Simulation-heavy domains
Autonomous vehicle development is one of the clearest examples of synthetic data in practice. Companies developing self-driving systems run millions of simulated driving miles to expose their models to dangerous or rare scenarios, like sudden obstacles, extreme weather, or sensor failures, that would be unsafe, expensive, or simply impractical to reproduce that many times on real roads. Simulated environments also let engineers vary conditions systematically, testing the same scenario under different lighting or traffic densities in a way real-world testing can't easily replicate.
Robotics faces a similar challenge: physical trial and error is slow and can damage hardware, so robots are often trained or pretested in simulated physical environments before any real-world deployment. This lets a system encounter a wide range of situations quickly without the cost or risk of running them all on physical equipment.