I Was Convinced Nvidia Was a House of Cards. Here’s Why I Was Wrong.

For the longest time, my view on Nvidia was unequivocal. I was a professional skeptic, and frankly, the story wrote itself. I’d see the breathless headlines about trillion-dollar valuations and I would roll my eyes. To me, and to many others I spoke with, the warning signs were flashing bright red, and I wasn’t quiet about it. My narrative was built on what I saw as three undeniable truths.

First, I was convinced by the reports of key customers seeking alternatives. When news broke that a titan like OpenAI was actively using Google’s TPUs to cut costs and avoid vendor lock-in, I saw it as the first major crack in the dam. This wasn’t a rumor; it was a concrete, strategic move by the most important player in the AI space. It confirmed my bias that Nvidia’s dominance was a temporary monopoly, propped up by scarcity, that would inevitably crumble as the market matured and rationalized.

Second, the drumbeat of insider stock sales was deafening. Over a billion dollars cashed out by the very executives who were telling the world the journey had just begun. I read the explanations about pre-arranged selling plans, but the sheer volume felt like a tell. It was, in my mind, the ultimate red flag—a classic case of the smart money getting out at the peak while the public piled in. How could you believe in a $6 trillion future if the leadership was taking so much off the table now?

Finally, I was persuaded by the broader strategic argument, a view championed by influential minds like Masayoshi Son: the ultimate value in a gold rush accrues to the gold miners, not the company selling picks and shovels. I saw Nvidia as the ultimate shovel-seller. Essential for a time, yes, but destined to be commoditized and overshadowed by the true innovators, like OpenAI, who were building the world-changing applications on top of their hardware.

This was my story. I believed it. I argued it. And then, a single conversation forced me to rethink everything.

It was a late-night call with a friend, a systems architect at a major cloud provider—someone deep in the trenches, not just analyzing from afar. I was laying out my case, feeling smug about my airtight argument, when she patiently stopped me. “You’re looking at the chessboard,” she said, “but you’re not seeing the universe it’s in.” That one line was the catalyst. It forced me to move past the headlines and confront the uncomfortable complexity I had been conveniently ignoring.

My first pillar of skepticism—the OpenAI “defection”—was the first to fall. In my quest to prove Nvidia’s vulnerability, I started digging into the technical specifics of AI workloads. I had seen “OpenAI uses Google TPUs” and interpreted it as “OpenAI is abandoning Nvidia.” The reality I discovered was profoundly different. AI isn’t a monolithic task. There’s the brutal, computationally-intensive process of training these massive models from scratch, and then there’s the comparatively lighter task of inference, where the trained model answers a query. OpenAI was using TPUs for some inference workloads because it was more cost-effective for that specific task. But where were they, and everyone else, still going for the bleeding-edge training that creates the next generation of models like GPT-5? Nvidia. Exclusively.

It was a difficult realization. What I had framed as a defection was, in fact, diversification—the hallmark of a healthy, maturing ecosystem. A company like OpenAI would be irresponsible not to explore a multi-vendor strategy for different parts of its workflow. Nvidia’s position isn’t one of a fragile monopoly, but of the undisputed gold standard for the most difficult, most valuable part of the AI equation. Their dominance isn't in locking customers in, but in being the only platform powerful and developed enough to unlock the future. My certainty began to crumble.

Next, I had to confront the insider selling. The billion-dollar figure had seared itself into my brain as a sign of no confidence. But my architect friend had urged me to look at percentages, not just shocking totals. So I did. I pulled up the filings. Yes, CEO Jensen Huang and others had sold millions of shares. But then I looked at the shares they kept. In almost every case, the sales represented a tiny fraction of their total holdings. These executives weren't cashing out; they were still among the most heavily invested individuals on the planet, with the vast majority of their net worth tied directly to Nvidia’s future success.

I realized I had let my narrative bias misinterpret basic financial prudence. When your personal stock holdings grow by thousands of percent to constitute an overwhelming portion of your wealth, selling a small, pre-scheduled percentage isn’t a red flag; it’s a fiduciary duty to your own family. The pre-arranged 10b5-1 plans I had dismissed were actually the proof of this—a systematic, transparent method of diversification, not a panicked sprint for the exit. My “lack of confidence” theory was, in truth, a gross misreading of wealth management at an unimaginable scale.

Finally, I had to dismantle my favorite analogy: the picks and shovels. It felt so right, so historically sound. But it’s a deeply flawed comparison. A shovel is a simple, static tool. It doesn’t get better or change the nature of gold itself. Nvidia’s platform is the opposite. It’s not just a chip; it’s a co-evolving ecosystem of hardware, software (CUDA), and networking that is in a constant, symbiotic relationship with the AI models being built. A breakthrough at OpenAI doesn’t just use Nvidia’s current chips; it creates an immediate demand for the next, more powerful generation of chips that Nvidia is already developing. They aren’t just selling tools to the miners; they are inventing new physics that allows for the discovery of new kinds of gold.

Nvidia isn’t the shovel. It’s the entire technological substrate. It’s the geology, the metallurgy, and the physics of the AI revolution, all at once. The value isn’t just in the application; it’s in the foundational platform that makes exponential progress possible. The game isn’t zero-sum.

I’m not writing this to offer investment advice or to claim that Nvidia’s path is without risk. I am writing this to confess that my deeply held skepticism was built on a foundation of surface-level readings and simplistic analogies. I was so committed to the idea of a bubble that I failed to see the fundamental technological shift happening underneath. The real story isn't about market hype; it's about the construction of a new computational era. It was a difficult and humbling journey to see that, and my view is now clearer for it.