When ChatGPT launched in late 2022, it created a moment of sudden, visible capability: a general-purpose model that could write, reason and converse well enough to convince millions of people, almost overnight, that a technology had crossed a threshold. Some investors and founders in robotics have been predicting a similar inflection point for physical machines, where one large general-purpose model trained across many robot types and tasks would make today's narrow, hand-built robotics systems obsolete. A former Nvidia executive who spent years inside that debate says the industry is not close to that moment yet.

What is new

Amulya Vishwanath spent seven years leading product marketing for Nvidia's robotics platforms before founding Techable VC, an early-stage firm investing in what she and others call 'physical AI' — robotics, materials science and life sciences. Speaking on TechCrunch's Equity podcast with Dominic-Madori Davis, Vishwanath described robotics as being in a 'Lego era,' meaning the basic components the field depends on are still being built or rebuilt: actuators (the motors and mechanisms that let a robot move), robotic hands, the software stacks that control them, and the safety standards that govern how they operate around people.

Her view runs directly counter to a position laid out earlier this year by Pim de Witte, CEO of General Intuition, who told Equity that a single general foundation model would soon make much of today's specialized robotics work redundant — essentially robotics' own ChatGPT moment.

Background: what a 'foundation model moment' means

A foundation model is a large model trained on broad data that can be adapted to many downstream tasks rather than being built from scratch for each one. In language, this is what let a single model like ChatGPT handle writing, coding, summarizing and conversation without separate systems for each. The hope for robotics is similar: instead of engineering a bespoke control system for every robot arm, warehouse picker or humanoid, a single large model trained on enough demonstrations and sensor data could generalize across hardware and tasks, the way language models generalize across topics and writing styles.

That analogy has limits, which is effectively the heart of Vishwanath's argument. Language lives in text, a format that is already digital, abundant and relatively uniform. Robots operate in the physical world, where every actuator, gripper and sensor is a piece of hardware that has to be designed, manufactured, tested and made safe before any software model can act through it. A model can be as capable as it likes, but it still needs reliable hands, joints and safety systems to actually do anything.

Why it matters

Vishwanath's framing matters for how investors, engineers and policymakers prioritize their attention and money right now. If one believes robotics is about to have its ChatGPT moment, the sensible move is to bet heavily on general-purpose robot foundation models and treat specialized hardware and software companies as likely to be swept aside. If instead the field is still in a 'Lego era,' as Vishwanath argues, then the more valuable work in the near term is in the foundational pieces: better actuators, better robotic hands, more robust control software and clearer safety standards. In that view, a general model would have little to work with even if it existed, because the physical building blocks underneath it are not yet solid.

This distinction also shapes where early-stage capital flows. As the founder of a venture firm investing across robotics, materials science and life sciences, Vishwanath's own portfolio choices reportedly follow from this thesis — favoring the fundamental components of physical AI systems over bets on an imminent general-purpose robotics breakthrough.

  • Amulya Vishwanath spent seven years in product marketing for Nvidia's robotics platforms before founding Techable VC.
  • She describes the robotics industry's current stage as a 'Lego era,' with core building blocks like actuators, robotic hands, software and safety standards still being developed.
  • This contrasts with a view, voiced earlier this year by General Intuition CEO Pim de Witte, that a single general foundation model could soon make specialized robotics work redundant.
  • Techable VC, where Vishwanath is founder and managing partner, invests in early-stage physical AI across robotics, materials science and life sciences.
  • The discussion took place on TechCrunch's Equity podcast, hosted by Dominic-Madori Davis.

Limits and open questions

The source material is a podcast conversation rather than a research paper, so there are no benchmarks, datasets or technical results to weigh here — this is a debate about industry timing and strategy, not a specific technical claim. It is also not clear from the discussion how long Vishwanath thinks the 'Lego era' will last, what specific milestones would signal that robotics is ready for a more general foundation-model approach, or how she evaluates the robotics foundation models that already exist in research labs and startups today. Readers should treat this as one experienced insider's perspective on industry readiness, set against a differing view from another founder in the same conversation space, rather than as a settled technical conclusion.

Sources