Nvidia veteran says robotics is stuck in its 'Lego era'

Amulya Vishwanath spent seven years leading product marketing for Nvidia's robotics platforms before stepping out to launch her own fund. Her first move as founder and managing partner of Techable VC: betting that robotics won't get its "ChatGPT moment" anytime soon.

The veteran investor framed the field as still being in its "Lego era" during a recent Equity podcast hosted by TechCrunch's Dominic-Madori Davis. Basic building blocks — actuators, robotic hands, software, safety standards — are still being rebuilt from scratch, she argued, instead of snapping together.

That view runs counter to a popular thesis in AI. The thesis says a single, capable foundation model will soon replace most of today's specialized robotics work. Pim de Witte, CEO of General Intuition, told Equity earlier in 2025 that such a model would trigger the field's ChatGPT moment — the late-2022 shock when a chatbot suddenly did tasks nobody trained it for.

The disagreement matters because the two camps write checks very differently. A platform-first bet goes to one team building a general robot brain. A Lego-era bet spreads across dozens of small component makers. Vishwanath is clearly in the second.

What does "Lego era" actually mean?

Vishwanath's metaphor points at a foundational stage. The core components of a working robot — movement parts, end effectors, control software, and the rules that keep them near humans — are not yet standardized or cheap. Engineers still assemble robots the way kids stack mismatched bricks.

That has real consequences for venture capital. If the stack is fragmented, there's no single platform play to back. Instead, investors must pick individual components and bet that the pieces eventually snap together into something useful. Returns come from being early to the right brick, not from owning the whole box.

Hardware plumbing, not software magic, still gates progress in robotics. That's the practical read of the metaphor — and it tells founders their problem is integration, not intelligence.

Why define physical AI beyond robots?

For Vishwanath, physical AI covers anything a model can design, simulate, or control in the material world. She puts a new molecule, a new alloy, a new drug, or a single machine bolted to a factory floor into the same bucket.

This wider lens shapes her check-writing at Techable VC. The fund writes first checks across robotics, materials science, and life sciences — three areas that share a common need for models that understand physics and chemistry. A model that simulates a protein fold is solving the same kind of problem as a model that simulates a robot arm.

It also sets her apart from peers chasing humanoid demos. A family-run factory that can't find welders has a different problem than a humanoid that learns to dance on camera. One pays the rent, the other makes a press release.

Physical AI is a category, not a robot shape. Her fund treats it that way.

What does "conscious automation" add to the debate?

Vishwanath uses the term to describe systems that know what they're doing well enough to step in, step back, and explain themselves. A surgeon-facing tool that flags fatigue is one example. A line-sensing robot that pauses before a fault is another.

She frames these as the best opportunities in the space. The hardest problems — small factories short on workers, burnt-out clinicians, supply chains that can't hire — need automation that earns trust, not just speed. A model that explains why it stopped is more useful than a model that just stops.

That language also signals where she thinks returns come from. Software that wakes up only when it matters can sell into regulated industries where fully autonomous systems stay banned. Healthcare, aviation, and pharma all sit on that list.

She is underwriting automation with a brain, not automation on autopilot. That's the conscious part.

How does her Nvidia background shape her investing?

Vishwanath is one of the first women executives to spin out of Nvidia and launch her own early-stage firm. She spent seven years on the company's robotics product team, watching Nvidia's platform become the default training rig for AI-powered machines.

She now treats that network as her competitive moat. Founders she backs get warm intros to Nvidia customers, and her due diligence runs through engineers who helped build the underlying tools. A pitch she gives founders directly: "see your VC as your moat."

A fund that opens doors is more valuable in 2025 than a fund that just wires money. Distribution beats capital in a market flush with both.

Where do the US and China each have the edge?

Vishwanath declined to score the two superpowers head-to-head on robotics. She does see a clean split. The US, in her telling, leads on foundation models and the software that turns data into robot behavior. China leads on hardware iteration speed and on shipping working machines into real factories fast.

That matches the pattern investors have tracked for two years. US labs publish papers; Chinese vendors deploy units. The gap is closing on both sides, but the division of labor remains, and capital follows it.

Humanoids illustrate the gap. Dancing demos come mostly from Chinese vendors shipping quickly to YouTube. Real factory work still happens in controlled US and Korean deployments, where safety rules slow the pace but raise the bar.

The takeaway is simple: the race isn't one race, it's two, and the winning product looks different on each track.

What's the bigger bet behind Techable VC?

Vishwanath is underwriting a long, messy build-out rather than a single platform winner. Her Lego-era framing tells founders the pieces will take time to fit. Her wider physical-AI definition tells limited partners the fund is not just a humanoid hedge.

That positions Techable VC against two extremes. It is not the model-first fund Pim de Witte describes. It is also not a hardware-only shop betting on actuators alone. It is a portfolio of pieces that have to snap together, with a manager whose Rolodex helps each piece find a customer.

If she's right, robotics will look like the auto industry by 2035 — a stack of standardized parts from many vendors, sold into a market that no single company controls. If the foundation-model crowd is right, much of what she backs gets replaced by one generalist system. The Equity episode ends without a verdict, and so does this corner of the market.

For now, the smartest money in robotics may be the money that knows it can't pick the winner yet.