The Moat Outside Nvidia
When investors talk about “moats,”—durable barriers to competition—they usually mean things a company owns: patents, factories, or proprietary technology.
But some of the strongest moats in technology companies live elsewhere: beyond company borders, inside the minds of the people who build on their platforms. And those people are not employees. They are pursuing their own goals.
Michael Polanyi captured the phenomenon decades ago with a simple observation: “We know more than we can tell.” Much of what humans know is not written down in manuals or captured in formal rules. It is tacit knowledge—skill acquired through experience.
This kind of knowledge turns out to be an extraordinarily powerful moat not only because it is not written down, but precisely because the people with the knowledge are outside the company and pursuing not the companies interests but their own.
Nvidia’s CUDA (Compute Unified Device Architecture) ecosystem may be the clearest modern example: millions of programmers have learned how to solve problems on its architecture, writing software that runs directly on its graphics processors rather than traditional central processing units (CPUs).
Nvidia’s presentations and developer ecosystem reports often cite 4-5+ million developers. CUDA has been taught in thousands of university courses worldwide.
That knowledge lives in habits, instincts, and accumulated experience—the craft knowledge—of people who have spent years solving problems on a particular architecture.
That knowledge does not belong to Nvidia. It sits outside the company—in universities, research labs, startups, and open-source software projects. Yet it functions as a powerful moat.
A rival chip maker can build competitive hardware. But recreating 20 years of accumulated developer experience is far harder. Programmers would have to relearn performance tricks, rebuild libraries, retrain teams, and debug unfamiliar architectures.
The barrier is not the chip. The barrier is the experience curve.
The same dynamic appears in the software frameworks that dominate artificial intelligence (AI) research. PyTorch, now the standard environment for training advanced AI models, was designed from the outset to run most efficiently on Nvidia graphics processing units (GPUs). Millions of researchers build their models in that environment without ever thinking about the underlying hardware. Yet every experiment, every optimization, and every debugging session adds to the body of tacit knowledge surrounding CUDA-based systems.
This is why technological competition often hinges on ecosystems rather than components. The architecture that attracts the most developers tends to accumulate the most knowledge. And the architecture that accumulates the most knowledge tends to become the default platform for innovation.
This is not a new thing.
In the late 19th and early 20th centuries, British shipyards dominated global shipbuilding. Yet British firms did not possess secret designs. Ship blueprints circulated widely.
The Brit’s real advantage lay in the tacit knowledge of the workforce and the surrounding industrial ecosystem. Building large steel ships required thousands of subtle craft decisions—how to rivet steel plates, align hull sections under stress, and coordinate hundreds of specialized tradesmen.
Much of this knowledge was never written down. It lived in the habits of workers and the dense network of suppliers clustered around British ports.
Other nations could copy the machinery and the blueprints. Recreating decades of accumulated experience proved far harder. Shipbuilding leadership eventually shifted to Japan and South Korea but only after those countries—and their workers—spent decades climbing their own experience curves.
Microsoft Windows dominated the personal computer era not merely because of Microsoft itself, but because millions of developers already knew how to write Windows software. The same dynamic helped establish Linux as the dominant operating system of the modern data center.
In each case the moat extended beyond the company that created the platform. It lived in the broader community of engineers and organizations building on top of it.
So far artificial intelligence—at least as represented by NVDA—is repeating the pattern.
Nvidia’s ongoing dominance is far from assured. But if it is knocked off its perch, it will not necessarily be by AI platforms with the fastest processors, but those that attract the largest communities of developers and researchers sucking up knowledge no one can see.
For technology investors, this carries a crucial lesson: the deepest technological advantages are often invisible. The strongest moats often live not inside companies, but in the accumulated experience of the people who build on their platforms.