Latest / The CEO Diary with Fexingo: Leadership Lessons, Executive Decisions, and Corner Office Stories / How Nvidia Kept the Long Bet Alive
Transcript
- Lucas: There is a moment, I think, in every great CEO's tenure where they make a decision that looks borderline irrational at the time, and then, years later, it seems inevitable. Jensen Huang's decision to build CUDA in the mid-2000s is probably the cleanest example in modern tech. Luna: And at the time, Nvidia was just a graphics card company. I mean, they made chips that made video games look better. That was the business. Lucas: Right. And Huang's insight — which he's talked about in several interviews — was that the parallel processing power inside GPUs could be repurposed for general-purpose computing. Not just graphics, but physics simulations, scientific computing, eventually machine learning. But to make that happen, developers needed a programming model. So Nvidia built CUDA. Luna: That's the Compute Unified Device Architecture. And it was essentially a software platform that let developers write code that ran on the GPU, not just the CPU. But building it cost them hundreds of millions of dollars a year for years with almost zero revenue. Lucas: Exactly. And Wall Street hated it. You had analysts saying, why is a graphics company spending all this money on a platform nobody's using? In 2008, Nvidia's stock dropped something like 70 percent during the financial crisis. They were burning cash. There was serious talk about whether they'd survive. Luna: I remember reading that in 2009, Huang said publicly that CUDA was 'a bet the company' decision. He basically told employees that they were going to keep investing in it even if it took ten years to pay off. Lucas: And that's the part that fascinates me, because most CEOs wouldn't have the conviction to stay that course. Especially when, in 2013, Intel reportedly offered to buy Nvidia's GPU business for something like $6 billion. Huang said no. Luna: And Nvidia's market cap today is over three trillion. So that was a pretty good call. Lucas: One of the best in business history. But the interesting thing is that CUDA's payoff didn't come from the applications Huang originally imagined. He thought it would be used for scientific computing, for oil and gas exploration, for medical imaging. And some of that did happen — but the real breakthrough was deep learning. Luna: Right, around 2012, researchers at the University of Toronto used Nvidia GPUs to train a neural network that crushed the competition in the ImageNet challenge. That was the AlexNet moment. Lucas: And suddenly, the entire AI research community realized that GPUs were the only way to train these models at scale. Huang had built this platform for a market that didn't exist yet, and then the market arrived. But here's the thing — he didn't know that was going to happen. He just believed that parallel computing was the future. Luna: It's a reminder that long-term bets require a certain faith in your own thesis. I mean, CUDA was a massive engineering investment. Nvidia had to hire hundreds of software engineers, build a compiler, create an entire developer ecosystem. They released a free toolkit, held training workshops, even funded academic research. Lucas: And a lot of that was done in a period when the company's core gaming business was going through cycles. When console sales dipped, Nvidia's revenue dipped. So you had this volatile core business funding a long, expensive R&D project with no clear return. I think that's the part that's underappreciated about Huang's leadership — he was willing to tolerate financial pain for a strategic vision. Luna: What about the competition? AMD had a similar opportunity. They had GPUs. They could have built a competing platform. Lucas: They tried. AMD launched something called Stream SDK, which later became ROCm. But they never committed the same level of resources. And part of that is because AMD was always a smaller company with less margin for error. But I also think it's a cultural thing. Huang has this almost obsessive focus on the long term. He's said in earnings calls that he doesn't manage to the quarter, he manages to the decade. Luna: That's easy to say when your stock is up 200 percent in a year. But he was saying that in 2010 when the stock was at $5. Lucas: Exactly. And that consistency is what separates the CEOs who are genuinely long-term thinkers from the ones who just talk about it. So here we are in 2026. Nvidia is the backbone of the AI infrastructure buildout. Every major cloud provider — Amazon, Microsoft, Google — they're all buying Nvidia's H100 and B200 chips as fast as they can make them. And it all traces back to a decision made almost twenty years ago. Luna: I want to ask about the risk of that approach though. Because not every long bet pays off. Intel bet on mobile chips with Atom and basically wasted billions. Google Glass was a visionary bet that flopped. Lucas: Absolutely. And I think that's the other lesson. Huang's bet was on a technology trend — the shift from scalar to parallel computing — that he believed was inevitable. He wasn't betting on a specific product or a specific market. He was betting on a fundamental shift in how computing works. And when you frame it that way, the risk is actually more calculated than it seems. Luna: So it's a thesis, not a guess. But still, the execution mattered. Nvidia had to keep improving the hardware, keep expanding the software stack, and keep the developer community engaged. They did all of that. Lucas: And they did it while maintaining their lead in graphics. That's the other part. CUDA didn't replace the core business; it augmented it. The same GPU architecture that powers the GeForce gaming cards also powers the Tesla data center cards. So they were able to spread the R&D cost across two large markets. Luna: That's a smart capital allocation play. And it's something that a lot of companies miss when they try to pivot. They abandon the old business before the new one is ready. Lucas: Right. Kodak invented the digital camera but couldn't figure out how to transition without destroying their film business. Nvidia managed to hold onto both. Although, full disclosure — their gaming segment has become a smaller share of revenue as data center has exploded. In their most recent fiscal year, data center revenue was over $100 billion, dwarfing gaming. Luna: So the bet paid off beyond anyone's expectations. And now Jensen Huang is often called the godfather of AI. But I think the real lesson for CEOs is about patience and conviction. Not every company can afford to lose money on a platform for a decade, but every CEO can ask themselves: what is the long-term bet that my company should be making? Lucas: And whether they have the stomach to see it through. Because there will be quarters where the stock drops, where analysts are screaming, where employees are skeptical. That's when leadership matters most. Luna: On that note — and I know this is a little off-script, but it's something I've been thinking about — this show is itself a long-term bet. We've been doing it for over a year now, and the response has been incredible. And it only exists because of listener support. Lucas: That's true. We keep things ad-free, which is rare in the podcast world. And that's possible because a relatively small group of listeners chip in monthly over at buy me a coffee dot com slash fexingo. It's not a big ask — it's just the reality that listener-funded shows like this depend on that kind of support to keep going. Luna: If these conversations have added something to your work or your thinking, that's the way to make sure they continue. No pressure, just something to consider. Lucas: And speaking of continuing — one thing I want to circle back to on Nvidia is the question of what happens next. They've built this massive moat with CUDA, but competitors are finally getting serious. AMD's ROCm is improving. There are startups like Cerebras and Graphcore building custom AI chips. And cloud providers like Google and Amazon are designing their own chips too. Luna: So the question is: can Nvidia maintain its dominance? Or is the CUDA moat starting to erode? Lucas: I think the moat is still very deep. CUDA has a decade-plus head start. There's an entire ecosystem of software libraries, frameworks, and tools that are optimized for Nvidia hardware. If you're a data scientist or a machine learning engineer, your workflow almost certainly involves CUDA. Switching costs are enormous. Luna: But that doesn't mean they can rest. Jensen Huang has said that Nvidia is 'only 30 days away from going out of business' if they stop innovating. It's a dramatic statement, but it captures the paranoia that keeps them pushing. Lucas: And they are pushing. The new Blackwell architecture, the Grace Hopper superchip, the push into AI inference — they're not just selling chips, they're selling complete systems. They want to be the full stack for AI, from hardware to software to networking. Luna: It's a remarkable story. And it started with a bet that, at the time, seemed almost foolish. Lucas: That's why it's a CEO diary episode worth telling. Because the decisions that look foolish often turn out to be the most important.