Vertical integration captures physical AI value

This is the core platform risk for a horizontal physical AI company, because buyers often want one vendor that owns the robot, the autonomy stack, the sensor package, and the live deployment loop. Figure is training Helix on its own humanoid body and logistics tasks, while Anduril and Forterra package autonomy inside complete mission systems, which lets them sell outcomes, collect operating data directly, and keep more of the economics at the system level.

• Figure is not just selling a model. It sells a humanoid robot whose cameras, actuators, and task data all feed Helix. That means every warehouse deployment improves the same tightly coupled system, instead of splitting value between a robot maker and a separate intelligence supplier.
  1 sacra 2 figure
• Anduril follows the same pattern in defense. Lattice is embedded across its own air, sea, and land systems, so autonomy is purchased as part of a working operational stack, not as a standalone software layer that a customer has to stitch into hardware themselves.
  3 anduril 4 sacra
• Forterra shows how this plays out in industrial autonomy. Its AutoDrive is bundled into production vehicle programs with Oshkosh, Kalmar, and BAE, alongside networking and integration modules, which makes the buying decision about a finished autonomous vehicle system rather than a separate world model.
  5 forterra 6 forterra 7 forterra 8 forterra

The market is heading toward full stack physical AI vendors that win by proving reliability in specific environments and then compounding deployment data. For AMI to capture major value, it will need to become the control layer that integrated system providers cannot easily replace, or move closer to the deployment edge where data, workflow ownership, and budget already sit.