Amazon absorbs Covariant research team
Covariant
This deal let Amazon internalize Covariant’s hardest to replicate asset, the people and models behind robotic picking, without shutting down the product business that was already deployed in customer warehouses. Amazon got a non exclusive license to the core model technology and hired founders Pieter Abbeel, Peter Chen, and Rocky Duan, while Covariant stayed active as an independent company with Ted Stinson as CEO, preserving existing commercial relationships and field deployments.
-
The product is not a general robotics moonshot. Covariant sells AI software and picking systems for warehouse stations where a robot arm identifies items in bins, grasps them, and learns from each pick. That makes Amazon a natural home for the research team because its fulfillment network generates huge volumes of exactly this kind of training data.
-
The structure looks closer to a carve out than an acquisition. Amazon described the deal as a non exclusive license plus talent transfer, and outside reporting said Covariant kept operating under Ted Stinson and Tianhao Zhang. That avoids the full integration burden of buying the company while still moving the key technical brain into Amazon.
-
There is precedent in warehouse automation for Amazon pulling a critical capability in house. After buying Kiva, Amazon kept the robots for its own network, and later warehouse robot vendors like Locus built around easier drop in deployments for everyone else. Covariant’s remaining business now has to prove it can keep selling broadly without its original founders inside the company.
Going forward, this pushes physical AI toward a split market. The largest operators, like Amazon, will absorb frontier teams to build proprietary robotics stacks around their own warehouse data, while independent vendors will need to win by shipping faster, integrating into many customer workflows, and turning each deployment into reusable model improvement.