Open Weights Drive Robot Adoption

The real edge in robot foundation models is not just better policies, it is becoming the default starting point for everyone building on top. Physical Intelligence moved in that direction by releasing π0 code and weights, which lets outside labs, OEMs, and developers fine tune the model on their own hardware instead of waiting for access to a partner program or closed API. That makes adoption easier across a fragmented robot market where every arm, gripper, camera, and control stack is slightly different.

• DeepMind has taken a partnership driven path, tying Gemini Robotics to Apptronik and other selected OEM relationships. That can produce strong flagship deployments, but it channels outside adoption through a narrower set of approved integrations instead of letting the broader market modify and run the model freely.
  3 cnbc 4 sacra
• Meta has published robotics research such as V-JEPA 2, but its work is still mostly a research release rather than a productized third party ecosystem with deployment tooling, support, and broad commercial onboarding. That leaves room for a company built around open robot model adoption to become the practical developer layer.
  5 fb 6 sacra
• Covariant and Skild show the other path. They aim to be cross platform robot brains, but their advantage comes from proprietary deployment loops, data, and software sold to OEMs and enterprises. Physical Intelligence is betting that open weights create faster ecosystem spread first, then value accrues through fine tuning, tooling, and downstream services.
  2 sacra 7 sacra 8 sacra 9 sacra

This is heading toward the same split seen in language models. Closed labs will keep producing frontier systems, but the widest developer footprint can belong to the model that is easiest to download, adapt, and ship on real machines. If Physical Intelligence keeps improving open models while the robot base diversifies, it can become the shared base layer others build on.