For the longest time, my view on Nvidia was unequivocal. I saw it not as a triumph of innovation, but as a ghost of markets past. In every soaring stock chart, I saw the haunting silhouette of Cisco Systems circa March 2000. I was one of the loudest voices in the room—or at least in my own columns and conversations—arguing that this was a bubble. A hardware-fueled, hype-driven mania that was fundamentally unsustainable.

I pointed to the Cisco comparison as my Exhibit A, a seemingly perfect historical parallel of a company selling picks and shovels in a gold rush, destined for collapse once the prospectors went home. I seized on reports, amplified by outlets like The Motley Fool and Yahoo Finance, of so-called 'smart money' cashing out. When I read that billionaire Philippe Laffont of Coatue Management had sold over a million shares, I saw it as the ultimate confirmation of my bias—the canary in the coal mine, gasping for air. And as new whispers emerged, claiming that the explosive growth in AI was beginning to stall, I felt a sense of grim vindication. The end, I believed, was nigh.

My conviction was built on these pillars of skepticism. They were simple, powerful, and easy to defend. And I've come to realize, they were profoundly wrong.

My journey away from that certainty wasn't a sudden epiphany. It was a slow, uncomfortable process of dismantling my own arguments, brick by brick. The catalyst was an earnings call—something I usually skimmed for headline numbers to fit my narrative. But this time, I listened. I heard CEO Jensen Huang talk at length not just about chips, but about a concept he called “Sovereign AI.” He spoke of nations building their own AI infrastructure to protect their cultural and economic futures. It was a single phrase that lodged in my mind like a stone in my shoe.

That simple idea—that the customer base wasn't just a handful of Silicon Valley giants but potentially every country on the planet—created a cognitive dissonance I could no longer ignore. It forced me to go back and stress-test the very foundations of my skepticism. And one by one, they crumbled.

The Ghost of Cisco: A Flawed Analogy

One of the pillars of my argument was the belief that Nvidia was the new Cisco. It was my go-to historical lesson, my intellectual trump card. Cisco sold the routers and switches that built the internet's initial infrastructure. Nvidia sells the GPUs that build AI. The parallel seemed so obvious. Both were hardware suppliers for a technological revolution. But the more I investigated, prodded by that 'Sovereign AI' comment, the more I realized I had made a fundamental category error.

Cisco sold boxes. They were brilliant, essential boxes, but they were nodes in a network. Nvidia doesn’t just sell a GPU, a piece of silicon. It sells an entire, vertically integrated computing platform. The hardware (the H100s and Blackwells) is inseparable from the software (the CUDA programming model), the networking fabric (NVLink and InfiniBand), and the layers of application-specific libraries and AI enterprise software that sit on top. Companies aren't just buying chips; they are buying into a cohesive, walled-garden ecosystem that is now the global standard for AI development.

Comparing Nvidia to Cisco is like comparing the invention of the steam engine to a company that only manufactured railroad ties. One is a component; the other is the prime mover of an entire industrial revolution. The internet connected information. AI creates new information, new designs, new scientific discoveries. The demand it generates is not just for connectivity, but for computation on an almost unimaginable scale. My historical anchor wasn't just wrong; it was actively blinding me to the sheer scale of what was being built.

The ‘Smart Money’ Myth: A Story of Anecdote over Data

With my primary thesis shattered, I had to confront my second pillar: the narrative that 'smart money' was fleeing. The story of Philippe Laffont's sale was a perfect weapon for a skeptic. It was clean, simple, and scary. See? The billionaires are getting out while they can.

But a single data point is not a trend; it's an anecdote. Forced to look deeper, I examined the broader picture of institutional ownership. It remains incredibly high. More importantly, I started thinking why someone like Laffont might sell. When a single stock explodes in value to become an outsized portion of your portfolio, selling a fraction of it to rebalance and take profits isn't a vote of no-confidence. It's basic, prudent portfolio management. He still holds a massive stake. I had mistaken responsible financial stewardship for a fire sale.

Meanwhile, I had ignored the counter-signal: Nvidia's own strategic moves. The acquisition of CentML, a company specializing in optimizing AI models, isn’t a sign of retreat. It’s a move to deepen the moat, to make the Nvidia platform even more efficient and indispensable for its customers. I was so focused on one billionaire selling a sliver of his holdings that I missed the company spending its own capital to become more entrenched for the long term. I was watching the exits and ignoring the fortification of the castle walls.

The Stallacy of 'Stalling Growth'

Finally, I had to face my belief that the growth was inevitably hitting a wall. The numbers were astronomical, and the law of large numbers is undefeated. The narrative that the core generative AI business was slowing down felt like an obvious, logical conclusion.

This was the hardest bias to shed, but I realized it came from a misunderstanding of the market itself. I was seeing the end of the first wave and mistaking it for the tide going out. That first wave was the hyperscale cloud providers—Microsoft, Google, Amazon—buying tens of thousands of GPUs to build the massive models. That spending will moderate. But what I wasn't seeing were the next waves, already forming.

There's the enterprise wave, as every major corporation realizes it needs its own AI factory. There's the sovereign AI wave I mentioned earlier. There are new markets, like the one represented by the Cyngn partnership, putting AI into autonomous industrial vehicles. But the biggest wave of all is the shift from training to inference. Training an AI model is a massive, but finite, computing task. Inference—the act of using that model millions or billions of times a day in real-world applications—is a perpetual and exponentially larger market. Nvidia is not just a training company; its entire platform and future chip designs are built to dominate the inference market.

My belief in 'stalling growth' was a failure of imagination. I was looking at the completion of the foundation and couldn't fathom the skyscraper that was about to be built on top of it.

I am not here to offer financial advice or to tell anyone to buy, sell, or hold a stock. I am sharing a personal and humbling intellectual journey. The narratives that frame Nvidia as a bubble are seductive because they are simple. They rely on familiar history and easy-to-digest fears. I know, because I built my camp there and defended it fiercely. But they don’t stand up to a deeper, more honest appraisal of the technology and the market. We are not watching a replay of 2000. We are watching the dawn of a new form of computing, the very architecture of the next economy being assembled before our eyes. And I was too busy looking for ghosts to see it.