Siemens says its design AI is 1,000x faster, and explicitly off-limits for the parts where failure kills

Siemens' design-simulation AI, Simcenter PhysicsAI, runs up to 1,000 times faster than a traditional physics solver. The Siemens executive who runs the product says that is exactly why it is not the last word on a part whose failure could kill.

"PhysicsAI is a surrogate model that learns from historical simulation data and predicts how a new design will behave in seconds," said Sam Mahalingam, who leads the PhysicsAI business at Siemens Digital Industries Software, on the sidelines of Realize LIVE Asia-Pacific in Bengaluru. Asked whether the tool is suitable for safety-critical applications, his answer was direct: "No, it is not."

Siemens positions PhysicsAI as a filter in front of full physics simulation. Engineers can explore thousands of design variations in seconds, narrow the field to two or three finalists, and then hand those finalists to a full physics-based solver before any sign-off. In Siemens' own case studies, PhysicsAI's predictions vary by 1% to 3% from a physics-based solver when sufficient historical data is available. Close enough to filter, not close enough to certify.

The Continental airbag case Siemens has showcased sits inside that exploration stage, not a safety certification lane. The boundary Mahalingam drew is his own product's framing, not an industry-wide consensus, and no independent regulator, certification body, or customer engineer is yet on the record.