EDA
System
IP
SolvNetPlus
Training & Education
AI is quickly emerging from its digital confines as something new: physical AI.
This evolving incarnation pushes beyond the realm of information — code, text, images, video — and enables machines to sense, decide, and act in the real world.
“Physical AI is AI with a body,” said Drew Henry, executive vice president of physical AI at Arm. “It’s where AI operates within a machine.”
Joined by executives from NVIDIA, Schaeffler, and Synopsys, Henry set the stage for a thought-provoking panel discussion at the 2026 Synopsys Executive Forum, which is part of Synopsys Converge.
Covering everything from AI ‘brains’ to robotic bodies to high-fidelity simulations, the panelists suggested the implications of physical AI are nothing less than profound. Embedding intelligence into robots, vehicles, and equipment promises to extend AI’s impact into factories, hospitals, city streets, and beyond.
But it won’t be easy. Bringing physical AI to life, the panelists agreed, demands a different kind of engineering. One that can bridge digital intelligence with physical systems at scale.
Architecting Physical AI SoCs with Standards‑Based IP for Real‑World Intelligence
Transforming industries
The economic potential of physical AI is enormous. While global GDP is roughly $100 trillion annually, the information and communications technology (ICT) sector accounts for only a modest share of total value added, even as it exerts an outsized influence on productivity and innovation across the economy.
To date, AI workloads have primarily transformed knowledge work. Physical AI, on the other hand, targets the much larger share of economic activity: moving people and goods, assembling products, delivering healthcare, and operating in the physical world.
“If you step into a typical factory today, it doesn’t look that different than it did 10 years ago, even though computing has exponentially increased,” observed Rev Lebaredian, vice president of physical AI simulation at NVIDIA. “We’re finally creating a bridge from computers to atoms of the physical world. All the growth we saw in the information world is about to happen in the physical world. Essentially, every industry will be transformed.”
Making that vision a reality will require a more holistic approach to engineering, blending silicon-to-systems design, software-defined capabilities, and agentic AI with deep simulation and digital twins.
Ravi Subramanian, chief product management officer at Synopsys, framed that shift in concrete engineering terms.
“So many industries rely on motion control — medical, industrial, automotive — and physical AI is intelligent motion control,” he said. “What’s changed is that robotics is shifting from ad-hoc tinkering to a true scientific discipline.”
Scaling specialized skills
Physical AI may shift productivity in surprising ways. For instance, highly specialized skills — surgical procedures, complex machinery setup, advanced simulation workflows — are often concentrated in a handful of experts, limiting scalability.
With physical AI, however, skills can be infused into systems that act as extensions of human expertise. Henry used the example of a surgeon who spends years developing the fine motor control and judgment required for a successful operation.
“We can bring that capability out to the rest of the world to have surgical procedures anywhere on the planet, not just in certain locations,” he said.
The same dynamic applies to engineering itself.
“If you go to any company that needs simulation, it’s only a few people who know how to use those tools,” said Lebaredian. “Agentic systems that understand the physical world can drive the simulation tools.”
Many more engineers — and even non-experts — can gain the benefits of high-end simulation without mastering every detail of the software, he added.
It’s a preview of AI-driven workflows where engineers of all stripes define goals and constraints, and AI agents operate complex tools on their behalf — a trend already evident in chip and system design.
Creating feedback loops
Robots aren’t new, of course. Industrial robots have been welding, painting, and handling materials in factories for decades. And smaller household robots have taken on narrow tasks like vacuuming. These machines, however, are specialists: tightly constrained, painstakingly programmed, and difficult to adapt.
“We had the technology to create the robot bodies,” said Lebaredian, “but we didn’t have the technology to create a robot brain that would make those robots useful.”
Physical AI will be those brains — not scripted for every motion and operating scenario, but trained using extensive simulations, the laws of physics, and corrective feedback.
“We can take the physical world and the laws of physics, all of the rules around us, and represent it inside the computer,” said Lebaredian. That simulated world becomes an effectively infinite data source to train physical AI systems, which can then be embodied in many different robotic platforms.
This is not a one-time training step, the panelists emphasized. Virtual models — called digital twins — of robots and other physical AI systems are trained in virtual models of factories, warehouses, or clinics before deployment, exercised across millions of scenarios to validate their behavior. Once deployed, real-world data flows back into the digital twins to refine both the AI model and the virtual environment.
This creates a continuous loop: simulate → train → deploy → observe → retrain.
For robotics component manufacturers like Schaeffler — both a global supplier and a large‑scale user of industrial robots across roughly 100 factories — that feedback loop is critical for translating operational data into more capable, adaptable robotic systems.
“Physical AI is going to be the next giant leap in automation,” said Dave Kehr, president of humanoid robotics at Schaeffler Group. “You just tell [a generalist robot], ‘I need you to go perform this operation.’”
Instead of months of training and setup, robots will be able to shift between tasks as conditions and demand change.
Designing systems, not just robots
Building a brain for a robot is only part of the challenge. Physical AI systems are complex combinations of sensors, actuators, processors, networks, and software operating in environments that are often dynamic and unpredictable. Developing and verifying such systems requires methodologies that look more like modern chip design — which integrates silicon, software, interconnects, and packaging — than traditional mechanical engineering.
A robot brain takes more effort to verify than its body, the panelists emphasized.
Extensive simulations and training will increasingly rely on digital twins that model all aspects of the physical robot, including digital, electrical, and mechanical. These simulations must extend beyond initial deployment so designers can verify and implement incremental changes throughout the unit’s lifetime as it operates across real-world environments.
But that represents a significant shift to the engineering process. Instead of working in silos — one team for chips, another for control software, another for mechanical systems — organizations need to align their workflows and focus on end-to-end, cross-disciplinary co-design.
“You really can’t do this in silos anymore,” said Sam Abuelsamid, vice president of market research at Telemetry and moderator of the panel. “It's such a complex set of interactions across the entire system that you need simulation tools, operational design, domain simulation, chip design, and the target applications — and fit all these pieces together.”
Emerging capabilities that combine electronics design tools with physics-based simulations — such as Synopsys Multiphysics Fusion™ solutions — are built for this kind of engineering. From chip design and verification to multiphysics simulation and digital twins, engineers can build shared virtual environments where AI agents, control software, and physical models interact before hardware exists.
As AI-in-engineering practices mature, agentic AI systems can help drive these workflows — configuring simulations, sweeping design spaces, and flagging edge-case behaviors — while human engineers focus on intent, constraints, and trade-offs.
Rethinking GDP, labor, and “3D” jobs
In addition to changing many human jobs, physical AI may upend longstanding economic truths. For instance, the capacity for manufacturing goods will no longer be limited by the size of the workforce.
“The GDP of a country correlates with population,” Lebaredian explained. “You [can] educate your workforce more, but ultimately your output is dependent on the number of laborers you have. If you can manufacture physical [robotic] laborers, then we're going to untether the GDP from the population.”
That doesn’t mean people become unnecessary, but it does shift human work toward higher-value roles instead of narrowly defined tasks.
“There’s going to be a lot of tumultuous chaos in the workforce as things change,” Subramanian said, noting the silver lining of eliminating so-called “3D” jobs that are dull, dirty, or dangerous. “The net benefit to humanity will be that the three Ds will be addressed, and it will open up the ability for humans to have better, safer lives.”
Realizing that vision will take time, engineering rigor, and close collaboration across industries. It will also require the integrated design flows, simulation capabilities, and AI-in-the-loop methodologies already reshaping chip design and systems engineering.
With physical AI, those digital threads won’t just connect software and silicon — they’ll extend into the physical world itself.
Continue Reading
Enabling the Next Generation of AI Infrastructure with Ethernet for Scale-Up Networking (ESUN)
Advancing UCIe Performance: Enabling 40G for Next-Generation Multi-Die Designs
