Robotics isn’t about to have its ChatGPT moment. It’s still in its “Lego era,” according to Amulya Vishwanath, who spent seven years leading product marketing for Nvidia’s robotics platforms. TechCrunch AI reports that Vishwanath, now founder and managing partner of early-stage firm Techable VC, thinks the basic building blocks of robotics are still being rebuilt. Those blocks include actuators, robotic hands, software and safety standards.
That puts her at odds with one of the loudest arguments in the field right now.
Two camps, one big bet
The physical AI debate splits roughly into two views:
- The foundation model camp. General Intuition CEO Pim de Witte told TechCrunch’s Equity podcast earlier this year that one general foundation model would make much of today’s specialized robotics work redundant. In his view, that model would be robotics’ ChatGPT moment.
- The building blocks camp. Vishwanath argues the hardware stack, the safety standards and the integration work aren’t mature yet. A brilliant model can’t do much with a hand that can’t grip reliably or a robot nobody has certified as safe to work next to people.
What stands out here is that both views can be partly right on different timelines. Language models took off because the “hardware” was already in place: billions of screens and keyboards. Robots don’t have that installed base. Every deployment still needs physical parts, a supply chain and a customer willing to put a machine on their floor.
Physical AI is bigger than humanoids
Vishwanath’s most useful idea might be her definition. Physical AI, she says, can be “a molecule, a material, a medicine,” not just a machine on a factory line. Techable VC invests across robotics, materials science and life sciences.
This matters because most of the hype and capital has gone to humanoids. Vishwanath takes a dig at that too, saying humanoids should be good for more than dancing. If you count AI-designed materials and drugs as physical AI, the market looks much wider, and some of it may reach revenue faster than general-purpose robots.
She also talks about “conscious automation,” which means pointing technology at real, painful problems instead of building demos. Her examples:
- Family-run factories that can’t find workers
- Surgeons dealing with burnout
Those are labor shortage problems with buyers who already feel the pain. That’s a much easier sale than a general-purpose robot looking for a job to do.
The Nvidia network as a moat
Vishwanath wants founders to see their VC as their moat. Her years at Nvidia gave her a network she now uses for due diligence and customer introductions. She’s also one of the first women executives to spin out of Nvidia and launch her own early-stage firm.
This fits a wider pattern. Nvidia sits at the center of the physical AI ecosystem through its simulation and robotics platforms. That gives its alumni an unusual view of who is building what and which startups actually ship. Expect more Nvidia veterans to turn that knowledge into funds and startups.
The US vs. China question
Vishwanath also covers where the U.S. and China each have the edge. The full detail is in the episode, but the framing makes sense. China has the manufacturing depth and the hardware supply chain. The U.S. leads on frontier models and capital. If she’s right that we’re in the Lego era, hardware ecosystems will matter a lot over the next few years.
What this means for the next 1-3 years
A few things to watch, and to act on:
- Bet on narrow wins first. Startups that solve one expensive labor problem in one industry will likely beat generalist robots to real revenue.
- Watch the boring layers. Actuators, grippers and safety certification are where bottlenecks sit. Companies fixing those could become the picks and shovels of physical AI.
- Widen your lens. If you track physical AI only through humanoid demos, you’ll miss materials and life sciences, where AI is already changing how products get discovered.
- Don’t write off the foundation model camp. If a general robotics model does arrive, specialized software could lose value fast. Smart builders should design products that can plug in better models as they appear.
My take: Vishwanath’s view is the more grounded one for anyone writing checks or running a factory today. The ChatGPT moment may still come, but it’ll land on top of the Lego pieces being built now. The full conversation is available on TechCrunch’s Equity podcast.