While everyone's waiting for the iPhone of robotics, the industry is still arguing about what a screw thread should look like.
The Summary
- Amulya Vishwanath, seven-year Nvidia robotics veteran turned VC, says we're in robotics' "Lego era" — rebuilding basic components like actuators, hands, software, and safety standards from scratch
- This directly contradicts the foundation model hype that assumes one big AI brain will unlock general-purpose robots
- The real bottleneck isn't intelligence, it's the physical infrastructure layer no one wants to fund
The Signal
Vishwanath spent seven years at the center of Nvidia's robotics push, watching billions flow into AI foundation models while the physical layer languished. Her diagnosis: we're obsessed with the brain when we haven't built a working body.
The "Lego era" framing matters because Lego works precisely because the blocks are standardized. Right now, robotics has no standard actuators, no agreed-upon gripper designs, no common software interfaces. Every company is manufacturing its own bricks in incompatible sizes. You can't snap-together a warehouse robot from off-the-shelf parts the way you build a web app from APIs.
"We're rebuilding actuators, robotic hands, software, and safety standards from scratch."
This isn't just an engineering problem. It's an investment structure problem. VCs fund software because it scales infinitely at near-zero marginal cost. Hardware doesn't. A better actuator requires materials science, manufacturing partnerships, and years of iteration. Investors want ChatGPT moments, not screw thread standardization committees.
Vishwanath's view runs counter to the prevailing narrative that robotics is one foundation model away from explosive growth. That narrative assumes the hard part is making robots smart enough to generalize across tasks. Her bet is the opposite: we already have enough AI. What we lack is reliable, affordable, interoperable physical components that don't break after 10,000 cycles.
The irony is that Nvidia itself enabled the foundation model boom by standardizing compute infrastructure. Everyone trains on the same GPUs, uses the same CUDA libraries, speaks the same language. Robotics has no equivalent. Physical AI, as Vishwanath frames it, is bigger than just robots. It includes anything AI-driven that touches atoms: drones, autonomous vehicles, manufacturing equipment, agricultural machinery.
The Implication
If Vishwanath is right, the next wave of robotics winners won't be the companies building humanoid demos for YouTube. They'll be the ones solving unglamorous infrastructure problems: standardized joints, open-source control software, safety certification frameworks that work across platforms. That's where the leverage is.
For founders, this means the opportunity isn't in building another general-purpose robot. It's in building the components and standards that let everyone else build robots faster. Think arms-dealer strategy, not hero strategy.