Figure's Helix Threatens Model Layer
General Intuition
The core risk is disintermediation, because the best robot model may be the one bundled inside the robot, trained on that robot’s own mistakes, and sold as part of one monthly system price. Figure already runs Helix fully onboard, updates deployed robots over the air, and uses long running BMW deployments plus home data collection to improve control, which compresses the room for a separate model supplier to sit between hardware and customer value.
-
Figure is not just another customer for an upstream model. It owns the body, actuators, manufacturing, deployment, and policy stack. That means every warehouse shift creates closed loop data tied to the exact hand, wrist, battery, and motion system Helix controls, which is the strongest setup for improving dexterity quickly.
-
The Google DeepMind and Apptronik bloc points in the same direction. Gemini Robotics is already adapted to Apollo, runs in an on device form, and Apptronik is expanding robot training facilities. Even though this path is partnership based rather than fully owned, it still pushes intelligence closer to the robot and away from a standalone model layer.
-
The market is also rewarding full stack stories with capital. Figure reached a reported $39B valuation and about $1.9B in funding, versus Apptronik at roughly $5.3B and nearly $1B raised. Investors are funding companies that can capture robot revenue, software revenue, and deployment data in one loop, not just sell an abstract policy API.
Going forward, the winners are likely to be the groups that turn live deployments into a compounding learning loop, then spread those gains across bigger fleets. If humanoid adoption scales through factory, warehouse, and home rollouts, model providers will keep the most leverage only where robot makers still need an external brain instead of a tightly coupled onboard one.