Nvidia Has a Secret

To understand Nvidia, you've got to understand its mercurial founder. Jensen Huang was born in southern Taiwan in February of 1963. When he was four years old, his father traveled to New York City for work. Jensen's father realized right away that the family would have a better future in America. Soon after Jensen's father returned home, his mother bought an English dictionary. Every day she would pick 10 words and grill Jensen and his older brother. Five years later, the family decided the time was right. It was time to move. But. But they faced the challenge that confronts almost every prospective migrant. They didn't have the money. Jensen's parents were desperate to find him and his brother a good school, even if it meant delaying their own move. They finally found one that sounded like it fit the bill. Oneida Baptist College in Eastern Kentucky. Jensen would later say that he and his brother were the first foreigners to attend this school and possibly the first people of Chinese origin to ever set foot in the town of Oneida. It turns out that there was a good reason why this school was so cheap. It wasn't the prep school that Jensen's parents were hoping to send their kids to. It was a reform school for troubled kids. When Jensen arrived there as a 9 year old, his roommate was a 17 year old who'd been just released from prison and was actually recovering from some stab wounds. It could have ended very badly, but Jensen showed an early glimmer of the character trait that would come to define him. Where others saw adversity and he saw opportunity. Jensen helped his roommate with math. In turn, his roommate introduced Jensen to weightlifting. Jensen spent a couple of years at Oneida until his parents were finally able to save up enough money to join him in the United States. After a little time in Tacoma, Washington, the family settled in Portland. Jensen skipped two grades in high school and went to Oregon State for his undergrad. He loved math, so it was only natural that he chose to study electrical engineering. After graduating, he moved to Silicon Valley together with Lori, his wife and former Oregon State lab partner. It was the mid-1980s, the era of top Gun and Wall Street. It felt like anything was possible. Jensen joined AMD to work on designing new chips. It was an extremely interesting stage of the company's life cycle. In 1985, AMD made the Fortune 500 list for the first time. But by later that year, the American chip market was suffering a downturn. This was partly due to trade pressures from Japan, but it was also because the American market for chips was bloated and stagnant. Inspiration was in short supply. AMD rode out the crisis by innovating aggressively, which included devising the Liberty chip program, a commitment to design or manufacture a new chip or chipset every week. Jensen watched closely and learned a lesson that would stay with him for the rest of his career. If you're not looking for ways to innovate and expand your mission, you're falling behind. Working at AMD in the 1980s meant working with the fastest chips in the world. One of Jensen's early projects was working on a super fast 1 MHz CPU chip. For comparison, the Ryzen 7, AMD's top of the line CPU today has a clock speed of 3.4 gigahertz, more than 3,000 times faster. He quickly realized a couple of things. First, designing chips was really hard. There was good reason why at the time, only intel and AMD could apply the financial and intellectual firepower to make it happen at scale. And this led directly to Jensen's second realization. It wasn't as cool as he had imagined. At the time, designing chips was a highly manual process and leading companies all used their own internal full stack methods. So he began thinking about how to make the process more cool and cutting edge. Maybe he didn't fully grasp it at the time, but coolness was actually a great proxy for other important things like a company's speed of manufacturing and ease of replication and ability to attract talent. Jensen began to believe that one way to make designing chips cooler was to encourage specialization. That would let them develop better tools and methods, thereby improving the quality of the final products. That thought, which was already percolating, would become even more urgent for Jensen when a colleague left AMD for a startup called LSI Logic. LSI had just gone public and made Sequoia Capital $135 million, which which was at the time the biggest ever one day return on an IPO. It was one of the first companies making application specific integrated circuits, or ASICs. An ASIC is a chip designed for a specific technical function rather than general purpose use. General purpose chipsets traded off power and efficiency for versatility. That made sense for home and enterprise computing. But for companies that knew exactly what they needed their chips to do and needed to extract as much power as possible, it wasn't an ideal solution. More specific technical functions meant that chips could be integrated into niche industry systems. This enabled companies like LSI Logic to custom design chips without needing to own the full stack. Jensen realized this aligned with his own thinking about how chip design could be improved. So when LSI presented him with the opportunity to join. He jumped at it. No sooner than Jensen had got his feet under his desk, he he was assigned one of LSI's biggest customers, Sun Microsystems. He helped build chips for what became their first major workstation product, the SparkStation one. This collaboration would define Jensen's first few years at LSI. He worked directly with Sun's co founders Andy Bechtelsheim and with Vinod Khosla. Today Bechtelsheim and Khosla are legendary venture capitalists. Back then they were simply world class engineers and businessmen. Jensen built a reputation as a rising star. He was creative, charismatic and a great builder. Let's fast Forward to late 1992. Jensen had just finished his master's. Eight years after he started it. He was still at LSI. Then he was approached by two sun engineers, Chris Malachowski and Curtis Prime. After years of working with Jensen, they trusted him. They told him that they weren't loving life at sun and asked if he had the time to discuss an idea. They met at one of Jensen's favorite spots. Denny's demand for 3D graphics is increasing. Chris and Curtis began. We are great chip engineers. You're a great chip product manager. How about we make a chip that can enable a PC to handle 3D graphics by taking some of the workload off the CPU? There was definitely a bull case for the idea they were pitching. All the early evidence suggested that consumers loved Advanced Graphics. Wolfenstein 3D was released in May of 1992. By 1995 it had sold a quarter of a million copies. Doom came out in 1993 and and took the world by storm. Jurassic park was one of the first Hollywood movies that made extensive use of 3D graphics. It made a billion dollars at the box office. But there were lots of risks. The technology was very new and very expensive. You weren't only making a bet on consumers wanting games with 3D graphics. You were also making a bet on gaming becoming a lucrative industry, which if you were making chips, wasn't something that you had all that much control over. It was risky, but Jensen had already shown his flair for turning risks into opportunities. He said yes to Chris and Curtis pitch. They were all in. That dinner at Denny's would prove to be one of the most important meals and the best investment of their lives. Jensen went to the CEO of LSI Logic, a guy named Wilf Corrigan, to tender his resignation. Wilf asked Jensen some pointed who's going to use these graphics cards? Who makes PC games? What's the developer ecosystem like? Prior to his role at LSI Logic, Wilf had been the CEO of Fairchild Semiconductor. He knew what he was talking about. He must have liked Jensen's answers because he offered Jensen and his co founders an introduction to the most legendary venture capitalist in the world, Don Valentine. So Jensen went to see Don and totally botched his pitch. It was a disaster. Jensen was nervous, unprepared, intimidated by the man sitting across from him. He was getting ready to leave when Don stopped him and said, okay, that wasn't very good, but Wilf advised me to give you money, but so that's what I'm gonna do. What's the lesson here? Well, it's not that you can afford to be cavalier about VC pitches. It's that Jensen's pitch was much more than the hour that he had spent with Don Valentine. It was the culmination of an eight year process. Sure, he didn't stick the landing, but he'd already done enough throughout his career to deeply impress someone of the caliber of Wilf Corgan. Every day. When Jensen demonstrated his knowledge and earned the trust of his colleagues and superiors, he was making an investment. It it just took until that moment for the investment to truly pay off. Sequoia took part in a $2 million round that valued the new company at $6 million. The next step was deciding on a name. Inspiration came from the file naming protocols that Jensen and his co founders were using. They'd been saving designs as NV files, which stood for next version. The next step they took was to look in the dictionary for words. With that pair of letters they found a Latin word, invidia, which means envy. Great. They decided it was time to become the envy of the industry. Nvidia had seed funding. It had a name. It was time to define a purpose and build a business model. Jensen saw that gaming had the potential to become the best way to tell stories. Because unlike movies or tv, games weren't pre recorded. The outcomes weren't predetermined. They could be different every time you played them. But he also understood that technology was simply not mature enough to realize that potential. That led him to define Nvidia's purpose. To enable graphics to be a new medium, to tell stories. In order to build their business model, the Nvidia team needed to convince developers of 2D games to learn how to develop 3D games. That required the development of new APIs and frameworks that could accommodate the new graphics chip. Nvidia was promising. There weren't many of them around at the time. It was a tough Sell. Partly because there just weren't many developers that were ready for that kind of step up. And partly it was because Nvidia had no track record. But remember, it was backed by Sequoia. That carried enough weight to land a deal with one of the leading gaming companies of the mid-1990s, Sega. The two companies began working together, but Nvidia made a big blunder right out of the gate. When creating 3D graphics, you need to set what's called a geometric primitive. A primitive is the atomic unit of graphics. It's the simplest shape the system can render. The earliest graphics systems had the most basic primitives, points and lines. Nvidia and Sega chose to define quadrilaterals, four sided polygons, as their primitives. That would become a real problem. As the industry matured, other companies began realizing that kids wanted to play games on their PCs. One of those companies just happened to develop the dominant operating system. Microsoft wanted to encourage the development of 3D graphics directly onto Windows. That also meant cutting out the middleman. Nvidia. Microsoft called its new product Direct3D. We know it as DirectX. Today, Direct3D used triangles as its geometric primitive. In hindsight, it's easy to see why. As the most basic shape that enables the creation of any kind of curved surface, triangles are the building blocks of 3D graphics. They quickly became the industry standard. So much so that Nvidia's headquarters today is literally made out of triangles. As an homage to game developers, other companies and other developers did the sensible thing and joined the Microsoft ecosystem. At the time, not doing so would have been risky. Nvidia had been working on what they thought would become the chip underpinning the Dreamcast. Then Sega dropped a bombshell. They were walking away. What's more is that in the 18 months since Nvidia started working on this chip, the price of memory had gone down by about half. Nvidia's competitors were designing their chips later and adding more memory at lower cost. They were benefiting from Moore's law, the famous observation made by Gordon Moore. More that the number of transistors on an integrated circuit doubles approximately every 18 months. They were also benefiting from the fact that they were followers, not first movers. Being a first mover is great if you can define the product and establish consumer expectations. But if your product isn't perfect, your rivals will learn from your mistakes. Having lost their main customer, Nvidia was hemorrhaging cash fast. It now had about nine months of Runway left. Most normal founders would have admitted defeat. And pulled the plug. But Jensen wasn't ready for defeat. Jensen knew that there was one more thing that Nvidia could do to stay alive. It required laying off all but about 35 staff members. Those who remained had a crystal clear brief to design from scratch and ship a new chip before they ran out of Runway. Needless to say, that was impossible. Or at least it was according to the principles which governed chip design in the mid-90s. Remember how I said earlier that Nvidia's rivals were one upping it by shipping chips with more memory at lower cost, thereby exploiting Moore's Law? Well, Jensen was about to flip the script. If Nvidia couldn't benefit from Moore's Law, it would have to get around it instead. The basic idea was comprehensive verification of chip design is crucial to reducing development costs and timelines. The functional verification of a chip design is usually performed with logic simulation or physical prototyping on what's called field programmable gate arrays or or FPGAs. Jensen knew that physical prototyping had a few drawbacks. Firstly, the process required to implement a large design into several of these gate arrays could be long and arduous. Secondly, it made fixing design flaws time consuming. Therefore, Jensen recognized an inefficiency. Simulation was underutilized early in the verification process. When bugs were more frequent, the next step was finding a way to exploit it. He'd heard about new software that enabled the emulation of chips. The startup that produced the software was unproven, though. But Jensen thought it could still do what he needed. Besides, he was out of options. It was one minute to midnight. It was time to bet every dollar Nvidia had left in the bank on black. All or nothing, it came up. Black Emulation allowed Nvidia to circumvent established production and shipping cycles. What previously took 18 to 24 months was suddenly possible in six. The result was the Riva 128, one of the first consumer graphics cards to integrate 3D acceleration in addition to traditional 2D and video acceleration. Almost overnight, it became much easier for Nvidia to persuade developers to use its chips. Why? Because Jensen had identified and optimized for a fundamental principle of customer behavior, the perception that performance was king. Sure, his process contained some imperfections, but by radically compressing its design cycle, Nvidia was able to make big leaps every six months. It could learn by doing and make step changes in the quality and sophistication of its chipsets twice or even three times as often as its rivals. There's a Real lesson there. If you believe you can anticipate or shape consumer demand, run as fast as you can toward that goal. By about 1997, the Nvidia team was doubling the performance they were delivering for a given price every six months. They'd hacked Moore's Law to run at 3x speed. A few years later, they would help thousands of companies and researchers do something very similar. Moore. More on that later, though. By 1999, Nvidia had moved two generations beyond the Riva 128. With the Riva TNT and the Riva TNT 2, Jensen and the team were keen to find a new name for the Successor to the TNT2. So they held a contest called Name that Chip. There were over 12,000 entries. You might have heard of the winner, GeForce. The first ever GeForce was the 256, which shipped in 1999. It provided up to a 50% or greater improvement in frame rate in some games compared to its main main competitor products. Jensen called it the first gpu, a graphics processing unit. A lot of people at the time dismissed that as a nice piece of marketing, but he was right. Most people just couldn't see it. And in fact, the development of GPUs came just in time. The performance increase of CPUs had slowed to as little as 10% every year as chip makers encountered physical constraints, including size, yield and heat limits. With the development of the gpu, the Nvidia had once again outsmarted Moore's Law. To understand Moore, we need to dig into a concept called parallel computing. A very basic model of a standard CPU is that you run an instruction and then the CPU executes a long chain of operations to process whatever it is you want to output. This chain is sequential. One thing has to happen after another in a specific order. That's not true for graphics processing. Graphics can be rendered in parallel because pixels on a screen can be rendered independently of each other. Therefore, the only real constraint on how quickly you could process graphics was how quickly you could execute instructions in parallel. Raw computing power. The more cores you add to a system, the more quickly you can execute those parallel functions. Beginning with the GeForce series, Nvidia leveraged that parallelism. But it would still take years for use cases beyond gaming and graphics to emerge. More on that later. Before any of that happened, Jensen, Curtis and Chris decided that Nvidia was ready to go public. The company first sold shares to the public on January 22, 1999, at $12 a share and a $600 million valuation. If you had spent $10,000 at the IPO, you could have bought 833 shares of Nvidia. Once you factor in the four stock splits since their listing, the value of that stock would have peaked in December of 2021 at over $13 million. Not bad for a $10,000 investment. The massively successful IPO was a milestone for Jensen, Curtis and Chris, not to mention Don Valentine and Sequoia. Nvidia was suddenly flush with capital. Their IPO and the development of the G4 series also meant they had lots of another precious credibility. Microsoft approached Nvidia about supplying the GPU capability for a secret new project it was working on. Can you guess the secret project that Microsoft was working on around the turn of the millennium that was required? Advanced graphics. It wasn't the Zune. Microsoft and Nvidia shook hands on a $500 million deal for Nvidia to supply graphics for the Xbox. The chip Nvidia used was the modified version of its newest design, the GeForce 3. This was the first card to contain programmable shaders and lighting. Earlier graphics cards were basically fixed function graphics accelerators. They would be able to map textures onto a set of polygons, but the output was predetermined. Programmable lighting and shading meant dynamic lighting and shading. The GeForce 3 enabled developers to deliver a much more complete gaming experience. It was a big step forward for GPU architecture, and it was the fullest realization yet of what Jensen promised. Nvidia would unlock a new way to tell stories. The Xbox deal and GeForce series put Nvidia on an incredible growth trajectory. For the fiscal year ending January 31, 1999, the company made $158 million in revenue. The they more than doubled that the following year to 375 million. The year after that, 735 million. The next year, which was when the Xbox was released, Nvidia made about $1.4 billion in revenue, doubling, doubling, doubling year after year. Exponential growth is the pathway to becoming one of the biggest companies in the world. Financial success was made possible by technological success. As I said earlier, Nvidia had effectively doubled or even tripled how quickly they were able to ship new chipsets. They shipped their first GeForce card, the 256, in October of 1999. Its successor, the GeForce 2 shipped the following spring. The GeForce 2 Ultra shipped in late 2000. The GeForce 3 with its programmable shaders shipped in spring of 2001 and then six months later, the GeForce 3 Ti 500 Jensen started Nvidia because he knew there was a better way to make chips. Within a decade, he had totally proved his point. Nvidia had created a truly programmable GPU that that took graphics to the next level and allowed developers and designers to tell new kinds of stories. Nvidia had risen to the top of the industry. It had near perfect product market fit. And yet Jensen knew as well as anyone that if you don't continue aggressively innovating, you're moving backwards. By the middle of the 2000s, some clouds were gathering on the horizon. Gross margins were getting thinner. The original Xbox was about to be replaced by its successor, the Xbox 360, which was not powered by an Nvidia GPU. That meant another long term revenue stream was about to run dry. Nvidia's biggest rivals were getting increasingly serious. In 2006, intel announced Larabee, its own dedicated GPU. Don't get me wrong, the picture wasn't as bleak as it had been 10 years earlier. Before Nvidia pioneered chip emulation, they were still extracting huge revenues from PC gaming. But Jensen wasn't the type of guy to be satisfied with dominating a single market. He wanted to be ready for when GPUs could be used in other domains. Could Nvidia reinvent itself yet again? Could Jensen come up with yet another genius play? We're about to find out. The story of how Nvidia went from being a leader in building graphics cards for gaming into the world's eighth biggest company begins with a story that can't entirely be verified. Supposedly, in the early 2000s, a quantum chemistry researcher at Stanford called Jensen to say thanks for solving a problem. It apparently took up to two weeks for this researcher's machine to do important computations. Until, that is, the researcher's son suggested that he go to his local electronics store, buy a bunch of G Force cards and install them on his lab computers. That's what he did. And apparently the computations that used to take days or even weeks on his lab machine could now be completed in mere hours. The researcher ran the same computations on his old setup, which was basically a supercomputer, to double check that the results held up. They did. But on the rig of Nvidia GPUs, they finished 10 times faster. In truth, whether this happened or not is kind of beside the point. What matters is that it motivated Jensen to evolve Nvidia into a new kind of company. Do you remember what inspired Jensen, Curtis and Chris to start Nvidia in the first place? It was Their desire to enable a new way to tell stories. On one front, they'd already succeeded. Nvidia was a huge factor in games becoming deeper, more immersive and more interesting, entertaining. But there was nothing specifically about gaming that made it Nvidia's main source of revenue. Beyond being the most obvious use case when it began developing its chips, there were still so many stories to tell and discoveries to make. Jensen wanted Nvidia to help people like that Stanford researcher discover new insights. So he was about to make his biggest and boldest bet on scientific computing. His first step was to ask the first order question that everyone should ask before making such a big bet. How big is the addressable market? That's the question that Don Valentine at Sequoia would have been asking at the time. It was kind of hard to argue that there was a sufficiently big opportunity. The customers would mostly be universities or individual researchers in science specific domains looking for ways to perform faster computations. They'd be working on hard problems like computational modeling and drug discovery. Maybe the market could be expanded to other professional services like architecture, but it was still difficult to see the use case that would would merit such a big outlay. Being visionary means seeing things others can't. But the most visionary founders are also often the most stubborn. I'm sure Jensen did market research to understand the potential of scientific computing, but it's also clear he just fell in love with the concept. He wanted to develop hardware that would enable smart people to expand the frontiers of human knowledge and tell new kinds of stories. He was confident that innovation would increase the number of viable use cases for scientific computing. And he was confident that his play was the path to Nvidia's durable differentiation. You've heard the saying that if you build it, they will come. But Jensen was more interested in the opposite. His attitude was if you don't build it, they can't come. That was his way of articulating something really key. The best way to generate massive revenues is to create new markets. After ascertaining the size of the addressable market, the next question to always ask is no matter your size is what will it take to get where you want to go? If Nvidia was to capture the market that Jensen believed would soon emerge, it had to be ready. It had to provide a new kind of technological platform, one that would rival the invention of the transistor for how it transformed computing. When Nvidia debuted the GeForce series in the late 90s, Jensen called it the first GPU. Remember that the GeForce did something new. It allowed for parallel execution of graphics graphics processing tasks. In practice, that meant faster and more dynamic rendering of graphics. Supporting those functions required Nvidia to extend the C programming language to provide developers with new tools. Their extension of C was called CG C for Graphics. It was a high level shading language developed by Nvidia in close collaboration with Microsoft for programming vertex and pixel shaders. It's kind of like an instructional pamphlet for the next next generation of graphics. But to achieve Jensen's true goals, he needed more than a pamphlet. He needed a full stack API that could be harnessed by developers and users in other domains. He found the answer in a researcher named ian Buck. In 2003, Ian led a team which developed Brook, the first widely adopted programming model to extend C with data parallel constructs. By 2006, Ian had joined Nvidia. Together they got to work on a general purpose parallel computer computing platform which could leverage the parallel computing capabilities of Nvidia's GPUs to solve highly complex computational problems like the one facing our friend the Stanford quantum chemist. The result was CUDA Compute Unified Device architecture version 1.0 shipped in November of 2006. It was the skeleton key that unlocked a whole new world. Many scientific and AI workloads have inherent massive parallelism in their algorithms. And because of that they run very slowly on CPUs. CUDA was able to offload those complex and time consuming functions to run on GPUs exploiting their parallel capabilities. All of a sudden, developers could tap into incredible computational power to dramatically accelerate workloads in domains as varied as computational finance, fluid dynamics, weather forecasting, medical imaging, quantum chemistry, electromagnetics and more. If the invention of the transistor and microprocessor are are recognized as the greatest inventions in computing, then CUDA can't be that far behind. And would you believe the technical feat wasn't even the most extraordinary part of this. The most amazing part of CUDA was the business model. To start, Nvidia has never charged a dollar for it, but it can only be deployed on Nvidia's hardware. This idea might remind you of another huge technology company that built a brilliant development ecosystem which was proprietary to its expensive first in class hardware. Of course I'm talking about Apple. But what's incredible is that Jensen started on this journey before iOS, before the iPhone. There's a beautiful symmetry there. GPUs unlock the power of parallel computing and Apple and Nvidia created the business models that made them billions of dollars in parallel. But Even the progress of this new development platform and an equally elegant business model weren't enough to put it on the path it is on today. In fact, Nvidia Media's stock actually went down by around 50% in 2011 after the company missed its earnings targets. In a way, Cuda was invented before the rest of the world was ready. It would take a twist of fate for it to catch up. In the year 2000, an American computer science professor named Fei Fei Li began working on an image classifying project she called imagenet. She was inspired by a project at Princeton from the 1980s called Wordnet. Her aim was to create a database of millions of correct, directly labeled images to assist in AI image recognition. Once the database was up and running, she decided to start a competition so that other researchers could explore it and use it. In 2012, a team from the University of Toronto made up of Alex Kruzhevsky, Ilya Sutskever and Geoffrey Hinton won the competition. They didn't just win by a little though. They dominated. The algorithm they submitted, which they called Alexnet recognized about each 85% of the images it was presented with. As you might be able to tell from its name, Alexnet was a neural network. Deep learning neural networks weren't new. They'd been around for decades. But they were very computationally demanding. It was basically impossible to harness them for practical applications within a traditional computing architecture. The team from the University of Toronto did exactly what you've already realized. They trained their neural network in Cuda on Nvidia GPUs. They this was the big bang moment for artificial intelligence, the marriage of unprecedented parallel computing power with deep learning models that taught machines how to learn. It was CUDA's killer app, its perfect use case. And it was also a huge coincidence. But as Jensen says, it wasn't a coincidence that Nvidia was the company that benefited most from the application of deep neural networks to GPUs. Because it had done the years of work to pioneer GPUs in the first place, Nvidia made itself the hardware supplier of choice for the machine learning revolution by being ready for the opportunity when it came. Machine learning transformed Nvidia's business by enabling entirely new revenue streams. One of them was broadening the customer base for its high end GPUs like the RTX 3090. That includes gamers, researchers and people who just want top of the line processing power. But the real money was always going to be found in enterprise customers. They have the deepest pockets and the greatest need Today, top of the line Nvidia GPUs power thousands of massive data centers for the world's biggest enterprise customers. Google, Microsoft, Amazon, the biggest technology companies in the world. Companies that create and maintain the infrastructure of the Internet itself use Nvidia hardware. When Tesla announced its Dojo AI training center, they announced their intention to develop their own chips internally. But today, it still uses Nvidia chips. Think about that. One customer using chips for a single purpose netted Nvidia at least $50 million of revenue. Enterprise definitively answers the question that Wilf Corrigan and Don Valentine asked Jensen. How big is this market? The market is gigantic and growing in size every day. As some of the world's biggest and most ambitious companies invest in deep learning techniques to fight disease, build the next generation of autonomous vehicles, and revolutionize the natural sciences, there's a very good chance to chance they're doing so with Nvidia GPUs. There was no looking back. In April 2020, Nvidia spent $7 billion to acquire Mellanox, a next generation networking company. Since then, it's tripled the revenue generated by its data center stream. That's allowed Nvidia to pull off the sort of revenue growth that usually only hot startups can enjoy. In 2020, Nvidia announced it had generated about $11 billion in revenue over the previous fiscal year. Last year, that figure jumped to $16.6 billion. And in February of 2020, Nvidia announced it had generated an insane $26.9 billion in revenue. That's a 60% increase from the previous year for a 30 year old company. Remember the business model? An open development ecosystem that can be deployed only on Nvidia hardware. Cuda and Nvidia GPUs dominate several application areas, including deep learning, and underpin some of the fastest computers in the world. Nvidia generates $8 billion of free cash flow every year and has more than 20 billion on hand. Whereas at the turn of the 90s they had a 30% gross margin, today their margin sits at 66%. They have built and fortified one of the world's deepest moats. I want to talk briefly about power and how companies can wield it. One common form of power is the ability to choose where you operate on the price curve. That kind of power is usually reserved for monopolies. Another form of power is the ability to make your competitors react to you, rather than the other way around. That kind of power is reserved for companies that create new technologies in new markets. It's very Rare to wield both forms of power, let alone at the same time. But that's just what Nvidia was able to accomplish. By pioneering the emulation of graphics chips, it could deliver customers what they cared about sheer performance. By creating truly programmable GPUs, it leveraged parallel architectures to massively accelerate performance. And just as the revenue from that established technology was beginning to plateau, it found a new use case for its technology in the form of enabling huge advances in machine learning, all while locking in consumers and enterprise customers into a proprietary ecosystem. That's real power. No one can predict the future, but as long as Jensen is in the driver's seat, Nvidia will never stop innovating. They're making a play for automotive, meaning the electric vehicle and autonomous vehicle software stack. Jensen thinks that's a $300 billion addressable market. And another project with massive potential is Nvidia's Omniverse. The Omniverse isn't like the Metaverse. It's not a place where you can walk around, commute to work, or speculate on digital property. It's useful here to look at Earth 2, the supercomputer project Nvidia announced in late 2021. Earth 2 will combine GPU accelerated computing, deep learning, and observed and modeled data to simulate ultra high resolution climate models. In a world where one to one scale simulations are possible, it's easy to imagine any company simulating how their new product or service will work in the Omniverse before deploying it in the real world. It could be running the world's most sophisticated climate models on a replica digital Earth. Or space startups simulating atmospheric effects, or their prototype vehicles. Or new pharmaceutical firms finding hundreds of new molecules that could hold the key to curing some of the world's most persistent and devastating. If the technology is there, then there's no limit. The Omniverse is an extremely ambitious concept because nothing like it has ever really existed. But in a way, it's easier to see it working out. Partly that's down to Jensen's unbelievable track record of delivering moonshots. But it's also because it's easier to see the continuity between the Omniverse and the immense scientific discovery Nvidia has enabled through Cuda. Jensen, of course, saw the potential of simulation from a long way off. In 2006, Nvidia invested in a company company that was creating a navigable software model of the Earth. Back then it was called Keyhole, but today we know it as Google Earth. Simulation holds incredible promise for accelerating scientific discovery by an order of magnitude. One of the biggest reasons why is that simulation already saved Nvidia. It's what enabled it to defy Moore's Law and pioneer chip emulation. It was critical to the development of the first consumer GPUs and it is the principal underlying CUDA. But most of all, Nvidia's extraordinary capacity capacity for simulation has allowed millions of people to do what Jensen Huang promised when he started the company 30 years tell better stories. Thanks for watching. I hope you enjoyed this video and if you'd like to go a bit deeper, please just watch this video next.