Ex-Nvidia Robotics Exec Says the Field Is Stuck in Its 'Lego Era' — Not a ChatGPT Moment

Robotics is not on the verge of a ChatGPT-style breakthrough, according to Amulya Vishwanath, who spent seven years running product marketing for Nvidia's robotics platforms before moving to the venture side. In her view, the field is still in what she calls a 'Lego era': the fundamental parts — actuators, robotic hands, software stacks and safety standards — are being redesigned rather than assembled into finished systems. That assessment pushes back on a growing chorus of founders and investors who expect a general foundation model to suddenly make robots broadly capable.

The Lego comparison is more than a metaphor. Robotics lacks the equivalent of the common interfaces and interchangeable modules that turned computing into a plug-and-play industry. Each new robot arm or humanoid tends to ship with its own mechanics, its own control software and its own calibration process, which means progress on one platform rarely transfers cleanly to another. Until those pieces settle, Vishwanath suggests, scaling intelligence across hardware will stay slow and expensive.

That gap helps explain why so much robotics investment today flows into components rather than end-to-end products. Actuators — the motors and gears that move a joint — remain a bottleneck in cost, durability and supply. Robotic hands are still a research problem, with few designs that can match human dexterity without fragile or costly mechanisms. On the software side, simulation, reinforcement learning and teleoperation pipelines are improving, but there is no shared layer that lets a model trained in one environment deploy reliably in another.

Safety standards add another layer of friction. Unlike chatbots, robots share physical space with people, and rules for how they should behave around humans are still being written by standards bodies and regulators. Vishwanath's point is that you cannot bolt a powerful model onto hardware that has not been certified to operate safely. That regulatory and engineering work is unglamorous, but it gates every commercial deployment from warehouses to homes.

Her remarks land as humanoid robotics draws record attention and capital, with several well-funded startups promising general-purpose machines within a few years. The counterargument is that those timelines assume a foundation model can generalize across tasks the way large language models generalize across text. Vishwanath's experience inside Nvidia's robotics business — which supplies much of the simulation and compute stack the industry relies on — gives her a close view of how far the underlying tooling actually is.

What to watch next is whether component makers and software vendors converge on shared standards for actuators, hands and safety, or whether each robot company keeps building its own stack. If the Lego pieces do start clicking together, the foundation-model approach could accelerate quickly. If they do not, the ChatGPT moment for robotics will keep slipping, no matter how capable the models become.