The Companies Getting Smarter on Your Dime

Buy the Companies That Own the Problems

The biggest winners in AI and advanced manufacturing are not just building better products. They are learning faster than everyone else—and, more importantly, learning from work their customers do for them.

For most of the industrial era, competitive advantage lived inside the company. The company owned the factory, controlled the process, trained the workforce, and optimized the supply chain. Learning—the accumulation of experience—happened within the firm’s boundaries. Scale mattered because it accelerated that internal learning curve.

Today, the strongest companies still learn internally. But their real advantage comes from becoming centers of an ecosystem. They effectively turn their customers into contributors to their learning process; their tools become the requisite tacit knowledge of all the players in that ecosystem.

We summarized this phenomenon in a recent discussion of Nvidia (NASDAQ: NVDA):

Nvidia’s… real power lives in universities, research labs, open-source codebases, and the habits of millions of programmers… CUDA (Nvidia’s programming platform) gave developers a way to think about graphics processing unit (GPU) computing. And tens of thousands of them did. The company seeded universities, funded research labs, created CUDA teaching materials, and supported faculty training programs. CUDA programming began appearing in computer science curricula around the world. Students learned GPU programming on Nvidia’s tools before they ever entered the industry…

Nvidia did all this as a conscious strategy. But there is a growing phenomenon of companies that externalize their learning not through elaborate efforts like Nvidia undertook, but by virtue of their position in the chain of production. They sit at points of constraint—where systems fail, where the world is forced to do the hard work of iterative experimentation to move forward.

Many companies may participate in those experiments, but the learning is captured by the company that sits athwart the constraint.

Consider Nova (NASDAQ: NVMI).

In a modern semiconductor fab, engineers operate at the limits of physics. Structures are too small to observe directly. The constraint is simple but brutal: you cannot improve what you cannot see.

Every wafer run becomes an experiment in trial and error. One adjustment improves yield; a second introduces a fresh defect.

The fab experiences the problem and learns locally—what worked, what failed, under its own conditions. But trial and error in a single environment can make only modest contributions to understanding.

Nova sits at the point where that experimental experience must be interpreted. Its systems send signals into structures and read what comes back. They don’t “see”; they infer—proposing a mapping from signal to outcome.

Over time, that mapping improves. Across many runs, many customers, and many process variations, Nova learns that one signal pattern correlates with a hidden defect; another predicts downstream failure; a third indicates process drift before it becomes visible.

The fabs generate the experience. Nova learns what that experience means. The advantage accrues not where the work is done, but where the results are interpreted.

That learning—which, in the Information Theory of Economics, we would call “growth”—informs not only innovation in Nova’s hardware, but the guidance it provides customers.

Following a similar pattern is Camtek (NASDAQ: CAMT).

As chip design has evolved toward systems of subcomponents—chiplets, stacked memory, advanced substrates, hybrid bonding—what we once called “packaging” has become an architecture of interfaces.

The new constraint: complex systems fail at their interfaces.

Each connection point is a risk: a possible misalignment, a defect, a bond gone bad. Worse, failures do not remain local. A defect in one layer can compromise an entire stack. As complexity rises, so does the number of interfaces—and the number of ways things can go wrong.

This is where Camtek sits. Its systems detect and classify defects. Every failure becomes a data point: what failed, where it originated, how it spread.

At small scale, this looks like inspection. At large scale, it becomes a continuously expanding library of how complex systems fail.

Over time, patterns emerge. Detection improves. Classification becomes more precise. Prediction improves—not only of failure, but of how to avoid it.

The learning happens in production—across OSATs (Outsourced Semiconductor Assembly and Test companies) and packaging lines.

But that learning accumulates in the system that sees all the failures: Camtek’s.

Finally, consider Mobileye (NASDAQ: MBLY).

Mobileye, the world’s leading supplier of autonomous and driver-assist systems, faces the most implacable constraint of all: our unruly world.

Autonomous driving fails not on average conditions, but on edge cases—rotten weather and worse drivers, or what engineers call “ambiguous visual inputs.”

No simulation can fully capture the unexpected. The only way to learn is to experience it. The constraint becomes: seeing what happens on real roads.

Mobileye solves this by being present in millions of vehicles. Every mile driven generates data. Every anomaly—every edge case—is captured.

The system accumulates experience of reality—not just approximations of it.

This may be the ultimate outsourcing of the learning curve. The world drives. Mobileye learns.

Large markets create more learning. But these companies have more going for them than scale alone. They are positioned at very specific points:

• where progress is hard
• where problems are unavoidable
• and where solutions must be expressed through their systems

Not all constraints create competitive advantage. A constraint becomes a moat only if solving it requires working through your tools—and if the solutions can be captured and reused.

That is why Nvidia, Nova, Camtek, and Mobileye are so powerful. It is also why, of the four, Mobileye may ultimately be the most vulnerable—because, in principle, any competitor can accumulate similar real-world driving experience over time.

In the 19th and 20th centuries, learning curves were internal. Experience accumulated inside factories, organizations, and tightly controlled systems.

Today, learning happens across the ecosystem—but it is not captured equally. It flows back to the nodes that sit where problems get hard and capture how those problems are solved.

For investors, the right questions now go beyond “What does the company own?” and “How big is the market?” to: Where are the hardest problems and whose systems do the solutions run through?

Because that is where learning concentrates.

And where learning concentrates, wealth follows.