OpenAI Dissolves Robotics Unit; Agile Robots Valued at $8.5B

OpenAI has disbanded its standalone robotics team and reassigned the researchers to general model work. On the same day, Agile Robots closed a $500 million Series F led by SoftBank at an $8.5 billion post-money valuation, with total funding reaching $2 billion. The pairing is the day’s clearest signal: a leading foundation-model lab is narrowing direct hardware exposure while China-linked physical AI assets are still being priced upward.

Capital is moving beyond whole machines. Xilect, co-founded by MetaX and UBTech, raised RMB 500 million in an angel round for an embodied-AI system-on-chip plus software stack; an MIT-linked team backed by Lei Jun raised hundreds of millions of yuan for camera-based tactile sensing. Data from these deals show the funding frontier shifting toward chips and sensors, where unit economics and supply-chain position are easier to underwrite than in humanoid assembly.

Commercial evidence remains thin but is becoming quantifiable. Six AGIBOT G2 wheeled humanoids completed 1,730 tasks in a 24-hour live convenience-store shift in Shanghai, with four units passing final assessment, according to the company. Honor registered a robotics subsidiary with RMB 100 million in capital. Such disclosures set a higher bar for pilots: continuous operating hours and task counts, not demo footage, will determine repeat orders.

Technology attention is also shifting from single-body autonomy to shared perception and deployment safety. Astribot’s CoMind lets robots share camera views so one can guide another through obstructed chores; Agility’s safety chief framed “cooperatively safe” deployment for mixed human-robot floors. These are preconditions for warehouse and service economics, because multi-robot coordination and liability rules decide how many units a site can justify.

The analytical conclusion is narrower than “exit” or “entry.” OpenAI’s move suggests foundation-model labs may enter embodied AI through model licensing and partnerships rather than owned robotics teams. Two observation points follow: whether other labs follow with partner-led deployment in the next two quarters, and whether China’s physical AI valuation anchor spreads from whole machines to chips and tactile sensing, with paid repeat deployments as the test.