Why Physical AI Isn’t Just About Robots — And Why It’s Way Bigger Than You Think
Amulya Vishwanath spent seven years at Nvidia pushing robotics platforms before jumping into venture capital with Techable VC, and he’s got a take that’s totally reshaping how investors see the space. While everyone’s chasing that so-called “ChatGPT moment” for robots, Vishwanath says we’re honestly still stuck in robotics’ “Lego era” — the foundational pieces like actuators, robotic hands, software stacks, and safety standards are all still being rebuilt from scratch. No joke, the idea that we’ll just bolt some What is AI smarts onto a chassis and call it done? That’s wishful thinking at this point. You know, it’s kinda wild how people keep ignoring the plumbing.
The real opportunity here, according to Vishwanath, isn’t in shipping the sexiest humanoid or the flashiest warehouse bot — it’s in the boring infra beneath it all. Think about it: you’ve got AI Models cranking out tokens left and right for chatbots, but the same energy going into reasoning about physical action in the real world is pretty low. When you start threading that through AI Tokens that actually represent physical decisions rather than just text predictions, suddenly the whole game changes. Let’s be real, man — most VCs are still betting on the wrong horses.
The Equity podcast episode dropped these insights from Vishwanath recently, and the takeaway is pretty clear-cut for anyone watching the space:
- Robotics won’t flip overnight just because someone slaps a foundation model on top — the hardware and safety layers are way more of a grind than people admit.
- The biggest upside lives in component-level plays (actuators, grippers, sim-to-real tools), not in the end-product robots everybody’s covering.
- Nvidia veterans see the gap between digital AI speed and physical AI reality first-hand; that perspective should reshape how you size these bets.