Nvidia's Growth Journey

The setup: a chip for playing games

Nvidia was founded in 1993 to make a specific kind of chip: the graphics processing unit, or GPU, which draws the images on a computer screen. Its first market was video games and PCs, where better graphics meant more realistic worlds, and gamers were willing to pay for the improvement. For its first decade, Nvidia was essentially a components company serving a demanding but narrow audience.

The technical detail that would matter later was hidden in what a GPU actually does. Drawing a complex three-dimensional scene means performing a huge number of similar calculations at the same time, one for every pixel and polygon. So GPUs were built to do many simple operations in parallel, which is a very different design from the general-purpose processor at the heart of a computer, which does a few complex things in sequence. Nobody set out to build a machine for artificial intelligence. They built a machine for drawing game worlds fast, and it happened to be good at parallel math.

None of this looked like destiny at the time. Nvidia was one of several graphics-chip makers fighting for a market tied to the ups and downs of PC gaming. It competed hard, won some rounds and lost others, and its fortunes rose and fell with each product cycle. The idea that this company would one day be among the most valuable in the world would have sounded absurd.

Nvidia's growth journey: the CUDA bet

Nvidia's growth journey turned on a decision that made little sense to many observers at the time: in 2006 it released CUDA, software that let programmers use its graphics chips for general computing, not just graphics. This is the pivot on which everything later depends, so it is worth understanding what it meant and why it was a gamble.

CUDA opened the GPU up. Before it, the chip's parallel-processing power was locked to graphics. After it, a scientist, an engineer, or a researcher could write ordinary programs that ran on the GPU and borrowed its ability to do thousands of calculations at once. Suddenly the gaming chip could accelerate weather models, physics simulations, financial calculations, anything that involved a lot of parallel math. Nvidia was turning a product into a platform.

The bet was expensive and unproven. Supporting CUDA meant devoting engineering resources and chip design to capabilities most of Nvidia's paying customers, gamers, did not need or want. For years it was not obvious the investment would pay off. The company was spending real money to make its chips useful for applications that barely existed yet, on the theory that general-purpose GPU computing would eventually matter. That is a platform bet: you build the general capability first and trust that valuable uses will find it. The strategic logic of building a platform others depend on is part of what identifying competitive advantages (moats) covers, because a platform with many users and much software built for it becomes very hard to displace.

What CUDA created, in the language of investing, was optionality. Nvidia did not know exactly which non-graphics use would become huge. It positioned itself to benefit from whichever one did. Optionality is the value of having many possible upsides you did not have to name in advance, and it is a large part of why the CUDA decision looks brilliant now.

Optionality is easy to admire after one of the options pays off and much harder to value before. At the time, CUDA was a cost with no guaranteed return, a bet that some unknown future application would justify the investment. Most such bets do not pay off, or pay off modestly. The ones that hit big, as CUDA eventually did with artificial intelligence, are visible only in hindsight, which is exactly why they are underappreciated in advance. A value investor who wants to give a company credit for optionality has to do so while the payoff is still uncertain, knowing that most options expire worthless.

The wave arrives: AI finds the GPU

The use that changed everything was one almost nobody at Nvidia specifically predicted: modern artificial intelligence. Around the early 2010s, researchers working on neural networks discovered that training them required exactly the kind of massive parallel computation that GPUs, made accessible by CUDA, were built to deliver. Work that would have taken a general-purpose processor an eternity could run on GPUs in a fraction of the time.

As AI research accelerated through the 2010s and then exploded later, demand for Nvidia's chips followed. The company that made graphics cards for gamers found itself selling the essential hardware for the biggest technology shift of the era. Because CUDA had been around for years, a whole generation of researchers had already learned to build on it, which deepened Nvidia's advantage: the software ecosystem was as much a moat as the chips themselves.

The honest way to describe this is that Nvidia rode a wave rather than predicted one. It built a general-purpose platform because it believed general GPU computing would matter, which was a good bet. It did not foresee that deep learning specifically would become the killer application and drive demand to extraordinary levels. The platform gave it the optionality; the AI boom happened to be the option that paid off spectacularly. The difference between riding a wave and predicting it is not a small one, and it matters enormously for how much credit to assign to foresight versus fortune. This is closely related to the humility that overconfidence bias demands: it is easy, after the fact, to believe someone saw the whole future coming.

The near-death moments hindsight erases

Nvidia's rise looks smooth in a chart, but the path was anything but, and remembering the near-death moments corrects the illusion of inevitability. Early in its history, Nvidia bet on the wrong technical approach for one of its first products and nearly ran out of money before recovering with a later chip. In 2008 it took a large charge over a defect in certain chips, and the stock fell hard. In 2018, a boom in using GPUs to mine cryptocurrency inflated demand, then collapsed, leaving the company with excess inventory and a sharp drop in its shares. The GPU business was cyclical and brutal, tied to product transitions and boom-bust demand.

Any of those moments could have gone worse. A company that nearly dies several times is not on a predestined path to greatness; it is surviving a series of genuine threats, some through skill and some through luck. The reason this matters for a value investor is that hindsight quietly deletes the moments when the outcome hung in the balance. Looking back, the survivors look inevitable and the near-misses vanish from the story. At the time, Nvidia's future was repeatedly in doubt.

This is why honest history refuses hagiography. The version of Nvidia's story that starts in gaming and marches triumphantly to AI leaves out every moment the company could have failed, and in doing so it teaches the wrong lesson. The right lesson includes the risk, the cyclicality, and the times the bet nearly broke the company.

Where it stands, and the hindsight caveat

Nvidia in July 2026 is worth roughly $4.74 trillion, according to Tenet data, among the most valuable companies on earth, on a share price around $196 after a stock split. The CUDA bet, the platform, and the AI wave together produced one of the most dramatic value creations in market history. It is a genuinely remarkable outcome, and the temptation is to treat it as proof of flawless foresight.

Resist that temptation, because it is the whole point of the article. Past results do not predict future returns, and this is doubly true for a company whose rise depended so heavily on a wave it rode rather than engineered. Nvidia made a smart platform bet that gave it optionality, survived several near-death moments, and then benefited enormously from a specific boom that arrived years later on its own schedule. Skill, survival, and luck are all in the story, and pulling them apart is the discipline. A business whose success rests partly on an external wave carries a risk that a chart of past triumphs hides: waves recede, and cyclical demand cuts both ways.

The reusable principle has two halves. First, a platform creates optionality, the chance to benefit from valuable uses you could not name in advance, and that optionality is worth paying attention to when you assess a business. Second, hindsight flatters, so separate riding a wave from predicting one, and be honest that some of any spectacular outcome is fortune. Both halves protect you: the first from missing genuine platform value, the second from mistaking a lucky, cyclical winner for a sure thing.

Where to go from here

Nvidia's journey pairs naturally with another company that reinvested through uncertainty, Amazon's evolution, and offers a sharp contrast to a technology leader whose moat proved provisional in the rise and fall of Nokia. To weigh the business against its price today rather than its history, open the live Nvidia report on Tenet and study the numbers with the hindsight caveat firmly in mind.