Servoservobrief.com, Sunday, November 15, 2026
Two bets on where robot intelligence lives: OpenAI says general models can run robots, while China funds domestic silicon to run them on.
-
OpenAI disbands its standalone robotics team, betting general models can run robots
OpenAI confirmed it dissolved its dedicated robotics team and moved researchers into its main model group, arguing its general model GPT-6 Astra can drive robots without a specialized robot brain. In an NVIDIA study, an Astra agent hit 74.8% on a simulated manipulation benchmark versus 51.0% for the best purpose-built robotics models. That gap is simulation-only, but it signals a shift away from training dedicated robot models toward adapting frontier LLMs.
Read more on OfficeChai, Pasquale Pillitteri (via Fortune) -
UBTech and MetaX chip venture Xilect raises 500M yuan to build a domestic robot brain chip
Xilect, a venture jointly owned by Chinese GPU designer MetaX and humanoid maker UBTech, closed a 500M yuan (about $75M) angel round co-led by GGV Capital and Matrix Partners China to build an edge compute chip for robots, benchmarked against NVIDIA's Thor. First chip is slated to tape out in late 2027 with mass production in 2028. It is a bid to cut China's humanoid builders off from reliance on NVIDIA silicon, though the product is still two years from volume.
Read more on MLQ.ai -
Foundation builds humanoids for the Pentagon instead of factories
San Francisco startup Foundation is developing humanoid robots aimed at US military use, breaking from rivals chasing warehouse and household work. No funding figure or deployment numbers disclosed, and the robots are not shown in autonomous field use. The move marks defense as an emerging buyer for humanoids, a market where demands and procurement differ sharply from commercial pilots.
Read more on The San Francisco Standard -
MOVIN raises $7.5M Series A for motion-capture data
MOVIN, a 2023-founded Korean startup, raised a $7.5M Series A led by IMM Investment, pushing total funding past $10M. The company works on motion data, the kind of human-movement recordings used to train robots. The round is small, but it adds to a wave of money chasing training-data pipelines for embodied AI.
Read more on Funding wire (Instagram)