OEM Proprietary AI Shrinks Third-Party Market

This risk is really about who owns the learning loop. A horizontal model like Skild gets stronger when many robot makers send back data from many real jobs, but a vertical player like Tesla, Figure, or Apptronik can keep that data, model tuning, and deployment inside its own stack. If enough OEMs do that, the biggest and best training pipelines sit behind closed walls, and the market left for third party robot brains becomes smaller and lower quality.

• Skild’s product assumes hardware makers will want to outsource intelligence. Its workflow is to map a robot’s joints, sensors, and cameras into Skild Cloud, then let customers call high level behaviors through APIs or natural language. That model works best when OEMs see AI as a shared software layer, not a core proprietary asset.
  1 sacra 2 sacra
• The vertical alternative is becoming more concrete. Figure says Helix controls perception, movement, and reasoning on board in real time, and is scaling from household data to logistics deployments. Tesla says Optimus is controlled by the same AI system behind its autonomy work. Apptronik is also funding a full stack Apollo push tied to its own humanoid platform.
  3 figure 4 figure 5 tesla 6 apptronik
• That changes data economics. A horizontal platform wins by pooling failures and successes across many robot types, but a vertical OEM can capture the highest value edge cases from its own fleet and improve only its own robot. The more important the data flywheel becomes, the stronger the incentive to keep AI in house.
  1 sacra 2 sacra 7 sacra

The market is likely to split in two. Large OEMs with enough capital, fleet scale, and deployment volume will build proprietary brains around their own machines, while smaller manufacturers and enterprises that cannot afford years of model training will buy outside intelligence. For Skild, the path forward is to become the default AI layer for everyone who does not have Tesla scale.