Loading summary
Transcript21 lines
- [00:09]
Narrator
June 2007 Santa Clara, California. It's lunchtime and several dozen financial analysts are tucking into sandwiches in a tent set up in the parking lot of Nvidia's headquarters. They've spent this morning listening to presentations about Nvidia's latest bet, Cuda, a platform that allows computer programmers to turn Nvidia's graphics chips into into the workhorses of high end computing. For years, Nvidia used its engineering prowess to design GPUs graphical processing units to make video games look and play better. Now it's chasing a different crowd, scientists and technical professionals whose work demands incredible amounts of computing power. But many of the analysts here today think it's a waste of time. At one table, Nvidia CEO Jensen Huang is getting grilled by a skeptical analyst. Even the rosiest estimates put the market opportunity for this at $100 million. You spent almost five times that just bringing it to market. How will this ever deliver returns for investors? Huang puts down his sandwich and trots out the official company line.
- [01:19]
Jensen Huang
We're creating an entirely new customer base for our company. Our graphics chips offer 10 to 200 times the performance of general purpose chips. We're about to enter the era of the gpu.
- [01:32]
Narrator
The analyst isn't swayed. You're sacrificing profit margin on a long shot.
- [01:37]
Jensen Huang
Our revenues and profits are still growing. We are delivering for shareholders today while also ensuring we deliver for them tomorrow.
- [01:45]
Narrator
But you could be delivering more for them today. Cuda. Add $0 to your market capitalization. One of the other analysts at the table senses that Huang's patience is wearing thin and interjects jensen, I'm interested in the potential use cases. I have a two year old and I'm taking a lot of high resolution photos of her. But when I transfer them to my Mac to edit them in Photoshop, my computer grinds to a halt. Could Cuda help with that? Huang's eyes light up.
- [02:16]
Jensen Huang
Yes. In fact, we've already partnered with Adobe on this. Photoshop with Cuda will hand the task of editing photos to the gpu. It won't just stop your Mac from slowing down, it will process the images faster. This is what we mean by the era of the gpu.
- [02:34]
Narrator
The analyst who asked the question is impressed, but it's clear the others at the table still don't get it. They're thinking in quarters, but Huang's thinking in years. Huang doesn't know yet who will use Nvidia's chips for high end computing or how, but he believes if he gives them the tools they will build amazing things. Still, Nvidia is a public company with stockholders to answer to, and their patience won't last forever. If the era of the GPU is going to happen, Nvidia will have to create a market that doesn't exist yet. Closing the books, getting your people paid, and bringing on new hires Running a small or mid sized business can be exciting and also a little chaotic. Workday Go makes simplifying your business, well, simple. Imagine all the important aspects of your company hr, finance and payroll, all on the AI platform. No more juggling multiple systems. No more worrying about growing too fast. Just the full power of Workday Go, helping small to mid sized businesses like yours scale and run more smoothly. Think about what that means. Seamless onboarding for new team members, real time insights at your fingertips and payroll that works perfectly every single time. So you can focus on the big picture and go after your big ambitions. And with Workday you can activate quickly in as little as 30 to 60 business days. So simplify your business. Go for growth. Go with Workday Go. This message is brought to you by Apple Card. Apple Card members can earn unlimited daily cash back on everyday purchases wherever they shop. This means you could be earning daily cash on just about anything, like a slice of pizza from your local pizza place or a latte from the corner coffee shop. Apply for Apple Card in the Wallet app to see your credit limit offer in minutes. Subject to credit approval. Apple Card issued by Goldman Sachs Bank USA, Salt Lake City branch terms and more@applecard.com from wondering I'm David Brown and this is Business Wars. Last episode, Nvidia went from an idea hashed out in a Denny's diner to the leader in video game graphics chips. Now CEO Jensen Huang wants to use the company's graphics processing units to bring supercomputing to the masses. But with mounting costs and no obvious customers to sell to, investor patience may run out before Huang's big bet pays off. This is episode two once in a lifetime it's 2007 and Nvidia stuck between a rock and a hard place. It's burned through hundreds of millions of dollars developing Cuda. The platform was supposed to usher in a new computing era. Nvidia imagined programmers would use it to harness the parallel computing power of its GPUs and deliver the next wave of technological breakthrough. But simply making the tools available wasn't enough. Developers are ignoring it, which makes Wall street fear that Cuda will never deliver for stockholders. Nvidia has the tech, but the market's not ready. You know, entrepreneurs often fall into this trap. They bill for tomorrow while customers are still stuck in yesterday. The key isn't just invention, it's translation, helping people see why your future matters. Now Nvidia decides to try to get universities to embrace CUDA and teach the next generation of computer scientists how to use it. The company's chief scientist, David Kirk, leads the charge, wooing computer science professors with offers of free Nvidia hardware if they teach their students about cuda. It's a tried and tested strategy. Hardware and software makers often donate money and products to universities not out of generosity, but to ensure students learn to use their products instead of the competition's. But time after time, universities decline Nvidia's offer because professors don't know CUDA and aren't familiar with how to program GPUs. They're more comfortable with CPUs, which work linearly, but GPUs use parallel processing. You know, you can think of a CPU like a single lane street where the vehicles represent individual computing tasks. They all move forward in a single line, one after the other. Each vehicle can only proceed down the road in the order they arrive. GPUs, on the other hand, are more like multi lane highways. There are thousands of tasks happening at once and they don't share the same lane. So they can all move forward at different speeds at the same time. That's the big advantage of parallel computing. The downside is that it's much harder to program. Developers need to coordinate thousands of simultaneous tasks. There is a steep learning curve for anyone used to programming CPUs, which is almost all computer science professors. And this leaves Nvidia in a catch 22 to grow interest in CUDA. It needs universities to teach it. But if the teachers don't know it, they can't teach it. Still, Nvidia is onto something here. Seeding adoption in schools is a proven approach. It's how Apple conquered design schools and how Autodesk ruled architecture labs. When marketing won't open doors, curriculums will. Smart tech companies don't just sell products, they teach people how to need them. Finally, Nvidia gets a break. Kirk contacts the University of Illinois at Urbana Champaign. The head of its computer engineering department tells Kirk that if he really wants students to learn about parallel computing, he should teach the class. Kirk hesitates. He is not interested in teaching. But the university offers to pair him with another professor who can teach him how to teach. So Kirk agrees to give it a try. He flies from his home in Colorado to Illinois every other week and teaches a course called Programming Massively Parallel Processors. He assigns students research projects using Nvidia's tech and then publishes their findings. These research papers pique the interest of other universities who start asking for Kirk's teaching materials. Soon, parallel programming classes start appearing at universities around the world. Wow. Think of this. Instead of buying billboards, Nvidia bought something a whole lot more durable. Credibility. Every professor who taught CUDA became an unpaid brand ambassador, and every student turned evangelist extended the reach. That's network growth the old fashioned way, through reputation, not algorithm. Sure, it's slower at first, but stronger once it catches on. If you can turn believers into evangelists, you'll never need to outspend your rivals. But academia's embrace still isn't enough. Cuda remains a novelty rather than a catalyst for a tech revolution. The era of the GPU still seems far away. What Nvidia really needs is a killer use case, something that can demonstrate the untapped potential of GPUs. It's 2010, and in Stanford, California, Nvidia's new chief scientist, Bill Dally, is catching up with a friend, Andrew Ng, over breakfast. Ng is the director of Stanford University's Artificial Intelligence Lab and is working with Google on an ambitious project. The project's goal is to train computers to recognize images. To do this, Ng and his team are using neural networks, a type of computer system that was inspired by how brains work. When fed enough data, neural networks learn to identify patterns and find connections using a process of trial and error. And Ng's team is training its neural networks on one of Google's biggest data sets, YouTube. They're using all those cat videos uploaded to YouTube to teach their neural network to recognize cats, even in images its machines have never encountered before. Dall? E is impressed by UNK's progress. Neural networks are an old idea. They were first proposed in the 1940s. But it's only now that computers are powerful enough and data sets big enough to make for meaningful progress. Even so, the project is a heavy lift. Ng informs Dali that the project's neural network uses 2000 CPUs. Dali smiles and tells Ng that he's sure Nvidia's GPUs could do the same job faster and better. After that breakfast, Nvidia tests that theory by helping the project replace its CPUs with GPUs. But it's not a simple transition. GPUs need tasks diced into small pieces and specialist programming, but the final results make the effort More than worth it, the cat recognition algorithm that once required 2,000 CPUs now runs on just 12 Nvidia chips. It's a great leap forward for artificial intelligence research, or AI, and a great leap forward for Nvidia. With CPUs, machine learning crawled. With GPUs, it soars. Every breakthrough needs its proof of life. For Cuda, it was a neural network that learned to spot cats. Silly on the surface, but revolutionary where it matters. When you're selling something complicated, stories beat specs every time people don't remember tenfold efficiency gains. They remember how cat videos taught computers how to see. When Nvidia CEO Jensen Huang hears about the project's success, he takes a personal interest in it. For years, Nvidia has been hunting for a reason for Cuda to exist. Now it feels like it's founded in machine learning, a field that's full of promise and one that Nvidia now has a chance to claim for itself. But even with Nvidia's chips powering new technological advances, investors still aren't convinced. Machine learning is a bet on future breakthroughs. And as the years tick by, the pressure on Nvidia rises. In 2011, John Nichols, the driving force behind Cuda, dies from cancer. He believed to the end that Cuda could transform the world, and he kept working on it, even as his health gave way. But outside of Nvidia, enthusiasm continues to wane. Downloads of Cuda's software tools keep falling year after year. The company's share price has barely changed since Cuda's introduction, and Most of Nvidia's $4 billion of annual revenue still comes from selling graphics chips to gamers. To investors and analysts, Cuda looks like a vanity project, an expensive toy for academic tinkerers. To them, Nvidia is wasting money that could be invested elsewhere or paid out in dividends. And Huang is seeming less and less like a visionary and more like a CEO too proud to change direction. As investor unhappiness festers, hedge funds start circling. They see a company obsessed with long term bets that may never pay off and one that's failing to deliver maximum shareholder value. So they start buying Nvidia's flatlining stock, preparing to strike and force change that will net them millions.
- [15:58]
Steve Nash
Hey, basketball fans. Steve Nash here. Ready to elevate your basketball IQ? I'm teaming up with LeBron James to bring you the latest season of Mind the Game. And we're about to take you deeper into basketball than you've ever gone before. We're breaking down the real game, the X's and O's that actually matter in every episode. We'll share elite level strategy, dive into career defining moments and explain the why behind plays that changed a game, a team or a championship. LeBron and I have lived this game at the highest level for decades. We've been in those pressure moments and made those game changing decisions and learned from the greatest basketball minds in history. Now we're pulling back the curtain and sharing that knowledge with you. Time to go beyond the highlights and get into the real heart of basketball. Watch Mind the game now on YouTube Prime Video or listen wherever you get your podcasts.
- [17:02]
Narrator
It's October 2012 in Florence, Italy. A tiny conference for neural network researchers is underway and on stage is a nervous young man from Toronto who's here as the keynote speaker. His name is Alex Kruzchevsky. He's in his mid-20s and he's not used to public speaking. He puts his script on the lectern, locks his eyes on the page and begins. His voice is monotone, with the faint quiver of nerves, and he never looks up. His presentation slides aren't helping. They're as devoid of flair as Khrushchevsky's voice. But if he did glance up at the audience, he'd see they're on the edge of their seats. Khrushchevsky tells them that earlier this year he bought two Nvidia GeForce graphics cards from Amazon for $1,000. Then, in the bedroom of his parents house, he connected them to his PC and used them to train a neural network on a collection of photos known as ImageNet. Using ImageNet to test neural networks is a common way to benchmark performance. The network must look at a photo and guess what it's an image of. At first the guesses are random, but every incorrect answer teaches the network what the image is not. Over time it learns enough to start getting things right, but eventually the gains plateau and the network tops out. The question everyone cares about is how high a percentage can your network get? Khrushchevsky tells the room that after a week of running up his parents electricity bill, his Nvidia GeForce powered network topped out at 80%. He can't tell the difference between a hatchet and a spatula, but still this percentage is staggering. Then Khrushchevsky looks up for the first time, tells the audience that's all he's got, and abruptly ends his talk. The room is silent, the audience too stunned to respond. People in this room have spent their careers training neural networks on supercomputers and they've only managed around 70%. Now some kid has blown past them using two gaming cards that are designed for powering Call of Duty. It's humiliating and inspirational, because now they realize the computer power they need to push AI forward is sitting on the shelves at their nearest Best Buy. Word of Khrushchevsky's experiment spreads fast through the AI research world and convinces many to start using Nvidia chips. But just as scientists finally wake up to Cuda's potential, its very existence is under threat. It's early 2013, and at Nvidia headquarters in Santa Clara, CEO Jensen Huang is face to face with a new opponent. Jeff Smith, CEO of the New York hedge fund Starboard Value, has come calling. The vibe at the meeting is cordial, but with an undercurrent of menace. Smith is the kind of man who gives corporate executives nightmares, a boardroom brawler willing to stage a coup if that's what it takes to get his way. And today, he wants Nvidia to take action to improve its share price. Starboard makes money by buying big stakes in companies it believes are undervalued. Then it pushes for changes that will boost the stock price so it can cash out with a hefty profit. Nvidia is its latest target. Starboard spent more than $60 million buying Nvidia stock. Now Smith wants Huang to take actions he believes will make that stock more valuable. And right now, Smith's questioning the company's ongoing commitment to Cuda. Huang has poured nearly a decade and a fortune into this project. But Cuda's user base is small. It's mostly academics who are dependent on unpredictable research grants. There's no sign that Nvidia's investment into Cuda will ever be recouped. Smith points out that most analysts believe Nvidia's stock would be worth more if it stopped trying to convince the world to embrace Cuda. But Cuda isn't The main thing Starboard thinks is dragging down Nvidia's stock price. Nvidia is also chasing the mobile and tablet markets and recently bought a manufacturer of mobile phone modems. Smith argues that Nvidia is trying to do too much and needs to focus. Smith also has a problem with Nvidia's bank balance. The company is sitting on $3 billion of cash. Smith wants the company to use this money to buy back its own stock. He believes this will lift the stock price and get investors excited about Nvidia again. So why would buying back stock help Nvidia's share price? Well, first it signals to the markets that Nvidia thinks its stock is undervalued. This simulates demand for the stock pushing up the price. At the same time, it reduces the number of Nvidia shares available to buy. And you know about economics 101. When demand goes up, supply goes down, prices rise, right? But there's another benefit. By reducing the number of shares in circulation, Nvidia's annual dividend payments are spread out among fewer investors. This means investors earn more from their stock holdings, making Nvidia shares more attractive once again. And, of course, stockholders get to profit from selling their shares to Nvidia at a good price. Huang refuses to abandon Cuda, but after a few more meetings, he agrees to spend 1 billion buying back Nvidia's own stock. The buyback announcement rallies the company's stock price, pushing it up 20%. Soon after, Starboard sells its stake in Nvidia at a big profit. But Huang knows this was a near miss. Next time, it may find itself battling an activist investor who's determined to kill off the Cuda project. Activist investors can feel like a hostile takeover in slow motion. They want quick returns, not long bets. But a smart CEO can use that tension to prove conviction, trimming the fat while keeping the heart intact. You'll notice Jensen Huang didn't cave. Instead, he made a tactical retreat. In business, survival depends on knowing when to give in and went to dig in. It's mid-2013 in Santa Clara, and Nvidia employee Brian Catanzaro is walking towards a conference room at the company's headquarters. Catanzaro is a research scientist who helped show Andrew Ng and Google what Nvidia GPUs could do. Now he wants to create a specialized version of Cuda for AI developers. He calls it CUDA Deep Neural Network. CUDNN for short. But thanks to the way Nvidia is structured, there's only one guy he needs to win over to get the project approved. Catanzaro steps into the conference room. Inside, Jensen Huang is hard at work, as usual. This might be a conference room, but it's also Huang's office, the heart of the company. Because Huang doesn't have a C suite, he is the C suite. There is no chief marketing officer, chief technology officer, or chief operating officer at Nvidia. It's just Huang and the more than 30 vice presidents who report directly to him. It's been like this since Nvidia was founded. Huang believes this structure lets him reshape the business quickly without having to worry about bureaucratic turf wars between senior executives. It Also means anyone can bring an idea straight to the top. And that's why Catanzaro's here today. His own managers don't think Coup DNN is important, and his performance reviews suggest he's far from a model employee. At most large tech corporations, he'd probably be sidelined, silenced, and eventually cast aside. But at Nvidia, he actually has a chance to lobby the CEO for support. They met for the first time a few days ago. Catanzaro expected a brief, curt meeting. Instead, Huang was intrigued and said he wanted to look into it more. And now he's called Catanzaro back to talk again. Catanzaro barely has time to say hello before Huang unloads his thoughts.
- [26:09]
Jensen Huang
I cleared my entire schedule so I could spend the weekend reading and making calls about this. I believe this project could be the most important one in Nvidia's entire history.
- [26:20]
Narrator
Huang points to the whiteboard. It's been wiped clean, apart from a mysterious acronym, oial. Catanzaro stares at the letters, confused. They mean nothing to him. He's got a newborn at home and is sleep deprived. But he's pretty sure those letters would still mean nothing even if he were well rested. Huang grins.
- [26:44]
Jensen Huang
O I A L O. It stands for Once in a lifetime opportunity. That is what Coup DNN is.
- [26:57]
Narrator
Huang gushes about the potential he sees in KU Diann. Catanzaro is stunned. Over the weekend, Huang's gone from knowing little about the proposal to arguably understanding it better than he does.
- [27:12]
Jensen Huang
Brian I asked myself, if a computer can use CUDNN to recognize images, what else can we teach it? My conclusion was everything.
- [27:23]
Narrator
Huang's eyes are now wild with the excitement of someone who's just seen the future.
- [27:29]
Jensen Huang
Neural networks are going to revolutionize society. And with cudnn, we can corner the market on the hardware needed to make them. From this second, we are no longer a graphics company. We are an AI company.
- [27:43]
Narrator
Catanzaro stares in shock. He thought he was pitching a small research project. Now Huang is talking about reshaping Nvidia around it. Catanzaro wonders if he's dreaming. Huang smiles at him.
- [27:59]
Jensen Huang
Now imagine this. I've marched all 8,000 Nvidia employees into the parking lot outside. Now, pick who you want on your team. Anyone you want. You choose them, they join your team.
- [28:13]
Narrator
And just like that, Huang has reorganized Nvidia around the belief that neural networks will be as transformative as the Internet. Now, the company's goal is to move quickly to build the hardware that will power these neural networks. But he's not just Trying to accelerate the future of AI, he wants to capture the market before his rivals even realize there's a market to own. It's August 2016, and in San Francisco's Mission District, Nvidia CEO Jensen Huang is pushing a dolly with a large cardboard box on it. He's dressed all in black. Black shoes, black jeans, black tee, black leather jacket. He looks like he raided YouTube singer Bono's closet. But this is his new uniform, and just like Steve Jobs, turtlenecks. This is no accident. It's branding, a way for Huang to stand out in a sea of graying CEOs. But today, what's inside the box is the real star. Huang wheels it into the Pioneer Building, the historic home of a startup called OpenAI. Inside, OpenAI co founder Elon Musk. Musk is waiting to greet him. They take the dolly into a nearby room, where, with help from an OpenAI employee, Huang lifts the box onto the floor. Then Musk moves in, box cutter in hand. Inside is a large rectangular steel machine, the first ever Nvidia DGX1 server. Nvidia calls it an AI supercomputer in a box, and that's no exaggeration. Packed inside this machine are eight of Nvidia's most advanced GPUs. Together, they're capable of carrying out up to 170 trillion calculations per second. This is the machine OpenAI will use to start training the system that will eventually become ChatGPT. The DGX1 retails at $129,000, but Huang's giving it to OpenAI for free. After lifting it onto a table, Huang takes a black marker and scrawls a message on the DGX1's steel casing.
- [30:55]
Jensen Huang
To Elon and the OpenAI team. To the future of computing and humanity, I present to you the world's first DGX1.
- [31:07]
Narrator
The DGX1 is a computer built specifically to train AIs, and it's spearheading Nvidia's push to claim ownership of this emerging market. When it comes to sales, it's still a sideshow for Nvidia. Many entrepreneurs face moments of truth just like this. The paying customers want one thing, but the future points somewhere else. Nvidia could have stayed comfortable selling gaming chips forever. The smart move wasn't abandoning the old market. It was using it to bankroll the new one. You don't grow by burning bridges. You let today's revenue build tomorrow's Runway. And if you can master this balance, you're not just chasing trends. You've actually got a shot at shaping them. In 2016, Nvidia's revenues reached nearly $7 billion. But most of its income still comes from video games, selling GPUs to PC gamers, its Shield family of gaming tablets, and its game streaming service, GeForce. Now, Nvidia's data center products, which include its earnings from AI, contribute just 12% of the company's income. But non gaming demand for Nvidia tech is about to explode. Thanks to crypto, the digital currency bitcoin is having a boom. In November 2017, the price of one bitcoin reaches a new high of $10,000, up from just $1,000 at the start of the year. And this ignites demand for crypto. But bitcoins don't grow on trees. To get new bitcoins, people get computers to generate or mine them by solving incredibly difficult and time consuming math problems. This means crypto miners need computers that can do complex math fast, which is exactly what Nvidia GPUs are built to do. Crypto miners strip stores of Nvidia graphics cards on ebay. Nvidia's cards sell for twice their suggested retail price. People start building homemade supercomputers out of dozens of Nvidia GPUs, all of which boosts Nvidia's sales and its stock price. Normally, that's a good thing. However, some inside the company worry that Nvidia's worth is now tethered to the roller coaster fortunes of the crypto bubble. But Huang's not about to ignore this unexpected windfall. So Nvidia cranks up production of its chips to try and meet demand. But then the crypto wave crashes and demand collapses, dragging Nvidia down with it. The company's stock price falls 31% in 2018. In the aftermath, Nvidia acts to keep crypto from distorting its business again. It launches a dedicated crypto mining card and then impairs its gaming cards so they can't mine crypto efficiently. But while the crypto circus has been grabbing the spotlight, Nvidia has been pushing forward with its plan to build and control the hardware needed to create AI. And this puts it on a collision course with an old FOE, Intel. In January 2019, intel makes a $6 billion offer to buy Mellanox, an Israeli company that creates high speed switches, routers and cables that move huge amounts of information through AI data centers. Without these super fast data pipes, even the world's fastest GPUs are bottlenecked. And this makes it an attractive acquisition target for intel, which is eager to strengthen its presence in the data center business. But Huang's not about to let intel get a stranglehold on Mellanox's tech. He swoops in and starts a bidding war. Intel fights back and the two chip giants slug it out to claim the prize. After two months of bids and counter bids, Nvidia emerges Victorious with a $6.9 billion offer. The move also makes Nvidia more of a one stop trip shop for customers seeking hardware for their AI data centers. Now this is what you call playing offense in a defensive market. Acquiring Mellanox wasn't about expansion. It was about stopping a competitor getting a foothold. Offense can be the best form of defense if you can afford to swing first. But Nvidia's quest to own the hardware of tomorrow isn't over yet. The following year, Nvidia strikes a $40 billion deal to buy ARM, the British chip design company behind the microprocessors used in most of the world's smartphones, tablets and smart TVs. Nvidia believes this takeover will help it become the premier computing company of the AI age. The combined company would allow Nvidia to offer data centers all the kit they Nvidia GPUs, ARM CPUs and Mellanox Networking hard. And that would help it compete more directly with intel and amd. But the move alarms the world's anti monopoly regulators and Nvidia's competitors, who worry this will give Nvidia too much power. Regulators from the us, the uk, the EU and China all oppose the deal. Nvidia is no longer just a gaming chip maker. Its tech is now used in numerous critical fields, from pharmaceuticals and engineering to car manufacturing and cloud computing. Investors see it as the chip maker with the brightest future. It's already valued higher than intel, even though intel generates more revenue for regulators. The idea of Nvidia absorbing the company that dominates the market for smartphone microprocessors is just too much. They fear it will stifle innovation and lead to exploitative prices. So in February 2022, facing fierce resistance from regulators in every major market, Nvidia abandons its bid to buy arm. Every business eventually faces the deal that doesn't close. The one you poured months, maybe years into. The real test isn't losing it. It's how you move on the next morning, notice that Nvidia didn't waste time sulking over regulators or missed opportun. It folded the lesson back into its playbook and kept Building the best leaders know when to walk away and when to double down on the reason they started chasing the deal in the first place. The arm setback doesn't slow Nvidia down. Nine months later, its far sighted investment in AI tech finally hits pay dirt. Two weeks ago, a new artificial intelligence program called ChatGPT made its debut online. This project from the OpenAI research lab can write essays and carry on convincing written conversation. So does ChatGPT represent a breakthrough that will spawn new businesses? Or is it more of a gimmick? The arrival of ChatGPT in November 2022 finally wakes the entire world up to AI. Debates about what it means for society and jobs fill the news. More than a million people sign up to use it within five days. And AI becomes the new tech gold rush. Billions of dollars of investment flood into AI companies and projects. Much of this money is spent on buying the hardware needed to build, train, and operate their AI systems. And they're mostly buying Nvidia. In May 2023, just months after ChatGPT's debut, Nvidia becomes the sixth most valuable company in the world. The following month, Nvidia's market capitalization hits $1 trillion. Nvidia spent more than a decade and an estimated $30 billion plus investing in the technology that's now fueling the AI boom. But now that long term vision and resolve is paying off big time. In 2024, the company claimed 70 to 95% of the AI chip market. And those chips are wildly profitable. Nvidia is making gross profit margins of 78% on its chips, nearly double its leading rivals. Over the next two years, Nvidia's valuation just keeps rising. It passes $3 trillion, then $4 trillion. Then in October 2025, $5 trillion. To put this in perspective, that's more than the annual value of the economy of Germany. Nvidia is now the most valuable company that has ever existed. And Jensen Huang's estimated personal net worth hits nearly $180 billion. But given the strategic importance of AI, Nvidia now finds itself caught up in global power struggles. In the final days of the Biden administration, export restrictions on AI chips were introduced. When President Donald Trump returned to the White House, he rescinded that policy, but blocked exports of the most advanced AI chips to China. Will you allow the chip maker Nvidia to sell their most advanced chips to China? No, no, we won't do that. But it's not on the table at all. We will let them deal with Nvidia. Nvidia is the prime company in the world for that. And we will let them deal with Nvidia, but not in terms of the most advanced. The most advanced. We will not let anybody have them other than the United States. The restrictions have left Nvidia locked out of the 50 billion dollar Chinese market, a situation Huang argues will only encourage China to develop its own chips rather than relying on Nvidia's. But export rules aren't the only threat. AMD, Nvidia's main rival in GPUs and AI chips, is stepping up its game. It's racing to develop chips powerful enough to pull customers away from Nvidia. AMD has also forged an alliance with OpenAI that will see the ChatGPT creator use AMD's chips and give it the option to buy a 10% stake in AMD. But it's not just rivals Nvidia needs to worry about. Companies it regards as customers are also turning into potential competitors. Amazon, Google, Meta and Microsoft are trying to build their own AI chips in order to reduce costs and avoid ending up vassals of Nvidia. And the competition from Chinese companies like Huawei can't be ignored either. But Huang isn't getting complacent. His old mantra that Nvidia is just 30 days away from bankruptcy may not fit anymore, but the mindset still does. And he's still making moves designed to take on the competition. He's made a pact with intel to co develop CPUs that are better suited to working with Nvidia's AI platforms. He's also invested billions in CoreWeave, a cloud computing provider that competes with Google, Amazon's AWS and Microsoft Azure. Nvidia still has the hardware advantage and is releasing new chips at a rate its competitors can barely keep up with. But it's also got a software advantage. Its Cuda software environment makes its hardware sticky. AI developers who switch from Nvidia's chips have to leave Cuda behind. And that means re engineering some of their code, which takes time and money. Plus, there's a risk that the alternative isn't as good as Cuda. All of which makes it hard to imagine that anyone can topple Nvidia from its position as king of the hill. But that's the thing about dominant companies. They seem unbeatable right up until they're not. After all, who 20 years ago would have predicted that a company making graphics chips for video games would become the most valuable company of all time? From wondering this is episode two of how Nvidia owned AI for business wars. A quick note about recreations you've been hearing in most cases, we can't know exactly what was said. Those scenes are dramatizations, but they're based on historical research. We've used many sources for this season, including the Invideo Way by Tae Kim and the Thinking Machine by Steven Witt. I'm your host, David Brown. This episode was written and produced by Tristan Donovan of Yellow Ant, with research by David Wolinski sound design by Kyle Randall fact checking by Gabrielle Drollet voice acting by Theodore Chin. Our managing producer is Desi Blalock. Our senior producers are Jenny Bloom and Emily Frost. Karen Lowe is our producer emeritus. Our executive producers are Jenny Lauer, Beckman and Marsha Louie. For Wondery.