The Three 'Red Flags' That Made Me an Nvidia Skeptic, and Why I Was Wrong About All of Them

For the longest time, my view on Nvidia was unequivocal. I was a skeptic, armed with what I believed were undeniable, concrete facts. I wasn't just a casual doubter; I was convinced the company was a magnificent house of cards, built on unsustainable hype and teetering on the edge of a great fall. I saw the headlines about key customers diversifying, the billion-dollar insider sell-offs, and the intellectual arguments positioning them as a mere toolmaker, and I nodded along. Each report felt like another brick in my wall of certainty. It all made perfect, logical sense: the emperor, for all his processing power, had no clothes.

My argument, which I shared with colleagues and friends, rested on three pillars. First, the concrete reporting that OpenAI, the undisputed leader in the AI space, was actively turning to Google’s TPUs to cut costs. To me, this was the smoking gun. If the world’s most advanced AI lab was strategically reducing its reliance on Nvidia, it signaled the end of the company’s monopoly and pricing power. The moat, I argued, was being breached. Second, the relentless drumbeat of insider stock sales, cresting over a billion dollars. This wasn't just noise; it was a clear signal from the people who knew the company best that the peak was in. And third, the compelling 'picks and shovels' analogy, championed by brilliant investors like Masayoshi Son. Why own the company selling the gear, the argument went, when the real, lasting value would be captured by the gold miners like OpenAI? These weren't just threats; to me, they were the writing on the wall. I was so sure, and now, I feel a professional and personal obligation to say: I was profoundly wrong.

My transformation didn't happen overnight. It began with a crack in my first, and strongest, pillar of skepticism: the OpenAI-TPU narrative. The story from TechPowerUp felt like the final nail in the coffin of Nvidia’s indispensability. But a late-night conversation with a friend, a data center architect who lives and breathes this hardware, changed my entire frame of reference. I laid out my case, expecting a knowing nod. Instead, she laughed.

“You’re thinking of this like a street fight between two companies,” she told me. “It’s not a fight. It’s an exploding universe.” She explained that the global demand for AI compute is expanding at a rate that is difficult for outsiders to comprehend. It’s not a finite pie they’re fighting over; the universe of that pie is doubling every few months. OpenAI turning to TPUs for some of its inference workload isn't a rejection of Nvidia; it's a testament to the astronomical, world-altering scale of the compute they now require. It's a “both/and” reality, not an “either/or” choice. The heavy lifting, the bleeding-edge model training that pushes the frontiers of what’s possible, overwhelmingly remains the domain of Nvidia’s CUDA platform. Using other chips for less-demanding, high-volume inference tasks is simply smart, workload-specific optimization. It’s what any mature, hyper-scale company would do. My 'smoking gun' wasn't a sign of Nvidia's weakness, but a symptom of a market growing so vast and so fast that it requires a multi-faceted ecosystem to support it—an ecosystem where Nvidia still provides the foundational, irreplaceable engine of innovation.

With my primary argument weakened, I turned a more critical eye to my other certainties. The insider selling. Over a billion dollars. The optics are, admittedly, terrible. It’s the easiest red flag to point to. But out of a sense of journalistic diligence, I forced myself to do what I had previously avoided: I read the dry SEC Form 4 filings. I dug into the details of the 10b5-1 trading plans under which these sales were executed. What I found wasn’t a story of panic, but one of prudence. These are pre-arranged, automated selling plans, often set up months or even years in advance, precisely to avoid any suggestion of trading on non-public information. Many of these executives have been with the company for decades, with the vast majority of their personal wealth locked into company stock. To see the stock appreciate by 10x or more and not diversify a small fraction for their families, for philanthropy, for life, would be financial malpractice. The billion-dollar headline is shocking, but it’s a function of the stock’s meteoric rise, not the executives' lack of faith. When you shift your focus from the absolute dollar amount to the percentage of their total holdings being sold, the narrative flips. They aren’t cashing out; they are rebalancing, while still leaving the vast majority of their fortunes tied directly to Nvidia's future success.

This left me with the final pillar: the elegant, intellectual 'picks and shovels' argument. It’s so compelling because it feels historically true. During the gold rush, it was the merchants like Levi Strauss who got rich, not most of the miners. But this analogy, I’ve come to realize, is a profound and dangerous oversimplification of what Nvidia actually is. The catalyst for this final realization was forcing myself to watch a full, five-hour GTC keynote from CEO Jensen Huang. I went in expecting a hardware sales pitch. I left feeling like I’d just had the architecture of a new industrial revolution explained to me.

Nvidia isn’t selling shovels. It’s not even selling advanced, robotic mining machines. Nvidia is selling the entire physics of mining. It has built the foundational operating system—the very ground upon which every AI 'gold miner' is forced to build their operations. The CUDA software platform, a decade-deep and profoundly complex ecosystem of libraries, compilers, and tools, is a moat so wide and deep that competitors are finding it nearly impossible to cross at scale. An application company like OpenAI can be brilliant, but its brilliance is expressed and executed on top of Nvidia's full-stack platform. Nvidia is not just a chipmaker. It’s a systems company, a software company, a data center architecture company, and an AI research lab rolled into one. To call it a 'picks and shovels' provider is like calling the person who invented the steam engine a simple metalworker. The analogy collapses under the weight of this reality. They are the platform, the standard, the fundamental infrastructure upon which this entire new economy is being built. In this kind of revolution, owning the foundational platform is the most powerful and durable position of all.

I was wrong because I was reading the headlines and not the footnotes. I was captivated by simple analogies instead of studying the complex architecture. My skepticism was born from a surface-level view that, I confess, was easy and convenient to hold. Dismantling it required a deeper, more difficult inquiry, and the realization that the story of Nvidia isn't about hype, but about a three-decade-long journey of deep, foundational, and often inscrutable engineering. It's a lesson in humility, and a reminder to always question the narrative—especially the one you find most comfortable.