The Loop is the Moat

What we look for in physical AI companies

By Ian Heinrich

If we can't build thousands of autonomous surface vessels, we shouldn't build any at all. - Dino Mavrookas, Founder/CEO, Saronic

Ensemble was founded nearly a decade ago on two main beliefs:

1. ‍proprietary data will reshape the most important companies of the future, and‍
2. America's physical economy needs to be reinvented.

These ideas might seem separate. We've come to see them as one.

The idea of reindustrialization, reshoring, defense, energy, housing, has shifted from an outlier view to something close to consensus. We share in the excitement. But the more time we spend with founders building in the physical world, the more clearly we see what separates the companies that will matter from the ones that won’t.

The companies that will define the future of the physical world treat every build as a chance to generate data, feeding what they learn back into the next unit, improving efficiency and cutting costs with each iteration. The only path onto the compounding curve is through disciplined repetition (see Wright’s Law). The real asset isn't just the thing they built, it’s the system that builds it, and the data that system generates over time.

Useful data means less variance. The winners reduce degrees of freedom until cause and effect becomes clear; then they scale. Every decision that makes the process more controllable makes the learning faster and the moat deeper.

There are two kinds of AI. Only one builds a moat.

There are two things a physical company wants AI to do, and they have opposite data requirements.

Operational AI (quality control, predictive maintenance, anomaly detection) can run on imperfect data. These tools are real and valuable, and increasingly becoming commoditized. Any company can buy them.

What can't be bought is systems AI: the model that learns how to build the thing better, predicts how design changes will affect production outcomes, and eventually drives toward autonomy. That model requires comparable observations of the same underlying process, repeated enough times to have genuine predictive power. AI can find patterns in chaos. It cannot predict the future of a process that has no consistent past.

The best founders we've seen aren't just thinking about how to use AI on their production data. They're designing their production process as a data generation strategy from day one. The loop isn't a prerequisite for AI. It's what makes your AI impossible for anyone else to replicate.

What this looks like in practice

Founders who get this entirely flip the conventional sequence. Instead of the canonical advice of prototype, find product-market fit, then figure out manufacturing, they let manufacturing requirements shape the product from inception.

One of the best examples from our portfolio is Saronic. Saronic's head of manufacturing, recruited from SpaceX, was their third employee, hired before they had a prototype. The result: 80% hardware consistency across platforms, seven core modular components, a fully vertically integrated supply chain. As Saronic’s founder Dino Mavrookas puts it, "If we can't build thousands of autonomous surface vessels, we shouldn't build any at all."

In construction, ICON built their manufacturing system - the Vulcan printer, the Magma mixing unit, the BuildOS software - before they had commercial scale. Every house instruments the process, not just delivers a unit. The result is meaningful cost reduction with every doubling of cumulative production. The home was never the product. The system that prints it is.

What we’ve seen work is to figure out which 80% of your build is predictable and design systems that absorb the other 20% without breaking the loop. Customers want custom, regulations vary, site conditions differ. Instead of ignoring that variance, design around it so the exception never breaks the system. Every custom engagement is a tax on the loop. So, price bespoke work accordingly, and importantly, decide before the customer is in the room which deals you'll walk away from.

The project trap

Most companies never get there. Not because of the vision or the market, but because they fall into what we call the project trap: each output begins from a similar baseline as the last. The system fails to learn and costs don't decrease. The tenth build is just as unpredictable and unprofitable as the second.

The trap is insidious because it feels like progress. Custom projects bring in revenue, please investors, and attract attention. Katerra raised nearly $3 billion on the right pitch: factory-built components, vertical integration, costs driven down through scale––and still fell into the trap. They said yes to everything. The factory never repeated the same thing twice and thus never learned. Six years later they filed for Chapter 11. 

Fisker, Proterra, Solyndra: different industries, same ending.

The incentives that sustain the trap feel reasonable at every step. That's what makes it hard to escape from inside it.

A simple test

When demand doubles, does the company get better or just get busier? 

Projects get busier. Products get better.

The founders who escape the trap do so from inception, before the prototype, before the first customer, before the first hire. They instrument what they can't yet interpret and walk away from revenue that breaks the loop. They understand that the data their system generates is the asset that eventually no one else can buy their way into.

We've built Ensemble the same way. A dedicated data science and engineering team has been part of our firm since inception, applying the same systems AI logic to venture that we look for in the companies we back. The loop isn't just a thesis. It's how we work.

Some of our most important relationships started with tinkerers and back of the napkin sketches. If you're still figuring out the shape of the thing but you're thinking about it the right way, that's enough to start a conversation. We invest from pre-incorporation through Series A and we're most useful earliest.

If any of this resonates, we want to meet you.

ian@ensemble.vc