A Strange Thing

By the early 1990s, personal computers were already in almost every second home, yet the graphics on those machines were shockingly poor. Ordinary users - everyday gamers - still genuinely enjoyed it; for them this rough-looking imagery felt like some kind of unbelievable magic. But for professional engineers, the visuals were painful to look at.

For example, Wolfenstein 3D wasn’t truly 3D at all. It was “pseudo-3D”: enemies were flat, cardboard-like sprites running around in flat corridors. And all of that at a mere 20 FPS - if you were lucky. If too many enemies appeared on screen, the frame rate on a weak machine could drop to 10.

Why? Because the poor central processor had to do absolutely everything - the thinking, the calculations, and even the rendering of the image. Engineers Jensen Huang, Curtis Priem, and Chris Malachowsky noticed this problem. And suddenly it hit them: what if they created a dedicated graphics chip? Gamers would buy such graphics cards like hotcakes, and the three of them would be able to earn serious money.

They would easily outrun competitors like ATI, S3 Graphics, and Matrox, who were all selling basic 2D solutions. Those chips couldn’t handle any meaningful math, had no understanding of geometry, and simply drew whatever the CPU had already prepared for them. But proper 3D was different - real 3D required a whole “brain” for geometry. A chip capable of constructing worlds from building blocks entirely on its own.

The 1990s had already begun, and strangely, no one had yet created a true, universal 3D chip for regular PC users. That’s how Nvidia was born on April 5, 1993, in Santa Clara.

Where Will You Be in 10 Years?

But first, they needed investment. They didn’t have their own money. So the team approached the legendary Don Valentine from Sequoia Capital - the same man who had invested in Atari, Cisco, and Apple. In Silicon Valley he was known as a “Godfather-like” figure. The meeting, however, went very poorly. Valentine fired sharp, rapid-fire questions at them, as if from a machine gun, trying to see whether they were simply inexperienced dreamers. He dismantled their presentation completely. The team struggled to explain what exactly they were creating, who it was for, and how they planned to succeed.

“Where will you be in ten years? Who exactly are you? What about gaming consoles? Are you a graphics company? Are you an audio company?” Valentine asked. And indeed - nothing seemed clear. Their revolutionary chip combined so many functions at once: not only graphics, but also sound, game ports for consoles, and even network connectivity.

The presentation was a complete failure. However, the founders still had an excellent reputation in Silicon Valley as highly experienced top-level engineers. This saved them. Investors ultimately provided $2 million - one million from Sequoia and one from Sutter Hill Ventures. Valentine, however, warned them: “If you lose my money, I will be extremely disappointed.”

Jensen became the CEO. Valentine joined the board of directors and became something like a “shadow mentor” to him. He guided Jensen, teaching him how to survive in the world of high-stakes business.

The NV1 Embarrassment and Doom. 1995

So, the team bought equipment and began building their revolutionary new chip: the NV1. They didn’t want to create just another 3D chip - they aimed to build a true “hardware king” of the virtual world: highly advanced and packed with innovation.

But when they finally launched the NV1 in September 1995, the chip was completely overshadowed - even crushed - by a computer game: Doom II, a first-person shooter where the player travels across Mars fighting demons. Although the game was also technically only pseudo-3D - using 2D graphics rendered by the CPU instead of the video card - its creators had pushed the engine to its absolute limit. The kinetic visual effects and fast, intense battles were unlike anything gamers had seen before.

Doom II became a massive hit and sold millions of copies right as the NV1 hit the market. And suddenly every gamer rushed to stores to buy any video card - except one based on the NV1. Why pay $399 for an expensive NV1 when Doom ran beautifully on cheap $100 two-dimensional cards (VGA)? On the pricey NV1, performance was worse - the game stuttered heavily. As a result, Doom effectively destroyed the NV1’s reputation. Even worse, the very next year, in 1996, the revolutionary shooter Quake was released - this time truly 3D. But Quake was built entirely on triangular polygons, while the NV1 used quadratic surfaces. This made the NV1 almost completely incompatible with Quake: the game lagged, glitched, and textures distorted badly.

Quake also relied heavily on the CPU and could run even on basic 2D video cards. And if you wanted a real 3D experience, the obvious choice was the Voodoo Graphics card from 3dfx, which was optimized for triangles and delivered groundbreaking visuals. The NV1 itself wasn’t a bad piece of hardware at all - its unusual geometry let it run faster while requiring less RAM, which at the time was extremely expensive.

But game developers were in no hurry to adapt their titles for the NV1. For them, supporting NV1’s architecture was like creating a second version of the same game - a huge waste of time. So the chip ended up supporting only a few titles, and to run many games on the NV1, users had to rely on emulators. Performance was terrible.

The team had unintentionally created a chip so complex and unconventional that nobody needed it. Gamers wanted one thing: a card that was fast and affordable. Nothing more. So people bought the competitors’ products - especially the Voodoo Graphics card from 3dfx. On paper, it looked modest, but in real gaming conditions it completely outperformed Nvidia’s NV1. Customers simply returned NV1-based cards to stores.  It was a huge, spectacular failure - an attempt at brilliance that backfired entirely.

The Chaos of Standards (Direct3D)

And then, in 1996, things got even worse for Nvidia. Because the video card malfunctioned so frequently, characters in games could fall through the floor, teleport through walls, or simply cause the system to crash. Sometimes the chip even triggered the dreaded Windows Blue Screen of Death. Microsoft noticed how unstable graphics cards could disrupt the entire Windows operating system - and decided to intervene. They introduced a unified standard called DirectX for game developers.

Before that, the market was complete chaos. There were too many competing graphics standards, causing endless headaches for developers. Creating a PC game meant navigating a labyrinth of incompatible interfaces - practically seven circles of technical hell. Developers usually chose only one standard for a game; otherwise, production would take years.

The industry desperately needed someone to bring order. DirectX brilliantly solved the problem: a developer wrote code once, using Direct3D, and the game worked on any compatible video card. But there was one massive issue. Direct3D relied on 3D triangles - and the NV1 simply couldn’t interpret them. The NV1 was entirely incompatible with Direct3D. It required its own special driver and its own API. Even today, people still write angry comments online such as: “They never apologized for how terrible their four-sided architecture was!”

Stop Trying to Fix the Unfixable

By spring 1996, Nvidia’s distributor, Diamond Multimedia, returned almost all of the 250,000 ordered NV1 chips. It was a complete disaster. Nvidia was heading straight toward bankruptcy. Employees were demoralized. The atmosphere in the company was grim. Jensen felt ashamed and scared; every morning he woke up with a sense of anxiety. They could have shut everything down and walked away.  But Jensen wasn’t the type to give up. He finally realized something critical: innovation for the sake of innovation is meaningless. Compatibility matters far more. Ordinary people don’t care about clever engineering tricks if they can’t play their favorite games. You don’t need to be the smartest - you just need to give people what they want. He gathered everyone together and declared:

“Enough. We’re done polishing something that cannot be saved. From now on, we build chips using triangles, like everyone else. Forget our exotic geometry.”

Well, a Mistake Is Just a Mistake…

But Nvidia had already burned through about $15 million. And building a new chip meant redesigning the entire architecture from scratch - which required a lot of time and a lot of money. And financially, they were already deep in the red. How could they even think about a new chip when the company was practically a corpse?

They urgently needed more funding. But many investors now saw Nvidia as doomed. Why put money into a team that had failed so spectacularly? Yet Don Valentine refused to “cut the cord” and shut the company down. Instead, he simply said:

“Well, a mistake is just a mistake.”

Not because Nvidia had a strong team of engineers - although they certainly did. But because Valentine had one core investment rule: always invest in obsessed founders.

And Jensen was exactly that - a completely driven workaholic, fanatically devoted to his craft. He was persistent, unwilling to quit, ready to adapt instantly, learning fast, admitting mistakes, honest, straightforward. A whole combination of rare qualities that Valentine deeply valued. He believed in Jensen.

Jensen did not sugarcoat anything when speaking to investors. He openly admitted how bad things were: Sega had also canceled their orders; the architecture had been fundamentally flawed; the product was shockingly poor; and altogether they had flushed $15 million down the drain. He told them plainly:

“We tried to do too much and ended up doing many things poorly. It’s better to do fewer things well.” 

But now, finally, they had a clear and realistic plan. The PC graphics market was still young. There was still a chance to jump back onto the wave. Jensen swore that from now on he would learn everything he could about running a business so he would never repeat such catastrophic mistakes.

So Sequoia and several other investors agreed to add another $3 million. They decided to give the team one last chance. They too saw not the end - but a bottom from which Nvidia might push off.

The Ugly Giant (The Emulator)

But Valentine set one condition: from now on, Nvidia had to focus entirely on speed. Jensen laid off most of the staff. Out of roughly one hundred employees, only about thirty-five remained. Those who stayed worked overtime, practically living at the office - because they now had to build a new graphics processor under impossibly tight deadlines. Jensen himself worked from early morning until past midnight. Some employees even slept at the office.

To survive, they couldn’t create another overly complicated chip. They had to build something extremely powerful - as powerful as physically possible - so that it would surpass everything else on the market and defeat all competitors.

So the team began developing a chip with a record-breaking 128-bit bus, 3.5 million transistors, and the fastest pixel-generation pipeline in the world. But the additional $3 million from investors would last the company only nine months.

And then Jensen did something bold: he spent one million dollars on a chip-emulation machine - an enormous device the size of a refrigerator, covered in wires and circuit boards. It looked strange and almost comically ugly. This huge expense reduced the company’s remaining lifespan from nine months to just six.

But the emulator allowed them to test the chip without manufacturing it. They could build the chip virtually - in software - instead of in silicon. This let them skip the extremely expensive prototyping stage and go straight to mass production, saving an enormous amount of time.

But the move was incredibly risky. No semiconductor company had ever done anything like this. If the emulator made even one tiny mistake, that error would be “baked into” the chip, and an entire production batch would end up in the trash - a failure that would kill the company instantly. Some employees insisted they shouldn’t take such a risk - investors would surely provide more money if needed. And perhaps they would have…But Jensen was unshakeable:

“WE WILL NOT FIND ANY MORE MONEY!”

The emulator ran 24 hours a day. And incredibly, it cut a full year of development down to just three months. The team delivered the new chip, RIVA 128, exactly on the six-month deadline - though such work normally took nearly two years.

RIVA 128 (1997)

This became the world’s first fully emulated graphics chip - and it exploded onto the market. In 1997, people were stunned. According to performance tests, it outclassed the Voodoo and everything else available at the time. All competitors suddenly looked outdated next to Nvidia. 

It was powerful, compatible with everything, and games that previously crawled now came alive and ran like a storm. The RIVA 128 chip became wildly popular. Every serious gamer felt obligated to buy it. The most popular video cards based on the Riva chip were the Diamond Viper 330 and ELSA.

The chip was a true breakthrough for its era. It was one of the first unified architectures capable of rendering both 2D and 3D games - two-in-one. This was a dream come true for the average user’s wallet. Revenue surged; competitors were shocked.  In just four months, Nvidia sold over a million chips. Total revenue for 1997 reached $29 million.

After this success, Jensen invested even more money into emulation technology and abandoned physical prototypes altogether. But Nvidia’s position was still fragile. Anyone can succeed once - but sustaining success over several years is the real challenge.

30 Days. The Speed of Light. The Innovator’s Dilemma (1997)

Jensen realized something important: they had won not because they were inspired - but because they were desperate. They succeeded only because they were one step away from collapse.

So he announced to all employees that from now on, and forever, Nvidia must operate as if it were always in danger - as if bankruptcy were always just 30 days away. Even when the company had strong profits, they had to behave as though they were on the edge of failure. Otherwise, complacency would creep in, and Nvidia would slow down - and the competitors would immediately overtake them. Fear, on the other hand, sharpens focus. From that moment, Jensen repeated the same mantra constantly, everywhere, to everyone:

“The company is always 30 days from shutting down.”

Inside Nvidia, there was a culture of “do it or die.”

In the world of technology, one wrong decision or one failed product launch could be fatal. Luck would not last forever. Mistakes and downturns were inevitable. Jensen demanded that employees work at the speed of light. There could be only one limitation - the laws of physics. No delays. No pauses. No wasted time. Products had to reach the market faster than the competitors could even react. If you rested on your achievements, someone was already behind you, preparing to surpass you.

Around this time, Jensen discovered a book that would profoundly reshape his thinking: The Innovator’s Dilemma, published in 1997. He had promised Valentine that he would learn how to become a great business leader - and this book became his guide. From it, he realized that true threats come not from the top of the market but from below, where disruptive innovations begin.

Jensen loved this book so much that he talked about it constantly - morning to night. For employees, it became almost a daily topic. 

RIVA TNT and Quake. 1998

After this book rewired Jensen’s mind, he began probing the market very carefully, noticing things others missed. He noticed that the super-popular game Quake at the time couldn't handle multiplayer battles with a dozen participants in one arena. Most video cards weren't suitable for this. Jensen saw an opportunity in this. Whoever could render pixels faster would win the graphics wars. Optimizing a chip for this universally loved game would become a powerful advertisement among gamers.

And so, the very next year (June 1998), they rolled out the new chip Riva TNT, which was optimized specifically for this game by being able to calculate in parallel. It processed millions of operations simultaneously through a dual-channel mode. John Carmack, the creator of Quake, was thrilled. He advised his vast legion of gamer fans that Quake must be played only on Nvidia chips. Although the Riva TNT chip lagged and had glitches in other Direct X games, it was a monster for Quake! 

As a result, they generated $158 million in revenue for 1998. This was a fivefold growth compared to the $29 million a year earlier, thanks to this cleverly devised tactic. However, the profit margin was small, earning only $7 million. How could they continue development? To survive, Nvidia needed to grow faster than the competitiors, and to grow faster, they needed much more money. Therefore, in 1998, they decided to go public on the stock exchange (IPO) to raise the desperately needed funds.

Just then, many chips (RIVA 128ZX) produced by their outsourced manufacturer suddenly turned out to be defective and glitchy. More than half the chips had to be discarded. Revenue fell, and expenses rose. Nvidia was losing money at an alarming rate. Furthermore, a severe crisis was raging in Southeast Asia. Because of this, the underwriter, Morgan Stanley bank, rejected Nvidia's IPO - it was considered too risky.

Nvidia once again found itself on the brink of collapse. But Jensen found a way out: he turned to major customers who were purchasing millions in chips from Nvidia. They liked Nvidia's chips, so they loaned Jensen $11 million.

Fatal1ty and RIVA TNT2. 1999

In January 1999, Nvidia released the new chip RIVA TNT2. Although it ran hot, it was a performance monster. The legendary esports athlete Jonathan Wendel, known by the nickname Fatal1ty, played Quake on this chip for 8-10 hours a day. He used a video card with the Nvidia TNT2 chip that year because other cards gave a maximum of only 30 frames per second (FPS). But Nvidia’s new parallel technology delivered insane FPS - up to 60, and sometimes even up to 70 frames per second. This gave him an advantage in battles.

Nvidia supplied him with free equipment, and in return, he promoted Nvidia at every opportunity. He upgraded his card every six months, and soon every professional gamer began to do the same. Since then, he won dozens of tournaments and took first place seven years in a row. Just as Michael Jordan was for Nike, he became the face of the brand for Nvidia.

IPO (1999)

On January 22, 1999, Nvidia finally went public and raised $42 million from selling shares. Shares jumped by 64% on the first day. The capitalization reached $626 million. Wall Street analysts were highly impressed that Nvidia could roll out as many as two new video cards a year, which was twice as fast as any other competitor.

The IPO was not only for money but also for authority - so that everyone would consider Nvidia a solid company, not some small-time operation. Nvidia automatically received a “seal of trust”. This was crucial because major giants do not want to partner with small startups that might suddenly collapse tomorrow. Therefore, after the IPO, Nvidia rushed to partner with authoritative major companies like Microsoft, Dell, and Compaq. They even began supplying chips to Apple.

GeForce 256 GPU (1999)

In that same year, 1999, they rolled out the GeForce 256, which Jensen proclaimed was the first chip with a dedicated GPU (Graphics Processing Unit)! At the time, this sounded like a bombshell - as if a computer had been given a second brain. The chip with the GPU offloaded the CPU from the most resource-intensive tasks: geometry and lighting.

Everyone thought Jensen had completely lost his mind, saying, “It’s cool, sure, but why do you need it? The central processor handles this fine, anyway. Why do ordinary people need such enormous power?” Everyone thought this venture would simply devour all the money and bankrupt the company, as happened with the first NV1 chip. But Jensen believed the company would fail anyway if it didn't keep rolling out new things. This market demanded genius or death.

Just a year later, everyone realized that this feature significantly accelerated the computer's operation and relieved the CPU. It became a genuine breakthrough for games and 3D graphics, making NVIDIA the market leader. The chip was to games what the internal combustion engine was to cars - a catalyst for revolution. Games ceased to be just a picture; they became entire worlds.

In 1999, Nvidia pulled in $374 million, which was again more than double the previous year. Although competitors like ATI and Matrox were bringing in billions, they had no profit - they were operating at a loss. Nvidia was the only graphics company that was making a profit. This was because they were innovative, while competitors were simply repeating and copying.

Furthermore, Nvidia outsourced chip manufacturing to TSMC, while ATI and Matrox were pouring huge money into their own production for some reason. While competitors released only one chip, Nvidia released THREE! Competitors thought Nvidia had some secret magic allowing them to release chips so fast. In reality, everyone at Nvidia was simply working relentlessly. The first one to market reaps all the rewards. Competitors simply couldn't keep up with these technological speedsters who were doing the right things at the right time. GeForce 2 followed in 2000, and GeForce 3 in 2001.

3dfx

The relentless six-month hardware production cycle destroyed competitor 3dfx. They simply couldn't handle the race and eventually went bankrupt. The first one gets everything; the second gets nothing. 3dfx conceded defeat and offered to sell itself to Nvidia. But Jensen refused to buy 3dfx completely; he simply took on all their talented employees - about 100 people - for free.

When the 3dfx engineers came to Nvidia and saw the software code, they were simply shocked. As perfectionists, their own code was ideal, but Nvidia's was complete chaos. Everything was sloppy and disorganized. No one cared about the code. Such code was increasingly difficult to maintain, but this was Nvidia's defining characteristic - the need to rapidly iterate and execute, to be fast enough to survive! It was like a battle scar of a survivor.

Nvidia now employed about 600 people. The new employees were shocked by the non-stop deadlines and the constant feeling of running behind schedule. Jensen was demanding. This annoyed everyone, but it worked. People were working hard. Competitors tried to stuff the chip with a larger number of features, while Nvidia, without waiting for all the features, tried to release the chip exactly on time, and features that weren't finished were pushed to the next chip.

GeForce 3. Shaders. 2001

In February 2001, they rolled out the GeForce 3 model with programmable shaders, which gave game developers the ability to write their own rendering functions and visual effects, similar to what was seen in movies. This was once again a very expensive gamble because it required completely restructuring the chip architecture again. But this made the video card an absolute sensation.

If all games before were similar, somewhat plastic and rubbery, now the worlds in games suddenly came alive and became realistic. For example, Max Payne 2, which featured dynamic lighting, reflections on wet streets, and the shine of metal. And Nvidia's annual sales volume reached a whole billion dollars! It became the fastest-growing semiconductor company. Shares also went to the moon, becoming 20 times more expensive after the IPO. Jensen became a millionaire.

Microsoft. Xbox. 2001

In the same year, they also ventured into the gaming console Xbox from Bill Gates' Microsoft, thanks to the affection of game developers who told Microsoft they wanted an Nvidia chip. Nvidia’s shares immediately soared up to $100, as it meant Nvidia chips would be in every second American living room.

However, Nvidia ultimately earned almost nothing from this. Microsoft was selling the Xbox below cost to conquer the market. Jensen was furious, feeling he had been tricked. But on the upside, Nvidia chips became a mass standard. Every game developer now optimized their projects for the Nvidia GPU, creating an ecosystem beneficial to Nvidia. This was like "cost-free advertising": millions of players interacted with Nvidia graphics daily, meaning they didn't have to spend millions on advertising. The initial "horror" of the low margin turned into wings for growth.

The push for diversification led to Apple. In the early 2000s, Nvidia won a small contract to supply graphics chips for the iMac G4 computer. But they needed to sell Apple more chips. They showed a real-time rendering of a Pixar cartoon directly to Steve Jobs, and he was amazed. Jobs decided that the premium versions of the Power Mac G4 must have the GeForce 3 chip on board. They also began supplying chips for Apple laptops in 2003.

The Golden Age of Computer Games

Next came the golden age of computer games, and PCs spread incredibly fast in those years. PC gamers were the best "addict" customers for Nvidia. They spent even more than console clients. They custom-built their systems and upgraded their hardware so that every new game delivered a slightly smoother picture. They overclocked their motherboards and boasted about who had the faster rendering. Therefore, Nvidia achieved record revenues every year. The company then employed already a thousand staff. Jensen immediately invested all the money earned into new risky technologies.

THE MOST INTERESTING CHAPTER. 2003

One day, scientists from Stanford managed to bypass the GeForce 6 video card to use its GPU to calculate protein folding. They were able to do this thanks to the programmable shaders, which served as a kind of loophole. Jensen noticed this and realized it could open up unlimited possibilities in new fields, rather than just releasing video cards for the rest of his life. Simple hardware improvements in video cards offered no good long-term prospects - due to Moore's Law, they were bound to hit a ceiling at some point. In other words, there was nothing to gain in hardware alone; they had to think bigger. Jensen didn't want to be forever stuck in the gaming market and inevitably lose leadership.

These pioneering scientists were the key he could seize. Jensen began luring such scientists to work at Nvidia on a secret project called CUDA. The goal was to retool computations from video games to science. Jensen took an unimaginable step: he didn't create a separate, special video card for this. Instead, he implemented this technology into all gamer video cards so that CUDA was available to everyone. This was to saturate the entire market with the new technology, make it the standard, and ultimately distinguish themselves from competitors.

The decision to create dual-purpose chips sparked controversy within Nvidia, as it increased the production cost of video cards. Gamers didn't even realize they were sponsoring scientists. This additional cost was called the CUDA tax.

The Era of Doubt. 2006-2012

The first G80 chip with CUDA was rolled out in 2006. The CUDA library itself was made open to all - it was a software package with a set of functions for scientists' needs. In short, CUDA became a way to access the machine.

The unimaginable happened: CUDA technology could now be enabled on ordinary GeForce cards. Because of this, even simple, poorly funded scientists could perform work that researchers with multi-million dollar supercomputers were doing. Before this, computational power was available only to a few. Supercomputers cost astronomical amounts of money and were only found in elite universities. This severely slowed scientific progress in many fields and was a bottleneck.

Now, scientists from various fields began gradually buying more and more GeForce cards for experiments completely unrelated to graphics. These included simulating galaxy formation, weather phenomena, the ignition process of a nuclear bomb, or even determining stock quotes. CUDA technology became their salvation. Even young graduates could immediately conduct their experiments and scientific research without time limits. A student could put four $500 GeForce gaming cards into a computer and get a workstation comparable to an entire cabinet of servers. This was a revolution! 

But CUDA also brought problems - gamers complained about video cards in laptops that stopped working after a few weeks. Nvidia's shares fell again, losing almost 90% of their value (the second time in six years). Jensen allocated $200 million to compensate customers for the losses. For the first time since the IPO, Nvidia lost money. There was no profit. Critics accused Nvidia of wasting time on useless pursuits.

But Jensen was unstoppable. In 2007, they released a complete video card with the G80 chip specifically for scientists - the Nvidia Tesla - for tasks like simulating protein molecule folding, DNA sequencing, weather modeling, and financial market analysis. Investors, however, were skeptical of this idea. This move turned out to be extremely expensive. The development of the G80 took 4 years and cost a staggering $475 million in various research and development. Consequently, profit declined from 2008 to 2010.

Wall Street considered CUDA technology useless. Analysts and investors believed Nvidia had lost its way. They saw only one thing: that the new direction was negatively affecting the profit margin.

But Jensen didn't give up. On the contrary, they even began supplying video cards to universities. He made financial donations so that universities would use Nvidia equipment in graphics programming classes. Schools were offered machines with CUDA if they taught courses on the subject. Employees visited universities, telling students and professors that the approach to computer science education needed to change, as parallel computing would become much more important in the future. However, there weren't many takers. No one wanted to listen about CUDA.

But Jensen, stubborn as a tank, continued to push in this direction. Lectures on GPU computing were recorded and posted online for public access. They established strong relationships with scientists from absolutely different fields. They held technological summits where Nvidia employees interacted with scientists and exchanged feedback. Then they added the features requested by these scientists to the chips. Scientists were literally guided step-by-step on how to use their technologies. They even released a book in 2010 about CUDA and parallel computing. Jensen lured many scientists to work at Nvidia, who then openly shared their scientific publications, which served as a kind of advertisement that attracted even more other scientists. But this was a drop in the ocean.

Despite all efforts, interest in CUDA was not growing; in fact, it was declining, reaching a minimum by 2012. The situation became dire. The problem was that the broader market had no tasks for CUDA. Jensen was bringing supercomputers to the masses, but the masses didn't want it.

Investors were celebrating, believing their predictions had come true and CUDA had failed. “They spent a fortune on this new chip architecture,” said Ben Gilbert, host of the popular Silicon Valley podcast Acquired. “They were spending many billions, targeting a little-known corner of academic and scientific computing, which was not a big market at the time”.

It reached the point where the audacious investor Jeff Smith from Starboard Value even wanted to force his way onto the board of directors to remove Jensen and stop the investments in CUDA. Fortune magazine called him “the most feared man in corporate America” because he was adept at penetrating boards of directors and ousting any CEO. Smith believed Jensen had made too many mistakes. In 2012, he even visited the headquarters and urged Jensen to finally focus. The air was thick with a hidden struggle for control of Nvidia.

But even then, Jensen refused to abandon CUDA. As a true fan of The Innovator's Dilemma, he knew he needed to cast a line into other areas. On the contrary, he bet everything and even rolled out the NVIDIA Tesla K20, which was only needed by a handful of scientists at Oak Ridge National Laboratory for their most powerful supercomputer in the world - Titan. The decision to double down on CUDA in the face of Jeff Smith was the riskiest. Smith viewed this as an ultimate failure because the costs for the K20 were astronomical, with no apparent benefit. Unlike gamers, supercomputer customers were volatile and always short on cash.

Semiconductor companies also considered CUDA an imprudent direction. It was a bet that made Jensen who he was, a gamble that set him apart from the crowd. Jensen resisted to the last, doing everything possible to persuade investors, but they were all set against him. Even discontent began to surface on the board of directors. Tench Cox thought they had lost their way. Daud Hudson, who was also on the board, considered Nvidia a stagnant company with a poor reputation. Jensen was viewed as insane. Why complicate perfectly good chips? It’s madness!  Why should they venture into some scientific computations? It’s completely absurd! 

What Christensen had predicted came true: "investors will resist." This was the real secret of The Innovator's Dilemma that readers overlooked. The book was not about how to succeed; it was about how not to fail. Christensen's book was not a guide for aspiring entrepreneurs but a counter-insurgency manual for senior managers of stagnant firms. Thirty years later, Jensen felt that Nvidia risked becoming just such a stagnant firm, and it was paranoia, not optimism, that prompted him to enter the market of experimental science.

Despite the total resistance, Jensen felt a breakthrough was approaching. He sensed it in the frantic enthusiasm in the eyes of his employees. He felt it so strongly that he was ready to reduce his profit to zero and jeopardize his main product. He felt it enough to risk his job. He felt that somewhere, he wasn't sure exactly where, but definitely somewhere, there was a revolutionary scientist who would soon validate the ideas of CUDA and start a revolution.

2012. The Turning Point (AlexNet)

And in 2012, that moment finally arrived! Exactly what Clayton Christensen spoke about in his book happened: breakthrough technologies often emerge in communities of amateurs who use components for unintended purposes.

The pioneering scientists wer Using just two Nvidia GTX 580 chips, they rolled out the neural network AlexNet, which could super-accurately determine what was shown in a picture. This was a revolution. Until then, neural networks, which mimic the structure of a living brain, were in deep disfavor among researchers. For many, the attempt to create AI was like trying to find Bigfoot. Starting in the 1950s, AI had repeatedly faced hype cycles that ended in embarrassing failures each time.

But now a turning point had occurred, and everyone realized that incredible computational power was needed to train neural networks. And it turned out that Nvidia video cards were perfectly suited for this, training neural networks hundreds of times faster than a regular central processor because they work in parallel. This is their main feature. The CPU is a sniper rifle, and the GPU is a 258-barrel shotgun.

When Jensen learned about these scientists, he was immediately intrigued by neural networks. He frantically began studying everything about AI. He had been sailing, seemingly nowhere, for so many years, and suddenly, among this endless void, a priceless treasure miraculously appeared. The more he studied AI, the more his excitement grew. He instantly had an epiphany. By mid-2013, Jensen was vibrating with intense energy. The entire puzzle that had been building itself blindly suddenly came together in an instant, and a marvelous picture for world domination presented itself to Jensen. Computer vision was just one direction in AI; there could be thousands of such directions. It became clear that neural networks would inevitably revolutionize society.

Some old employees thought they shouldn't enter the AI field, as it was just a fleeting fad. But Jensen then announced to everyone that from now on, Nvidia was no longer a graphics company; it was now an AI company. And CUDA began to be adapted for AI.

cuDNN. 2013

Events began to unfold rapidly. In 2013, Nvidia started creating the cuDNN function library purely for neural networks within CUDA. This was now a complete ecosystem for AI professionals. And in 2014, cuDNN was rolled out to the market. They even optimized the hardware itself for AI. Matrix Multi-Engines, created purely for deep learning, were added to the chips. These accelerated deep learning three times faster than regular CUDA GPUs.

Jensen was intensely focused on AI and spoke only about it. He was over 50, but he began working even more. In addition to Google, other giants like Amazon, Microsoft, and Oracle were massively buying chips for their cloud services to sell computational power as a utility. And CUDA users didn't even have to buy their own hardware from Nvidia.

Large companies began massively integrating AI into their products. And investors rushed to throw money into AI startups. In 5 years, investments grew from zero to $5 billion by 2015. All these startups were running on the Nvidia platform.

In 2016, they rolled out the supercomputer DGX-1 - the most powerful one Nvidia had ever created, tailored purely for deep AI learning. It was priced at $129,000 and included 8 NVIDIA Tesla P100 GPUs. Elon Musk was the first to receive such a computer for his AI startup, OpenAI. The computer weighed so much that a dolly was needed to wheel it into the office.

The Mining Fever. 2017

In 2017, cryptocurrency miners suddenly appeared. Bitcoin was growing wildly. The entire planet suddenly started mining with video cards. And mining turned into a craze. GeForce cards doubled in price, and all warehouses were emptied. There were even super-miners who placed blocks of GeForce cards in entire cabinets and racks, in bedrooms and garages. It looked like madness.

The most interesting thing was that CUDA was perfectly suited for the mining algorithm. The CUDA package was downloaded 2.7 million times in 2017. That was almost double the year before, and 15 times more than in 2012.

The craze peaked in 2021, when video card prices increased fourfold. Gamers were furious. The top gaming video card was the NVIDIA GeForce RTX 3090, whose price tripled to $4,000.

By the end of 2021, NVIDIA's revenue reached a record $16.68 billion (more than a fourfold increase compared to 2017), and net profit was $4.33 billion. The company's capitalization at its peak in November 2021 exceeded $800 billion. By this time, NVIDIA was firmly associated not only with gaming but also with artificial intelligence and data centers.

ChatGPT. 2022

And when OpenAI, the same company that was buying supercomputers from Nvidia, suddenly rolled out the chatbot ChatGPT in 2022, it caused an immense press frenzy. Everyone was amazed that a program could compose poems, song lyrics, and food recipes. When people encountered ChatGPT, their minds were blown because the computer communicated with them like a rational being. It was like meeting an alien. Suddenly, the whole world became obsessed with generative AI.

The number of ChatGPT users skyrocketed to 100 million active users in just two months! And the most interesting thing is that training these "aliens" required unreal computational power. Everyone suddenly remembered that only Nvidia chips could handle such a load.

Demand for the H100 chip soared. Major tech companies lined up to get the coveted chips. Prices for them reached tens of thousands of dollars per unit. Nvidia couldn't produce as many as everyone wanted.

In March 2023, OpenAI introduced GPT-4, which was able to pass exams for law, art history, US history, biology, and statistics. It passed every exam. For example, you could send it your lab results, and it knows how to treat you even better than a real doctor. It can create a natal chart, check the text of a contract to ensure a contractor doesn't cheat you, or even write lyrics for a rap hit. People were surprised, amused, and even scared. It was magic because this bot could perform any job for you. It even reached the point where AI started using AI to accelerate the work of AI. That is, AI was used to speed up the work of matrix multiplication, which helps AI think faster.

In May 2023, Nvidia's earnings report was published, and everyone learned that the company expected an astonishing output of $11 billion for the second quarter. Such figures were simply anomalous. Where did all this money come from? Nvidia was forecasting the largest revenue growth in the entire history of its industry. Even Google never had such positive reports. Shares suddenly, sharply, started to jump up wildly, rising by 24%. Because of this, the capitalization swelled by $184 billion! This single increase was more money than the entire value of Intel combined! Someone on Wall Street said, “There’s a war going on in AI, and Nvidia is the only arms dealer.”

Jensen added fuel to the fire and spoke at a technology conference that week, where he announced the new super-mega-ultra-powerful AI computer: the Nvidia DGX GH200 AI. It had as many as 256 graphics processors - a computer purely for AI, so that various chatbots could easily churn out their answers to millions of people worldwide. Jensen was offering incredible power at a low price. “The more you buy, the more you save,” Jensen said.

This was a hugely successful move. Everyone rushed to stock up on AI hardware because there wasn't a single company left in the world that wasn't trying to implement AI. AI isn't just about some chatbots entertaining housewives; it also includes speech recognition, video analysis, and computer vision - the possibilities are simply immense. And now, all these high-tech gadgets will transform humanity beyond recognition.

The demand for AI computational power is growing faster and faster as the range of problems that AI can solve becomes wider and wider! Even empty factories with only robots have started to appear. In the first quarter of 2024, Nvidia’s data center business grew by 427% compared to the previous year. All this is due precisely to the demand for artificial intelligence chips.

Therefore, to ordinary people from the outside, such growth seems like a sudden miracle or a bubble. But in reality, Jensen had been clearing the ground for this for many years while everyone else was doing nothing. He didn't capture the market; he created it! That's why Nvidia's revenues and stock are growing wildly.

Nvidia's products for AI are also very expensive. For example, the Blackwell GB 200 server rack, designed specifically for training AI models and equipped with 72 GPUs, costs from $2 to $3 million. This is the most expensive machine Nvidia has ever created.

THEREFORE, NVIDIA IS RAKING IN MONEY BY THE CONTAINERLOAD

The market position allowed Nvidia to set high prices for hardware. This attracted competitors. AMD and Intel offered open-source alternatives to CUDA. But few AI researchers used their products. Nvidia had better designs, better software, and a better ecosystem.

Between 2012 and 2022, Nvidia achieved a thousandfold increase in performance. Of this, a 2.5x acceleration was achieved through improvements in transistor technology, and the remaining 400x through Nvidia's mathematical tools. Competitors made silicon that was just as good, but they couldn't accelerate the computations themselves.

In addition, Nvidia created specialized tools for specialized programmers. These were for automotive research, drug discovery, medical imaging, cybersecurity, and even for recording fatal shots in Fortnite. By 2022, Nvidia offered over 400 such kits in its NVIDIA NGC catalog and through SDKs, covering gaming, animation, climatology, planetary science, mathematics, physics, finance, biochemistry, and quantum computing. These software packages were freely available to anyone. Jensen called these kits his treasures, as they engaged researchers of all stripes with Nvidia's equipment.

That's why Nvidia is hard to compete with. As soon as a new area emerged, Nvidia was already there. The key was to be first. Later, competitors appeared in this field with more elegant tools, but it was already too late. The industry standard had been set. Nvidia was always a couple of generations ahead of everyone. This was a lesson Jensen learned back in 1995 when he lost the standards competition with the NV1 chip. Now, thanks to that lesson, he is invincible! 

Work Is Rest

In 2024, there were about 5 million CUDA developers, 600 AI models, and 3,700 applications. Nvidia is now a true giant in the field of AI. Jensen believes that praise is harmful, as it distracts. On the contrary, a dressing-down is necessary so that the company doesn't fall into complacency. “Every morning I look in the mirror and say - you are terrible,” Jensen once said at a meeting. The most important thing in the company is endless self-improvement. One cannot be proud of the past. One must concentrate on the future. One must simply do one's job.

Jensen says that work is a rest for him. He works even on vacation. He doesn't even remember the movies he watches on weekends because even then he continues to think only about work. He works every day. He works even in his sleep. He dislikes those who work little. He is constantly learning new things. And when he is called a perfectionist in hopes of insulting him, he is happy, because that is how it should be! After all, if you want to do something extraordinary, it shouldn't be easy!

Key business lessons:

1. Embrace Failure as a Critical Feedback Mechanism 

• Failure to Launch (The NV1 Chip): NVIDIA's first revolutionary product, the NV1, was a "spectacular failure." It was technically advanced but failed because it was incompatible with emerging industry standards (Direct3D triangles) and lacked market alignment (gamers wanted fast and affordable, not complex and proprietary).
• Lesson: Prioritize Market Standards over Proprietary Brilliance. Innovation is meaningless if your product doesn't easily integrate into the established ecosystem. As Jensen Huang realized, "Innovation for the sake of innovation is meaningless. Compatibility matters far more."
• Admit and Course Correct Immediately: After losing $15 million, Jensen was transparent with investors: "We tried to do too much and ended up doing many things poorly. It’s better to do fewer things well."
• Lesson: Own Your Mistakes. True resilience isn't avoiding failure; it's honestly assessing a catastrophic mistake and making a radical pivot, even if it requires laying off staff and rebuilding the entire architecture from scratch.
• 
  

2. Execute with Unrelenting Speed and Urgency

• The "30 Days from Bankruptcy" Culture: Jensen institutionalized a culture of paranoia and extreme urgency, operating as if the company was always on the verge of failure. This mindset, born from desperation, became their greatest strength.
• Lesson: Cultivate High-Pressure Execution. Demand that your team operates at the "speed of light." In a fast-moving technological market, speed-to-market often beats a marginally superior product. "The first one to market reaps all the rewards."
• Process Innovation to Gain an Edge: They invested $1 million (a huge risk at the time) in an emulator machine to test chips virtually. This process innovation allowed them to release the RIVA 128 in three months - a process that normally took an year.
• Lesson: Innovate Your Process as Much as Your Product. Look for high-risk, high-reward strategic bets that radically compress your development cycle and give you an insurmountable speed advantage over competitors.

3. Play the Long Game by Creating the Next Market (The Innovator's Dilemma)

• The CUDA Bet: Following the principles of The Innovator's Dilemma, Jensen made an expensive, decade-long bet on an unproven market (parallel computing/scientific research) while his core business was profitable. This bet - implementing CUDA on gaming GPUs - was controversial, tanked profits, and almost cost him his job.
• Lesson: Anticipate Disruption from Below. Don't be complacent in your current profitable market. Invest substantial resources into disruptive, adjacent technologies that seem niche or unprofitable today (the "CUDA tax") but could become the next foundational layer of the economy tomorrow.
• The Strategic Pivot to AI: After the AlexNet breakthrough, Jensen immediately declared, "Nvidia was no longer a graphics company; it was now an AI company." This clarity of purpose allowed them to reallocate all resources to the future.
• Lesson: Be Ready to Change Your Identity. Don't let your past success define your future. When a major technological shift occurs, execute a radical pivot and dedicate all company focus to the emerging wave.

4. Build an Ecosystem, Not a Component

• Build the Software Bridge: The true breakthrough wasn't just the GPU hardware; it was the CUDA software ecosystem. By making the CUDA library open and providing over 400 specialized software development kits (SDKs), they made their hardware the only viable platform for researchers and developers across diverse fields (AI, drug discovery, finance).
• Lesson: The Ecosystem Trumps Hardware Alone. Superior software tools, libraries, and an open, thriving developer community (5 million CUDA developers by 2024) create massive lock-in that is nearly impossible for competitors (like AMD or Intel) to overcome, even with comparable silicon.
• Set the Industry Standard: By being the first to deeply integrate with new fields, they set the industry standard for parallel computation and AI.
• Lesson: Being First to an Ecosystem Guarantees Longevity. Being first to define the rules means you are always "a couple of generations ahead of everyone," making your position invincible to latecomers.

5. Founder Obsession Sets the Company’s Destiny

• Obsession is Contagious: Investors backed Jensen not just because of his engineering skills, but because he was an obsessed founder who was persistent, honest, and willing to adapt.
• Lesson: Be Fanatically Driven. The founder's unrelenting work ethic and commitment - working long hours, demanding perfection - set the tone for the entire company's survival and success. If you want to do something extraordinary, it should not be easy.
• Constant Self-Criticism: Jensen believes praise is harmful and self-criticism is necessary to prevent complacency.
• Lesson: Fight Complacency. Maintain a state of endless self-improvement. Never "rest on your achievements," as the competitor is always behind you, preparing to surpass you.

Connection to Islamic Finance and Entrepreneurship

The NVIDIA story offers several powerful lessons that resonate deeply with the core principles of Islamic Finance and Entrepreneurship. While NVIDIA operates in a conventional financial system, its strategic decisions and ethical founder behavior align remarkably well with Shariah-compliant concepts focusing on risk-sharing, moral purpose, and diligence.

Here are the key lessons from NVIDIA's success through the lens of Islamic finance and entrepreneurship:

1. Financing Based on Risk-Sharing (Mudarabah) 

The funding structure that built NVIDIA strongly reflects the Islamic finance principles of Profit-and-Loss Sharing (PLS), which offer an ethical alternative to interest-based conventional lending (Riba). The initial $2 million investment into the company functions like a Mudarabah contract, where the investors acted as the Rabb-al-Maal (capital provider) and the founders (Jensen Huang and his partners) acted as the Mudarib (entrepreneur/manager). Crucially, when their first product, the NV1 chip, failed, the investors bore the financial loss of the $15 million burned capital. The entrepreneurs lost their time and effort, but the financier bore the monetary risk, adhering perfectly to the Mudarabah rule. Furthermore, investor Don Valentine's continued support. This approach fostered a collaborative environment where both parties shared ongoing risks and rewards based on true partnership, rather than a fixed-return debt agreement. NVIDIA’s reliance on equity capital, which requires partners to genuinely share in the ultimate risk and reward, aligns directly with the overall principle of risk-Sharing fundamental to Islamic finance.

2. Pursuing Public Welfare and Purpose (Maslaha)

The trajectory of a successful Islamic entrepreneur is not solely focused on profit maximization but also on contributing positively to society, a concept known as Maslaha (public interest). NVIDIA's controversial, long-term bet on the CUDA architecture embodies this principle. Jensen Huang consciously diverted hundreds of millions in resources to shift the GPU's function from mere graphics to scientific and parallel computing. This was a decision that risked short-term profitability - incurring a "CUDA tax" on their successful gaming cards and "tanking profits" - for a massive, long-term societal good. The ultimate reward was making supercomputing accessible for global AI and scientific discovery, aligning with the ethical imperative to prioritize collective benefit over short-term material gain. By making the CUDA library and SDKs open and widely available to researchers, NVIDIA fulfilled a communal obligation by actively building an ecosystem that empowers others, reflecting the community-building aspect of Islamic entrepreneurship.

3. Excellence, Honesty, and Trust (Ihsan and Amanah)

The ethical behavior and operational culture of the founder are cornerstones of Islamic business. Jensen Huang’s relentless work ethic, where he states that "work is a rest for him" and his philosophy of "endless self-improvement," perfectly aligns with Ihsan (perfection or excellence). This concept demands the highest level of diligence and sincerity in one's work. Furthermore, after the NV1 failure, Huang was transparent and honest with his investors, directly admitting the mistakes and the severity of the financial situation. This built deep trust (Amanah or trustworthiness) and truthfulness (Sidq) with his financiers, a foundational requirement for any Shariah-compliant relationship.

4. Reward begets risk (Al-Kharaj bi al-daman). 

It is an Islamic maxim which states that no one can expect to succeed in their endeavours without taking on some level of risk or loss. The relationship between Reward and Risk is an absolute in entrepreneurship, and the company's extraordinary success perfectly validates this Islamic maxim. NVIDIA’s journey illustrates this principle across three distinct phases of escalating risk, each unlocking a proportional level of success. First, there was the Risk of Catastrophic Failure with the NV1 chip, where the company nearly went bankrupt after losing $15 million in pursuit of market domination. The resulting reward was not immediate profit, but the foundational lesson that "Compatibility matters far more", forging a culture of urgency that led to their next successful product. Second, the Risk of Strategic Cannibalization involved the decade-long bet on CUDA, which created a controversial "CUDA tax" that "tanked profits" by sacrificing the profitable gaming core for an unproven scientific market. This massive near-term risk yielded the revolutionary reward of being the foundational layer of the global AI economy, granting NVIDIA an "invincible" industry-standard position when Deep Learning emerged. Finally, the Risk of Personal Obsession refers to CEO Jensen Huang’s commitment, which demands the personal risk of total commitment and a relentless pursuit of "endless self-improvement". The reward for accepting this profound personal cost is a culture of demanding excellence that ensures the company is constantly "a couple of generations ahead of everyone," thereby maintaining its technological and market dominance.

In summary, the NVIDIA story confirms that in business, the path to groundbreaking rewards is paved with the courage to embrace and manage potentially company-ending risks. The degree of reward is a direct reflection of the severity of the risks successfully overcome.



Source of the original story: https://companystories.ru/

Adapted for a Muslim audience by Zakat App.