An Nvidia insider-turned-VC on why robotics hasn’t hit its ChatGPT moment, yet

The Lego Era Delusion

Everyone is looking for the killer app in robotics. They want to see a humanoid walk into a kitchen, look at a pile of dirty dishes, and understand the physics of grease and ceramic well enough to clean up without breaking a plate. We see the videos on social media—shiny metallic torsos doing backflips or making lattes—and the hype machine tells us we are one firmware update away from the ChatGPT moment for physical machines. But if you talk to the people who actually built the platforms these robots run on, the reality is a lot less cinematic.

Amulya Vishwanath spent seven years at NVIDIA leading product marketing for their robotics division. She has seen the guts of the industry from the inside of the company currently providing the brains for almost every AI project on earth. Her take is a necessary cold shower for the founder community: robotics isn't waiting for a smarter brain; it is struggling because the body is still a collection of mismatched parts. She calls this the Lego era, and it is a period defined by rebuilding the basics rather than scaling the breakthroughs.

Building Without a Blueprint

In the software world, we take a lot for granted. If you want to build a new AI-powered SaaS, you have standardized APIs, cloud infrastructure that scales with a click, and programming languages that play nice together. You are building with finished bricks on a level foundation. Robotics is the exact opposite. Every time a founder starts a new robotics company, they aren't just writing code; they are often reinventing the wheel—or in this case, the actuator, the hand, and the safety protocol.

The physical world is stubborn. You cannot simply port a large language model into a bipedal frame and expect it to navigate a warehouse. The hardware is fragmented. Actuators, which are essentially the muscles of the robot, aren't standardized. The sensors that help a robot perceive depth and texture vary wildly in quality and data output. When your building blocks don't fit together, you spend 90% of your venture capital just trying to get the machine to stand up, leaving only 10% for the actual intelligence that makes the product valuable.

The Foundation Model Myth

There is a prevailing narrative that if we just throw enough data at a transformer model, robotics will solved. The idea is that a foundation model for physical movement will emerge, allowing robots to learn by watching videos of humans just like GPT learned by reading the internet. While that is a nice vision for a pitch deck, it ignores the friction of the real world. A chatbot doesn't have to worry about gravity, friction, or battery life. A robot does.

Vishwanath’s perspective is that we are skipping steps. You can't have a ChatGPT moment when the hardware still feels like a science project. We are currently seeing a massive push toward humanoid robots because they look like us, but from a builder's perspective, they are the hardest possible form factor to master. We are trying to build the most complex version of the technology before we have even standardized the basic components. It is like trying to build an iPhone before anyone agreed on how a battery or a screen should work.

What This Means for Builders

If you are a founder in this space, the temptation is to lean into the AI hype. It is easier to raise money by talking about neural networks than by talking about the torque density of your joints. But the real value in the next five years isn't going to be in the flashiest demo. It’s going to be in the companies that solve the plumbing of robotics.

We need better middleware. We need hardware that is modular and reliable. We need safety standards that aren't just reactive but are baked into the silicon. If you are building a robotics startup today, you shouldn't be asking How do I make this robot smarter? You should be asking How do I make this robot easier to build and maintain? The winners won't necessarily be the ones with the most sophisticated AI, but the ones who figure out how to move robotics out of the custom-built, bespoke phase and into the mass-production phase.

The Skeptic's Advantage

It is easy to get caught up in the hardware-is-hard trope, but there is an opportunity in this friction. Because it is so difficult to get these disparate parts to work together, the barrier to entry is massive. Unlike the AI software space, where a wrapper can be built in a weekend, a functional robotics company requires deep institutional knowledge and a lot of physical trial and error.

Vishwanath’s insights suggest that we should stop looking for a single moment where everything changes. There won't be a single day where robots suddenly become ubiquitous. Instead, it will be a slow, grinding process of standardizing those Lego bricks. We will see improvements in how robots handle objects, how they navigate tight spaces, and how they interact with humans, but these will be incremental wins, not a sudden explosion of capability.

The industry is currently obsessed with the finish line, but we haven't even finished paving the track. We are rebuilding the basic building blocks of what it means to be a robot.

The Takeaway

For those of us watching from the sidelines, the lesson is clear: don't mistake movement for progress. A robot doing a dance on YouTube is a marketing feat, not a technical revolution. The real work is happening in the unglamorous parts of the stack—the sensors, the actuators, and the low-level software that makes them talk to each other.

Robotics will have its moment, but it won't look like a chatbot. It will look like a standardized set of tools that allow any developer to build a physical machine as easily as they build a website. Until those Lego bricks actually snap together, we are just playing in the sandbox.

Read the original at TechCrunch Venture →