In a move that’s half “conscious uncoupling” and half “hold my circuit board”, Figure AI CEO Brett Adcock announced this week that his humanoid robotics firm is parting ways with OpenAI more swiftly than a robotic vacuum avoiding a staircase. The reason? A self-proclaimed “major breakthrough” in fully in-house AI – and a belief that outsourcing robot intelligence is as logical as outsourcing one’s skeleton.
The Split That Shook the Robotics World
Last year’s OpenAI-Figure partnership was the technology equivalent of a power couple: the ChatGPT virtuoso meets the humanoid pioneer. But like many high-profile partnerships, it concluded with a tweet. Adcock’s X post revealed Figure’s new philosophy: “We can’t outsource AI any more than we’d outsource actuators or batteries”. In other words: “It’s not you, OpenAI – it’s your lack of robot-specific focus.”
Industry sources suggest two catalysts:
- The DeepSeek R1 Factor: Speculation suggests that open-source AI models (like China’s DeepSeek-R1) demonstrated that OpenAI’s LLMs aren’t essential for teaching robots practical skills.
- OpenAI’s Robot Aspirations: With Sam Altman’s team filing trademarks for “user-programmable humanoids”, Figure likely concluded that competing with their investor was unwise.
The Secret Sauce: Vertical Integration or Vertical Ambition?
Adcock’s announcement reveals a fundamental robotics truth, as reported by The Decoder: “LLMs are the smallest piece now. The real challenge is high-rate control AI”. Put simply: Teaching a robot to engage in conversation is straightforward; enabling it to navigate a BMW factory safely is the true challenge in embodied AI.
[Continued translation follows the same style and principles for the remaining sections…]