Ex-Nvidia-turned-VC Amulya Vishwanath: ‘Physical AI is in its LEGO era’

The Lego Block Phase of Robotics

For a long time, robotics felt like a science project that never quite made it out of the lab. You had massive industrial arms that could do one thing perfectly a million times, but if you moved the target two inches to the left, the whole system broke. We were building specialized tools, not intelligent agents. But the conversation is shifting. Amulya Vishwanath, a former Nvidia hand now navigating the venture capital space, recently noted that we are entering what she calls the Lego era of physical AI.

As a founder, this metaphor resonates because it describes the shift from monolithic, hard-coded engineering to modular, data-driven intelligence. We aren't just building better hardware anymore; we are building the connective tissue that allows hardware to see, feel, and react. The blocks are finally starting to click together.

Beyond the ChatGPT Moment

Everyone in the valley is obsessed with finding the ChatGPT moment for every sector. For robotics, that moment isn't just about a chatbot that can tell a joke. It is about foundation models for movement. In the same way LLMs consumed the internet to learn how we talk, physical AI models are beginning to consume video and sensor data to learn how the world works.

This is a massive shift for builders. Historically, if you wanted to build a robot that could fold laundry, you had to program every joint rotation and grip pressure. Today, we are seeing the rise of general-purpose models that understand the physics of fabric. The goal is no longer to code the action, but to train the intent. Vishwanath’s perspective highlights that we are moving toward a world where the intelligence is portable across different physical forms.

The Silicon Realignment

Coming from Nvidia, Vishwanath understands something most software founders miss: the sheer compute cost of moving a physical object in real-time. Training a model on text is relatively cheap compared to simulating a three-dimensional environment where gravity, friction, and light all matter. This is why the Lego era is so important. We can't afford to rebuild the brain every time we change the body.

For those of us in the crypto and AI space, there is a clear parallel here to open-source protocols. When you have modular components—standardized ways for sensors to talk to actuators—you lower the barrier to entry. You don't need a billion-dollar R&D budget to experiment with physical AI if you can snap together pre-trained vision models with standardized robotic chassis.

What This Means for Founders

If you are building in this space, the opportunity isn't necessarily in building the hardware itself. That is a capital-intensive race to the bottom where giants like Tesla and Boston Dynamics have a massive head start. The opportunity is in the middleware of intelligence.

• Data Collection: How do we gather high-quality teleoperation data that isn't just noise?
• Simulation-to-Reality (Sim2Real): How do we bridge the gap between a robot learning in a digital void and performing in a messy warehouse?
• Edge Optimization: Physical AI can't rely on a slow cloud connection. It needs to make decisions locally and instantly.

We are seeing a move away from the idea that a robot is a computer with wheels. Instead, we are seeing robots as physical manifestations of a broader AI ecosystem. This creates a playground for founders who can solve specific bottlenecks in the supply chain of movement.

The Skeptic's Corner

I’ve seen plenty of hype cycles, and robotics is famous for overpromising. We’ve been told for a decade that self-driving cars were a year away. The reality is that the physical world is much harder to solve than the digital one. A hallucination in a chatbot is a funny tweet; a hallucination in a two-ton autonomous vehicle is a tragedy.

The Lego era implies that things are easy to put together, but anyone who has stepped on a stray brick knows that physical reality is unforgiving. We are still missing the standardized operating system that makes these components truly interoperable. Vishwanath is right about the direction, but builders should be wary of the timeline. The infrastructure is being laid, but the skyscraper isn't finished yet.

Building for the Physical Future

For the crypto-native audience, think of physical AI like decentralized infrastructure (DePIN). We are looking at a future where physical assets are governed by intelligent models. The convergence of AI's decision-making and robotics' execution is where the next decade of value will be created.

The takeaway for builders is simple: stop trying to build the whole robot. Find the specific Lego block that is missing—whether it's a better way to process tactile feedback or a more efficient way to compress spatial data—and own that piece of the stack. The era of the generalist robot is coming, but it will be built by specialists who understand that modularity is the only way to scale.

The Lego era isn't just about building toys; it is about creating a standard language for how machines interact with the real world.

We are currently in the messy phase where all the pieces are dumped out on the floor. It looks chaotic, but for the first time, we actually have all the parts we need to build something that moves as well as it thinks.

Read the original at TechCrunch AI →