The Transformation Nobody Saw Coming

The Transformation Nobody Saw Coming

In the spring of 2016, Jensen Huang, the co-founder and chief executive of NVIDIA, walked onto a stage in San Jose and handed a small black box to a research lab at Stanford University. The box was roughly the size of a desktop computer, weighed about thirty pounds, and cost $129,000. Inside were eight graphics processors—the same kind of chips that, until very recently, had been used almost exclusively to render explosions in video games and to shade the polygons of fantasy worlds. The lab that received this machine was not a gaming company. It was a deep learning research group, and the machine was one of the very first "DGX-1" servers ever built. The new owner of the box was the Stanford Artificial Intelligence Laboratory, and the people who would operate it were about to begin training neural networks on a scale that had never been possible before.

Three years later, in the same city, a small research lab at OpenAI used a warehouse full of those same black boxes to train a language model called GPT-2. By 2022, the successor to that model—ChatGPT—reached one hundred million users faster than any consumer technology in human history. By 2024, every major technology company on earth, every national government, and every meaningful hedge fund had reoriented its strategy around the chips inside that little black box. NVIDIA, the company that had spent its first two decades selling graphics cards to teenagers, became the most strategically important company in the global technology supply chain.

This is the book you are about to read, and the puzzle at its center is the one we just described. How did this happen? Not just in the sense of "how did NVIDIA's stock go up"—though it did, by more than 200-fold in the decade after the DGX-1 launch—but in the deeper business sense of: how does a company reinvent its identity not once, not twice, but three times, while remaining the same legal entity, with the same ticker symbol, and with most of the same employees?

The answer, we will argue, is that NVIDIA's transformation is not really a single transformation at all. It is a series of three transformations, each one built on top of the last, and each one invisible to almost everyone at the time it happened. To understand why NVIDIA matters, you have to understand all three. And to understand all three, you need a mental model that the rest of this book will rely on throughout: the difference between a product company, a platform company, and an infrastructure company. Most readers—even sophisticated technology investors—regularly confuse these three categories, and that confusion is exactly what makes NVIDIA's story so hard to read at first.

This chapter is the foundation. We will do three things. First, we will introduce NVIDIA's three identities—graphics chipmaker, platform company, AI infrastructure provider—and show you why each one is simultaneously true. Second, we will walk through the thirty-year chronology of the company in five distinct acts, so that you can see how the present moment was constructed one decision at a time. Third, we will give you a reusable analytical framework—the three-lens framework—that you can apply to any technology company, including ones you already know. By the end of the chapter, you will be able to look at Apple, Amazon, Uber, or any other company and identify which of its three identities is doing the work of generating value. That ability is, in our view, the single most useful mental model a reader of business strategy can carry into the AI era.

Let's begin with the puzzle.

The Identity Puzzle: Three Companies Inside One

The most common mistake people make when they first encounter NVIDIA is to ask the wrong question. They ask, "What does NVIDIA sell?" The answer to that question is straightforward: NVIDIA sells graphics processing units, or GPUs, and the systems built around them. Once you have the answer, you think you understand the company. You do not. The reason is that "what does NVIDIA sell" is a product question, but the interesting questions about NVIDIA in 2024 are platform questions and infrastructure questions. The product is the visible surface. The platform and the infrastructure are the deeper layers that actually generate the value.

To see this clearly, imagine three concentric circles. The innermost circle is the product layer. This is the layer where a company sells specific, finished goods to specific customers. NVIDIA's product layer includes the GeForce RTX graphics card you can buy at Best Buy for $500 to $2,000, the Quadro card that professionals use for computer-aided design, and the discrete GPU inside a gaming laptop. These are products. They are physical objects. They have model numbers, they have prices, they have SKUs, and they have direct competitors from companies like AMD and Intel.

The next circle outward is the platform layer. A platform is not a product you buy. It is a foundation on which other people build products. NVIDIA's platform layer is called CUDA, which is a software toolkit that lets programmers write code that runs on NVIDIA GPUs. CUDA has been in development since 2006, and by 2024 it had been downloaded and used by more than four million developers. Every one of those developers made a choice—often an unconscious one—to write their software against NVIDIA's platform rather than against an alternative. Once they have made that choice, switching to a different platform becomes expensive, time-consuming, and risky. This is what platform strategists call "lock-in," and it is the source of most of the durable value in technology markets.

The outermost circle is the infrastructure layer. Infrastructure is what an entire industry depends on but no individual user notices. You do not think about the electrical grid when you flip a light switch. You do not think about the international payment rails when you swipe a credit card. You do not think about the cloud when you open a website. Infrastructure is the layer where, if the provider were to disappear overnight, the rest of the economy would break. In 2024, the AI industry had become quietly dependent on NVIDIA in exactly this way. The training of frontier AI models—models like GPT-4, Claude, Gemini, and Llama—required tens of thousands of NVIDIA GPUs working in concert. The companies that owned those GPUs—Microsoft, Google, Amazon, Meta, Oracle, and a handful of national AI initiatives—had made multi-year, multi-billion-dollar capital commitments to expand their NVIDIA-based compute capacity. This is not platform lock-in. This is infrastructure dependency. It is qualitatively different and quantitatively larger.

Here is the most important sentence in this chapter, and we want you to write it down. NVIDIA's surface business is selling chips. NVIDIA's platform business is selling developer mindshare. NVIDIA's infrastructure business is selling the physical foundation of the AI economy. These are three different businesses with three different competitive dynamics, three different customer relationships, and three different ways of generating value. When you read a headline that says "NVIDIA reports record revenue," that revenue is generated by the product layer, but the strategic significance of that revenue is determined by what is happening in the platform and infrastructure layers. When you read a headline that says "AMD releases a new AI chip," that is a product-layer threat, but its strategic significance depends on whether it can disrupt the platform and infrastructure layers—and it cannot, at least not yet.

Let us make this even more concrete with an example that does not involve NVIDIA. Take Apple, a company you almost certainly already know. The product layer of Apple is the iPhone, the Mac, the iPad, the Apple Watch, and the AirPods. These are the things you can walk into a store and buy. The platform layer of Apple is iOS, the App Store, and the