From Scraps to Billions: How Intelligence Factory is Solving Robotics' Real Bottleneck

When you think about robotics, you probably imagine cutting-edge hardware: sleek mechanical arms, precision actuators, sensors that cost a fortune. But Yash Sinha and his team at Intelligence Factory have identified a fundamental shift in the industry that most people are still sleeping on: hardware isn’t the bottleneck anymore. Intelligence is.

This realization led Sinha and his co-founders to tackle one of the hardest problems in AI today: building foundation models specifically designed for robotic manipulation. It’s a bet-the-company decision that reveals something crucial about building in deep tech—sometimes the most defensible path forward isn’t the most profitable one in the short term.

The Bottleneck Has Shifted

For decades, robotics companies obsessed over hardware. How can we make the arm faster? More precise? Stronger? But as hardware has commoditized and improved, a new constraint has emerged: robots need to be intelligent. They need to understand the world. They need to manipulate objects with the kind of dexterity that humans take for granted.

Here’s the kicker: for robots to have any commercial value, they need to work at a 99% success rate. That’s not a nice-to-have. That’s a requirement. A robot that works 90% of the time isn’t useful in a warehouse. It’s a liability.

So you’re left with two massive problems. First, robotics data is scarce. Unlike computer vision, where you can scrape millions of images from the internet, there’s no public dataset of robots successfully picking and packing objects across thousands of different items and scenarios. Second, the diversity of real-world tasks is staggering. Your Amazon order looks nothing like mine. The objects are different. The packaging is different. The arrangement is different.

This is where Intelligence Factory comes in. They’re building the infrastructure to collect high-quality robotics data at scale, using it to train foundation models, and then deploying those models in real environments. It’s a vertical integration play in an industry that’s been fragmented for too long.

The Hardest Trade-Off: Data vs. Deployment

When you can collect high-quality robotics data at scale, you have options. You could sell that data to other robotics companies. It would probably be lucrative. In the short term, it might be the smarter financial move.

Sinha and his team chose differently. They decided to use their data advantage to build and deploy their own models. It’s riskier. It requires more capital. It requires hiring roboticists and ML engineers. It requires managing complex operations across multiple deployment sites. But it’s also the only path that builds a defensible, long-term business.

Here’s why: if you’re just selling data, you’re not building a moat. Your competitive advantage erodes the moment someone else figures out how to collect data at scale. You’re helping your competitors build better models. You’re commoditizing your own advantage.

But if you own the deployment, you own the relationship with the customer. Switching costs are high. Your model improves with every deployment because every deployment generates more data. You’re building a flywheel.

This decision—to play the longer game—is exactly the kind of thinking that separates companies that build sustainable businesses from companies that chase short-term revenue. It’s also the kind of decision that requires conviction. Sinha and his co-founders had that conviction partly because they both had deep robotics backgrounds and understood the stack end-to-end.

Building the Foundation First

One of the most insightful parts of Sinha’s approach is how he thinks about building a deep tech company. There’s a culture in startups right now where you raise money, you scale aggressively, you consolidate your position. But Sinha argues that in deep tech, you need to build an extremely solid foundation first.

For Intelligence Factory, that foundation is the data infrastructure. Before they scale deployments, before they build the perfect model, they need to ensure they can consistently collect high-quality data. That’s the unglamorous work. That’s the work that doesn’t make headlines. But it’s the work that determines whether your models will actually work in the real world.

Their approach is methodical:

1. Identify target deployments. Find 4-5 companies interested in piloting the technology.
2. Understand the task. What objects do they need to pick and pack? What’s the actual use case?
3. Collect data. Use their data factories to gather high-quality examples of these tasks.
4. Train the model. Build a model that can handle this specific deployment.
5. Deploy and iterate. Put the robot in production, collect more data, improve the model.

With each deployment, the model’s action space expands. It sees more objects, more scenarios, more edge cases. Over time, you build a model that can pick and pack anything. That’s the MVP. That’s already a billion-dollar market.

Then you expand to other tasks: blocking and plugging, cleaning, assembly. Each new task, each new deployment, feeds back into the model.

Hiring for the Frontier

When you’re working on problems that haven’t been solved before, traditional hiring criteria break down. You can’t just look at credentials and past experience. You need people who can think from first principles.

Sinha spends an enormous amount of time on hiring because he understands that the team is everything. YC’s job board has been helpful for inbound, but the real work is in the interviews. He’s looking for people who can extrapolate, who can reason about problems that don’t have established solutions, who can handle the ambiguity of working on the frontier.

It’s a different skill set than hiring for a typical software company. You’re not hiring people to execute on a well-defined roadmap. You’re hiring people to figure out what the roadmap should be.

The Lesson for Technical Founders

What’s remarkable about Sinha’s approach is how disciplined it is. There are no shortcuts. There’s no hype. There’s just a clear-eyed assessment of what needs to happen for robots to become a viable commercial product, and a methodical plan to make it happen.

For technical founders, especially those working in deep tech, the lesson is this: defensibility matters more than short-term revenue. The hardest decisions are often the ones that require you to leave money on the table today in order to build something more valuable tomorrow. And building that something requires obsessing over the foundation—the data infrastructure, the team, the first principles thinking—before you scale.

Want to hear the full story? Listen to the complete episode with Yash Sinha on Code Story to dive deeper into how Intelligence Factory is building the future of robotic manipulation, the specific challenges they’ve overcome, and what’s next for the company. Subscribe to Code Story on your favorite podcast platform and don’t miss this conversation with one of the founders pushing robotics forward.