicsX’s investor list is almost as significant as the funding amount. Nvidia’s continued backing reflects a belief that industrial simulation will become one of the most computationally demanding workloads on next-generation GPU hardware—and that PhysicsX is a likely driver of that demand. Applied Materials, a $70 billion semiconductor equipment company, sees faster physics simulation as directly accelerating its own roadmap for next-node deposition and etch processes. Siemens, with its deep industrial automation portfolio and Simcenter simulation platform, is positioned to integrate LPMs into products it already sells to hundreds of thousands of engineers.
Temasek’s decision to lead the round signals something broader: sovereign wealth funds are beginning to recognize that infrastructure-layer industrial AI—less glamorous than consumer chatbots but potentially more economically significant—warrants direct equity stakes rather than exposure through listed equities. Singapore has made advanced manufacturing competitiveness a national policy priority, and owning a position in the company most likely to define AI-accelerated engineering fits that strategy precisely.
Four Industries Where PhysicsX Is Making Its Bet
PhysicsX has organized its go-to-market around four verticals where simulation bottlenecks are most acute and commercial returns most visible:
Aerospace and defense: Airframe and engine development involve thousands of high-fidelity CFD runs per program, and development cycles that routinely span a decade. PhysicsX’s platform is particularly valuable for novel configurations—hydrogen combustion engines, blended wing bodies, urban air mobility designs—where traditional empirical databases are thin and computational simulation must substitute for flight experience.
Semiconductor manufacturing: Chip fabs continuously tune hundreds of plasma and chemical deposition processes for yield optimization. Process simulation has become one of the bottlenecks in sub-2nm node development, and Applied Materials’ strategic investment suggests PhysicsX is co-developing solutions in this space.
Clean energy: Fusion reactor design, offshore wind turbine structural optimization, and next-generation battery pack thermal management all involve complex multiphysics problems where the engineering talent pipeline cannot keep pace with the capital being deployed. LPMs offer a partial solution: multiplying the productive output of existing engineers rather than waiting for a generation of new ones to be trained.
Automotive and motorsport: The company’s F1 roots give it hard-won credibility with the world’s most demanding simulation consumers. Formula 1 teams generate petabytes of aerodynamic data per season and have pioneered many techniques that PhysicsX is now generalizing for broader industry.
The LPM Thesis and Its Risks
PhysicsX’s fundamental bet—that a generalist Large Physics Model can achieve the kind of cross-domain generalization that large language models achieved for text—is bold. The comparison is imperfect. Physical systems are governed by well-understood equations, but the training data requirements for high-fidelity multiphysics simulation are enormous and expensive to generate. Simulation accuracy requirements in aerospace and semiconductor contexts leave little tolerance for the occasional hallucination that is acceptable in a chatbot.
The company’s competitors include entrenched simulation vendors like Ansys (now part of Synopsys), Siemens’ own Simcenter platform, and a growing field of AI-native challengers including Monolith AI and Neural Concept. What differentiates PhysicsX’s pitch is the generality argument: rather than accelerating one simulation type, it is building a unified platform analogous to how GPT changed the economics of language tasks.
With approximately 350 employees, expansion planned into the US and Singapore, and backing from strategically aligned corporates who can both validate the technology and channel revenue, PhysicsX has assembled the resources to test that thesis at scale. For industries where competitive advantage is measured in grams of weight saved and fractions of degrees Celsius, the ability to explore millions of design iterations rather than dozens could prove transformational—not just for the company’s investors, but for the pace of physical innovation itself.