This also extends to physical AI models. Deployments on owned hardware are the moat.

The full article text is not available in MoonHub yet. You can read it at the original source.