Robotics Inflection Point
Amulya Vishwanath, who moved from Nvidia into venture capital, is betting that physical AI is nearing a breakthrough that could reshape robotics in the same way large language models transformed software. Speaking to TechCrunch's Equity, she described the current state of the sector as "its LEGO era," a phase in which builders are still assembling the core pieces of the stack rather than deploying fully mature systems at scale.
Her framing captures a broader tension in frontier AI: while generative models have already delivered a consumer and enterprise revolution in text, code, and images, robots remain far harder to productize. Machines that operate in the physical world must contend with messy environments, sensor noise, safety constraints, and the near-infinite variability of real-world tasks. That complexity has kept robotics from producing a single defining moment comparable to ChatGPT's launch in late 2022.
Vishwanath's point is not that the breakthrough has already arrived, but that the ingredients are finally converging. Compute is more powerful, foundation models are becoming more capable, and the robotics ecosystem is learning how to combine perception, planning, and action in ways that were impractical only a few years ago. In her view, the sector is moving from isolated demonstrations toward a more composable architecture, where hardware, software, and data pipelines can be mixed and matched to accelerate development.
The LEGO Era
The "LEGO era" metaphor is especially apt because it suggests both promise and incompleteness. In the same way that LEGO bricks enable endless combinations while remaining standardized, physical AI is beginning to benefit from reusable components: simulation environments, robot policies, vision-language models, and increasingly specialized hardware. Yet the field is still far from a universal platform that can be dropped into any machine and reliably perform useful work.
That distinction matters for investors and operators. Robotics has long attracted enthusiasm, but commercial adoption has often lagged behind the hype cycle. Industrial automation is mature in constrained settings such as warehouses and factories, but general-purpose robots that can navigate homes, hospitals, retail spaces, or construction sites remain difficult to deploy profitably. Vishwanath's thesis implies that the next wave of winners may not be the companies chasing a single humanoid headline, but those building the infrastructure that makes physical AI repeatable.
Her comments also reflect a growing shift in how the market talks about robotics. The conversation is moving away from narrow task automation and toward embodied intelligence: systems that can perceive, reason, and act in the physical world with increasing autonomy. That shift has drawn attention from major AI labs, chipmakers, and venture firms alike, all of which see robotics as a potential long-duration market for frontier models.
Waiting For The Breakthrough
The phrase "ChatGPT moment for robotics" has become a shorthand for the kind of leap that turns a technical demo into a mass-market phenomenon. But Vishwanath's remarks underscore why that moment has been slower to arrive in physical AI than in software. Language models can be trained on vast digital corpora and evaluated quickly; robots must learn from embodied experience, where mistakes can be expensive, dangerous, or irreversible.
That reality has pushed the industry toward hybrid approaches. Companies are increasingly blending simulation with real-world data, using foundation models to improve generalization, and designing systems that can operate within carefully bounded environments before expanding outward. The result is a field that is advancing quickly, but unevenly.
For Nvidia, the broader backdrop is significant. The company has become a central supplier to the AI boom, and robotics is one of the clearest extensions of that hardware-led ecosystem. Vishwanath's perspective, shaped by experience inside the chip giant and now in venture investing, suggests that the next phase of frontier AI may depend less on a single model breakthrough than on the industrialization of the entire robotics stack.
The investment implication is straightforward: physical AI may not need a dramatic overnight revelation to become transformative. It may instead emerge through a series of compounding improvements that make robots cheaper to train, easier to deploy, and more capable in unstructured settings. If Vishwanath is right, the field is not waiting for a miracle. It is waiting for the pieces to click together.