In the Beginning, There Was Hardware
Before software ate the world, hardware built it. Apple, Intel, Cisco were not just tech companies; they were the early blueprints of venture capital success. The first wave of Silicon Valley was, quite literally, silicon.
And yet, for the past two decades, venture has pivoted hard towards software. The logic was simple: scalable, capital-efficient, high-margin businesses that move at internet speed. Hardware, on the other hand, slipped out of the spotlight—code was king, atoms were a liability.
Of course, software-driven models reshaped the landscape and will continue to dominate many categories, but with AI, the playing field resets daily.
Open-source models, fine-tuning toolkits, and cloud APIs have democratized access. What used to be a moat of technical complexity and model IP is increasingly just a starting line.
While software can still build massive businesses, in some of the most defensible AI plays, hardware integration is becoming a powerful amplifier of differentiation.
Today, Real Defensibility Lies in the Fusion of Hardware and Software
AI software is evolving at breakneck speed. But with that speed comes a problem: it commoditizes fast. What’s cutting-edge one month can be yesterday’s demo the next.
This is where hardware steps back into the spotlight. In AI-native systems, hardware isn’t just the container, it’s the source of differentiation. The ability to generate, process, and act on unique data at the edge creates a feedback loop that software alone can’t replicate.
And when hardware and software are co-designed, deeply integrated, and infused with learning capabilities, something truly powerful happens. The system itself becomes intelligent, adaptive, and extremely difficult to replicate.
Exodigo: A Case Study in Full-Stack AI Moats
At Greenfield Partners, we’ve seen this firsthand through our investment in Exodigo. The company combines cutting-edge multi-sensor hardware with proprietary AI models to map the underground world—utilities, pipes, voids—with unparalleled precision.
Here’s the critical insight: the hardware itself isn’t just a delivery mechanism; it’s the source of proprietary data. Without Exodigo’s custom sensor stack, the AI has nothing unique to learn from; the sensor data is just noise. Only by owning and integrating both do you unlock the system’s full power.
This isn’t just a novel approach, it’s a defensible one. Every scan improves the model. Every deployment tightens the feedback loop. The result is a data moat that’s nearly impossible to replicate without matching the entire stack, something no one else in the space has done.
The Deep Tech Surge: Hardware-Heavy, Impact-Heavy
It’s not just Exodigo. Some of the most consequential companies in today’s private markets are deeply hardware-driven:
- SpaceX – $350B valuation, vertically integrated from rockets to satellites.
- Anduril – $28B valuation, blending AI with defence hardware and autonomy.
- The Boring Company, Helsing, Figure AI, Covariant, Agility Robotics — all combining physical systems with intelligence.
These companies were once seen as outliers. Now they’re blueprints for how bits and atoms reinforce one another.
VCs Are Catching Up
Firms like Valor Equity Partners and Lux Capital have long recognized this trend. They’ve consistently backed deep tech founders building complex systems that combine physics, software, and machine intelligence.
At Greenfield, we share this conviction, backing companies like Exodigo, as well as innovators in photonics (DustPhotonics), autonomous systems, as well as two unannounced investments in this space, and beyond. However, our lens isn’t limited to only hardware-heavy models. Whether it’s pure software or full-stack, what matters is defensibility: moats built from data, distribution, network effects, or, when the use case demands it, the integration of atoms and bits.
The Future Belongs to Full-Stack Builders
To deep tech founders: you’re not swimming upstream anymore. Investors now understand that moats aren’t just made of code—they’re built from integrated systems that learn and improve together.
For some, that means full-stack architectures where hardware and software reinforce one another. For others, it will mean software-driven moats built on proprietary datasets, vertical workflows, distribution advantages, or network effects. The common thread is defensibility.
---
If you’re building in this space, whether through bits alone or the fusion of bits and atoms, we’d love to hear from you.
Submissions are open for the AI Disruptors 60, our list recognizing early and growth-stage startups shaping the future of AI.
