Everyone in robotics is waiting for the day one giant model makes the whole field click, the way a chatbot did for text in late 2022. Amulya Vishwanath, who spent seven years inside Nvidia's robotics business, thinks the wait may be misplaced, because the parts still need to be built.
Vishwanath led product marketing for Nvidia's robotics platforms. She has since left to found Techable VC, where she is managing partner and writes first checks. In a recent podcast conversation, she laid out how she sees the field and where she is placing bets.
The 'Lego era' versus the 'ChatGPT moment'
Her description of the present is a modest one. Robotics, she says, is in its "Lego era." The basic building blocks, including actuators, robotic hands, software and safety standards, are still being rebuilt.
That puts her at odds with a popular and more thrilling story. In that version, a general foundation model for robots arrives and the field has its own ChatGPT moment. General Intuition CEO Pim de Witte has argued along these lines. He said earlier this year that a single general foundation model would make much of today's specialized robotics work redundant, and that it would be the moment in question.
The two views are not quite opposites, but they disagree about where the hard work is. One says the intelligence layer will swallow the specialists. The other says the specialists are busy making the bricks that any intelligence layer will have to stand on. Nobody in this conversation is shown settling the argument, which is fair, since neither camp has shipped the evidence yet.
Physical AI is not just robots
Vishwanath defines physical AI broadly. In her words it can be "a molecule, a material, a medicine," or a machine on a factory line. That is why Techable VC, an early-stage firm, invests across robotics, materials science and life sciences instead of keeping to humanoids and robot arms.
It is a convenient definition for an investor, since it widens the pool of startups. It is also a coherent one. If AI is going to leave the screen, the interesting question is what it touches when it does, and the answer is not always a machine with limbs.
She also talks about what she calls "conscious automation." The summary of the episode does not spell out her definition, so it would be unwise to guess at it here.
Start with the problem, not the robot
On what makes a good opportunity, she keeps returning to the size of the problem. Her examples run from family-run factories that cannot find workers to surgeons facing burnout. Those are not demos looking for a use. They are shortages and strains that already exist, and the technology is hired to relieve them.
The same instinct shows up in her comments on humanoids, which she says should be good for more than dancing. Anyone who has watched a robot video go viral will understand the jab. A backflip is a fine way to attract attention and a poor way to fill a factory shift.
The investor as moat
Vishwanath also has a view on what a venture firm owes its founders. She wants them to see their VC as their moat. Her years at Nvidia, she says, built the network she now relies on for due diligence and for customer introductions.
This is a sensible pitch from someone whose résumé is the product. It is also a useful reminder that in hardware-heavy fields, a check is the easy part. Access to people who can say whether a component will work, or who might buy the finished thing, is scarcer.
She is described as one of the first women executives to spin out of Nvidia and start her own early-stage firm.
The U.S. and China
The conversation also covers where she thinks the U.S. and China each have an edge. The summary available does not say what she concluded, so there is no verdict to report. What it does indicate is that she treats physical AI as a competitive field with national dimensions, not only a technology trend.
What to take from it
The useful signal is small and fairly plain. A person who spent seven years close to one of the largest robotics platforms is betting that the near-term value sits in the unglamorous layers: hands, actuators, software, safety rules, and the real customers with real labor shortages. If a single model does arrive and makes the specialists redundant, she will have been wrong in an expensive way. If it does not, the Lego bricks will still be needed, and someone will have to sell them.
Either way, the industry's big moment may turn out to be a lot of small ones that were never recorded as one.