Robotics Is Still in Its LEGO Era, Not Its ChatGPT Moment
Amulya Vishwanath, founder and managing partner of Techable Ventures, says robotics has not reached its "ChatGPT moment" and is instead in a "LEGO era." She gave the assessment in an interview with TechCrunch's Equity published Oct. 20, 2026. TechCrunch Techable Ventures focuses on physical AI, the term for AI systems built to run machines in the physical world.
Vishwanath spent nearly seven years as a product marketer at Nvidia before starting Techable Ventures. Before Nvidia, she held a product marketer role at a company later bought by Intel. Earlier, she worked as a digital signal processing engineer at Interconnect Systems, which was later bought by Molex.
Techable VC is an early-stage venture capital firm investing in physical AI. Berkeley Haas Berkeley Haas identifies Vishwanath as EMBA 25. The firm describes its remit as physical AI and sciences.
Her core point concerns the foundation. Foundational software and building blocks for robotics are being rethought to allow innovation to scale. People are currently building basic components like robotic hands and actuators, the motors and joints that create movement, that will eventually come together like LEGOs. The work, in her description, is at the level of piece parts and interfaces rather than finished general systems.
Talent figures centrally in her read of the market. Talent is critical and remains a key definer of the U.S. market. Much talent is going to large companies like Anthropic and Nvidia.
The Oct. 20 interview followed two related TechCrunch features published Oct. 16, a podcast titled "Physical AI is more than robots, says the Nvidia veteran now investing in it" and a video titled "An Nvidia insider-turned-VC on why robotics hasn't hit its ChatGPT moment yet," both featuring Vishwanath. TechCrunch
In my view, the LEGO image points to a specific technical condition for builders. Manipulation hardware, actuation, sensing, middleware, the software that links hardware to applications, simulation pipelines and policy architectures, the AI structures that select actions, are maturing on separate tracks, with integration still custom and brittle. Progress depends less on a single model breakthrough and more on stable abstractions, repeatable evaluation, and hardware that can be sourced, swapped and maintained without rebuilding the stack. Modular periods reward tooling, interoperability and manufacturing discipline. They also shift early venture bets toward picks, shovels and interface layers, with system-level defensibility arriving later once parts fit reliably. The hiring constraint she names matters, because when top researchers cluster inside large labs with compute and data advantages, startups must compete on access to robots, real-world data loops and deployment sites rather than on model scale alone. The long arc is constructive. If the piece parts improve in parallel, systems get cheaper to build, easier to test and faster to deploy.
