I still remember the first time I saw a demo that felt like actual magic. It was years ago, a tiny startup that had somehow taught a flip phone to do real-time object recognition. A flip phone! Everyone else saw a cool party trick. I saw a brutal, relentless data pipeline. They had spent months building a system to capture, label, and process millions of images in every lighting condition you can imagine. That company was acquired for a very, very nice number. The magic wasn't the demo; it was the data.
I get that exact same feeling watching the latest videos from Figure AI. You've seen them. A sleek humanoid robot, Figure 03, making coffee. Tidying up a factory. It's damn impressive. It gets millions of views. And as an investor who's written checks to over 200 companies—including foundational AI labs like Anthropic and OpenAI—my inbox explodes. "Is this it? Is this the one?" My answer is always the same: don't watch the robot. Watch the data.
The Demo is the Sizzle, Not the Steak
Let's be clear: the team at Figure is world-class. What they've pulled off in such a short time is remarkable. The videos of their Helix 02 model showing full-body autonomy are proof of serious engineering talent. It's easy to get lost in the visual spectacle. A robot that moves with near-human grace is a powerful thing.
But I've been in this game too long to be swayed by a good demo. I once backed a logistics company whose public dashboard was a work of art. It showed packages gliding around a warehouse in a beautiful symphony of efficiency. But the real business—the actual defensible moat—was the ugly, complicated beast in the background ingesting data from thousands of different barcode scanners, RFID readers, and manual entry points. The slick front-end was just a pretty picture of the data they had truly mastered.
That’s what I see with Figure. The demos are in a lab. The robot is doing things it has been specifically trained on. The real world isn't a lab. It's a chaotic mess of unexpected obstacles, weird lighting, and unpredictable humans. The gap between a demo and a commercially viable humanoid that can work 24/7 in a real warehouse won't be closed by shinier hardware. It will be closed by data.
The Data You Don't See
This brings me to the real story. The part that gets almost no attention in the breathless news articles. It’s what Figure calls “Project Go-Big.” While everyone is mesmerized by the robot making coffee, Figure is quietly building the single most important part of their company: a massive, internet-scale data collection and pretraining system for humanoids.
Think about what happened with language models. For years, progress was slow. Then, companies like OpenAI and Anthropic realized the key wasn't just a better algorithm, but training that algorithm on a simply staggering amount of text from the internet. The sheer scale of the data created a phase change. It took models from being interesting toys to tools that are actively changing the world.
This is the playbook for robotics. The company that wins the humanoid race will be the one that builds the best data engine. Figure gets this. They aren't just building a robot; they are building a data factory. Their partnership with Brookfield shows they are already putting robots in real-world environments to collect data. This is the alpha. This is the hidden truth. Figure's success doesn't depend on the robot's current abilities. It depends on the quality and quantity of the data it is collecting every single second.
My opinion is strong on this: it's a data race, plain and simple. The hardware will become a commodity. Different companies will build robots with similar physical specs. But the data flywheel—that virtuous cycle of collecting data, using it to improve the AI, and then deploying better robots to collect even more data—is incredibly hard to copy.
The Messy Reality of Working with People
Now, let's talk about the other side of the coin. Building a data-gobbling machine is one thing. Making it work alongside fragile, unpredictable humans is another thing entirely. A recent lawsuit filed by a former engineer, reported by CNBC, claimed the robot could “fracture a human skull.” This isn't just a PR headache; it's a fundamental challenge for the entire field.
This is the surprising truth the slick demos hide: human-robot interaction is the hardest, messiest, and most important problem to solve. It’s not just about avoiding collisions. It’s about reading intent. It’s about understanding social cues. It’s about building a machine that an average warehouse worker feels safe and comfortable being around.
I remember visiting a factory that had installed a new automated inventory system. The tech was flawless. But the workers hated it. The interface was confusing, the system was rigid, and it didn't account for the dozens of little exceptions and workarounds that are part of any real human process. That multi-million dollar system was eventually ripped out. It failed not because the tech was bad, but because the designers didn't understand the humans who had to use it.
With humanoid robots, the stakes are infinitely higher. A bad UI on a computer is annoying. A bad UI on a 150-pound robot can be lethal. The data Figure collects can't just be about navigating physical space. It has to be about navigating social space.
My Robotics Investment Thesis
So, when I look at a company like Figure, or any other robotics startup, I’m not just looking at the robot. I’m looking at the entire system through a specific lens. My investment thesis for this space comes down to three things:
First, a Data-First Obsession. Is the founding team obsessed with data? Do they talk more about their data pipeline than their hardware specs? I want to see a clear strategy for building a data flywheel. I want to see simulation environments, auto-labeling systems, and a relentless focus on acquiring unique, proprietary data.
Second, Pragmatic Deployment. Are they trying to boil the ocean, or are they starting with a specific, structured environment where they can prove their value and collect data in a controlled way? Figure's focus on logistics and warehouse automation is a smart move. It’s a massive market with clear, repetitive tasks. It’s the perfect training ground.
Third, Human-Centric Design. Is the team thinking about the human element from day one? Are safety, ethics, and user experience core to their design process? I want to see teams that include psychologists and UX designers, not just roboticists. The winning robot will be the one people actually want to work with.
These principles have served me well, whether I'm investing in a SaaS company or a deep-tech AI lab. The core challenge is always the same: how do you build a system that solves a real problem for real people in the real world?
The Quiet Revolution
It’s easy to get caught up in the hype. The videos of humanoid robots are compelling. They speak to a vision of the future we've been dreaming of for decades. But if you want to understand where this is all really going, you have to look past the shiny object.
The real revolution isn't being televised in a viral demo. It's happening quietly, in the background. It's in the whirring servers of a data center, in the lines of code that run a simulation, and in the careful, painstaking process of designing a machine that can earn our trust.
The winner in the humanoid race won't be the company with the most polished chrome or the most acrobatic robot. It will be the company that amasses the messiest, most complex, and most comprehensive dataset of what it truly means for a machine to interact with our world. That's the real story. That's the hidden data.
Frequently Asked Questions
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.
How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.