Open Is How Robotics Wins

The value in robotics will not belong to whoever owns the biggest closed model. The model is not the moat. Your data and your know-how are.

The moat is three things:

  • Your proprietary data. The data from your machines, in your environment, doing your task.
  • Your know-how. Knowing how to use that data, and how to break the problem down correctly.
  • Your creativity and grit. No matter how good AI is, building a product and a business is hard. Make it happen.

That’s your IP, and it should stay yours. Everything else should be open. If you want a step-function change in robotics, the foundation has to be open: open models, open weights, open coding systems.

We want to see one billion new Physical AI deployments over the next ten years, for the good of the world. Our most profound challenges are in the physical world, from how we grow food, to how we keep the lights on. The solutions will happen in the physical world, and those solutions will require both physical machines, like robots, and also increasingly creative and useful applications of AI.

That doesn’t happen if robotics belongs to a small handful of companies. Time and time again we have seen in technology that explosive innovation happens in open ecosystems, not in closed ones. For Physical AI to truly take off it will need democratization, not consolidation. It will require the wisdom of crowds, not the blessing of a few companies that want to own the future exclusively.

What does it take for transitions like the one in front of us? It takes open source, of course. It takes 10,000 startups, and it takes millions of developers, distributed across a vast expanse of the physical economy and the physical world.

Open source is not a nice-to-have. It is not going away just because one of its biggest proponents just walked off the stage. Open source, in the transition that is coming, is essential. Why do I think this?

Start with a fact that people in this industry don’t like to say out loud: the public robotics data everyone trains on is small and shared (Open X-Embodiment, DROID), so models trained on it converge. The difference comes from data collected on your own machines, and from what you do with the models.

Two objections deserve an answer. First, “scale wins”: whoever has the most data trains the best model. But the data that matters is spread across millions of deployments, in factories, farms, and warehouses no model company can see. That argues for open infrastructure owned by the people who have the data. Second, “open weights already exist.” Weights are a start, but the tooling, data pipelines, and deployment layer have to be open too, or the weights do little for the people who need them.

The data is already out there

For video, there is plenty of data. There is an enormous amount of video of humans doing things, and an enormous amount of video made specifically to teach you how to do something. There’s also a lot of research on taking that kind of data and turning it into training data for robots. Early results are promising. Research groups are already training robot policies on video of people doing everyday tasks.

Somebody has to make it happen

The people who were supposed to champion open models have stumbled. Jensen Huang was a huge supporter of open models, and now he’s out at Nvidia. OpenAI keeps its frontier models closed.

I don’t hear enough voices saying this, so I will: let’s make open source the standard in Physical AI.

I’m not standing on the sidelines. I’ve bet my own money on Viam. Its robot runtime is open source, most of the modules I write are open and available to all, and we charge for cloud services. That’s the same model as MongoDB and MongoDB Atlas, which I built. Business and open source are not opposed.

Fundamentally, I’m not looking for a world in which those millions of developers will all work for the same ten people who have pulled all the value behind a paywall inside a tiny handful of companies. No one should want that. Let’s instead create a world where open source Physical AI development will be a part of every business from a small machine shop to an enormous e-commerce company, from a global shipping behemoth to a local commercial laundry and every conceivable thing in between.