Vertical players capture AMI's data flywheel

This dynamic decides who captures the data flywheel, not just who builds the smartest model. In physical AI, the winner usually improves by running robots, vehicles, or sensors in the field, collecting edge cases, and pushing software updates back into the same system. A horizontal layer like AMI can supply core intelligence, but integrated players like Figure, Anduril, and Forterra also control the machines, customer workflow, and feedback loop where performance compounds fastest.

• Figure is the clearest logistics example. It is building the robot body, the Helix AI stack, and the warehouse deployment together, so every pick, carry, and failure case becomes proprietary training data tied to its own hardware and operating environment, not to a neutral intelligence layer.
  1 sacra 4 sacra
• Anduril shows why this matters commercially. It self funds products, packages software inside hardware the government can actually buy, and reached an estimated $2.1B of revenue in 2025. That model lets it capture system level economics instead of selling a narrow software component into someone else's program.
  2 sacra 3 sacra
• Forterra is the middle case. It is not trying to be a general prime like Anduril, but it still sells one autonomy system across military and commercial vehicles, including yard trucks and logistics equipment. That kind of stack is harder for a horizontal model company to displace because the autonomy is already embedded in the vehicle workflow.
  3 sacra 5 sacra

The market is moving toward companies that own more of the loop, from sensing to action to retraining. For AMI to keep strategic leverage, the path is to become the intelligence core inside important vertical systems, or to climb further down the stack until its models are inseparable from deployment and data collection.