The robotics industry has all the ingredients for a revolution and none of the recipe. That is the thesis Amulya Vishwanath is staking her new venture career on, after seven years leading product marketing for Nvidia's robotics platforms gave her a front-row seat to why the field keeps promising more than it delivers.
Vishwanath's diagnosis is bracingly specific: robotics is in its "Lego era," she says, with actuators, robotic hands, software stacks, and safety standards all improving in isolation but refusing to cohere into systems that work reliably in the real world. The comparison to large language models is instructive. ChatGPT did not invent transformers or attention mechanisms; it assembled existing components at the right scale, with the right interface, at the right cultural moment. Robotics, Vishwanath argues, is waiting for its own moment of integration — but the physics of moving through space impose constraints that text generation never faced.
Why hardware is harder
The software-first AI boom trained investors to expect exponential improvement curves. Robotics offers something more stubborn. A language model that hallucinates can be corrected with a follow-up prompt; a warehouse robot that misjudges a shelf edge destroys inventory and trust. Safety standards, regulatory approval, and the sheer expense of physical iteration mean that robotics companies burn capital at rates that would make most SaaS founders weep. Vishwanath's time at Nvidia taught her that the simulation-to-reality gap — the chasm between a robot performing flawlessly in software and performing adequately in a factory — remains the industry's defining challenge.
Nvidia itself is betting heavily on closing that gap. The company's "physical AI" push, announced with characteristic Jensen Huang showmanship, promises end-to-end platforms for robotaxis, humanoid robots, and industrial automation. The pitch is that Nvidia's simulation tools, combined with its dominance in AI accelerators, can finally make the Lego blocks snap together. Vishwanath is more cautious. She believes the enabling technologies are ready, but that the winning companies will be those that solve narrow, high-value problems rather than chase the humanoid dream.
The investor's edge
Vishwanath's new fund — she has not disclosed the name or size — will focus on what she calls "physical AI," a term she insists is broader than robotics. Autonomous vehicles, drone logistics, and even smart infrastructure fall under the umbrella. Her edge, she claims, is knowing which component suppliers are real and which are vaporware, a judgment call that requires the kind of vendor relationships she built at Nvidia.
The timing is deliberate. Venture capital has grown wary of robotics after a decade of underwhelming exits, but Vishwanath sees that skepticism as an opportunity. "Everyone wants to fund the next foundation model," she told TechCrunch. "Almost no one wants to fund the company that makes the gripper that actually works." Her bet is that the unsexy middle of the stack — the actuators, the safety certifications, the integration software — is where the real value will accrue.
Our take
Vishwanath is right that robotics has been overpromised and underdelivered for years, and her insider perspective gives her a credibility that most robotics VCs lack. But the Lego metaphor cuts both ways: Lego sets come with instructions, and robotics still does not have a clear blueprint for which pieces matter most. If Nvidia's physical AI push succeeds, it may commoditize the very integration layer Vishwanath is betting on. Her fund's success will depend on whether she can identify the few components that remain defensible even as the giants move in. It is a narrow path, but someone with her map might actually find it.
