Every few weeks a new robot foundation model drops a jaw-dropping demo—a humanoid folding laundry, an arm making coffee, a policy that generalizes across tasks. The models are genuinely good now, yet almost none of them are reliable enough to deploy for real work. The gap between the demo and the deployment has become mostly a data problem rather than a model problem.
I spend my days on the data side of physical AI, and this is the single most misunderstood thing about where robotics is right now.
If the models are ready, what is missing?
Robotics found its GPT moment in the vision-language-action model (VLA)—a single network that takes in what the robot sees plus a language instruction and outputs actions. Physical Intelligence's π0, NVIDIA's GR00T, and a dozen others all follow this recipe. The architecture works.
The difference is that an LLM had the entire internet to learn from, while a VLA has no Common Crawl of robot actions to draw on. The scarce ingredient is examples of a body actually doing the task, more than compute or model design.
Why can't you scrape your way out?
Language and images are effectively free. You can download a trillion tokens of text and billions of captioned images, and that gets an LLM most of the way there.
Robot data is different because of one modality that doesn't exist on the internet: embodiment state—the joint angles, end-effector pose, gripper width, and contact forces of a body, synchronized frame-by-frame with what it saw and what it was asked to do. YouTube has plenty of video, but none of the action data that goes with it. You cannot infer, from a video of someone folding a shirt, the exact torques and trajectories their hands produced—and even if you could, human hands are not robot grippers.
That missing modality matters more than anything else here. It is why "just use internet video" never quite works, and why real, action-labeled demonstrations remain the expensive, irreplaceable input.
Is all robot data equal?
Robot data varies a lot in value, and it helps to think of it as a ladder, from cheap-and-scalable to accurate-and-expensive:
- Internet and egocentric video — huge, cheap, diverse. Great for general visual priors. But it's human, not robot, and it has no clean actions.
- UMI-style handheld grippers — put a robot-like gripper (with a wrist camera) in a human's hand. Now you get real gripper actions without a robot. Cheaper and portable, but gripper-only and still an embodiment gap.
- Teleoperation on a real robot — a human puppeteers the actual robot; you record the exact embodiment, the exact actions, multi-camera. The most accurate and the most expensive.
Each rung trades cost for fidelity. The cheap rungs help with pretraining, while the top rung is what makes a model actually work on a specific robot.
Why is post-training where the value is?
The LLM world already taught us this. You pretrain on the cheap, abundant internet, then you spend real money on the expensive, curated, human-generated data—instruction tuning and RLHF—to turn a raw predictor into something useful. That post-training layer is small in volume and enormous in value.
Robotics follows the same pattern. Pretrain on video and cheap data; post-train on high-quality teleoperation demonstrations on the target robot and tasks. The demonstrations are the robot equivalent of RLHF data: scarce, expensive, and the difference between an 80%-success demo and a 99%-reliable product.
The difference between 80 and 99 matters a lot. A model that succeeds 80% of the time makes a great demo but a poor product, because if you chain a hundred steps and your odds of a clean run round to zero. Closing that last gap is almost entirely a data-and-evaluation problem.
What is the defining constraint of the decade?
So the bottleneck in physical AI is no longer the model. It is the cost, quality, and scale of real-world demonstration data—and the ability to measure whether that data actually made the robot better.
Whoever can produce high-fidelity, well-annotated, on-the-right-embodiment demonstration data—cheaply and at scale—supplies the input every robot company is starving for. That's the layer I'm betting on, and it's the layer we're building at PrismaX.
The models will keep getting better, as they always do. What will decide who ships a working robot is a quieter question, which is who has the right data.