Nvidia is trading at a valuation that assumes AI infrastructure spending will grow exponentially forever. History suggests that's exactly the assumption that precedes spectacular crashes. Just ask Cisco Systems shareholders who watched 80% of their wealth evaporate between 2000 and 2002.

The parallel is uncomfortably precise. In 1999-2000, Cisco was the pick-and-shovel play on the internet boom, selling the routers and switches that enabled the digital revolution. The company's revenue was growing 50% annually, margins were expanding, and every analyst agreed the internet was the future. At its peak, Cisco briefly became the world's most valuable company with a market cap exceeding $500 billion.

Then reality hit. Internet infrastructure spending didn't grow exponentially, it followed an S-curve. Early adopters overbuilt capacity. Telecom companies went bankrupt. Enterprise IT budgets normalized. Cisco's stock fell from $80 to $13 in 24 months, and it took 17 years to recover to its dot-com highs.

Today, Nvidia is the infrastructure play on AI, selling the GPUs that power large language models and generative AI applications. The company's market cap has soared past $2 trillion on expectations that AI spending will compound at high rates for years. Sound familiar?

The bull case is straightforward: AI is transformational technology, enterprises need Nvidia's chips, and data center demand is booming. The bear case is equally simple: current valuations already price in a decade of perfect execution and unlimited demand growth. When infrastructure buildouts reach diminishing returns, the stock has nowhere to go but down.

Here's what the bulls miss: infrastructure spending doesn't scale linearly with deployment. Early AI adopters are building massive compute clusters. But most enterprise AI applications don't require cutting-edge GPUs once the models are trained. Inference runs on cheaper hardware. The installed base eventually satisfies demand growth.

Nvidia is also facing the same inventory dynamics that killed Cisco. When customers expect prices to fall or next-generation products to arrive soon, they stop buying today's hardware. That creates a sudden demand cliff that's invisible until it shows up in earnings.

The counterargument is that AI is fundamentally different from the dot-com era. AI has real revenue, real applications, and real productivity gains. Unlike 1999 internet startups burning cash on Super Bowl ads, AI companies are deploying working technology that generates measurable value. That's fair, but it doesn't change the valuation math.

Cisco's products were real too. The internet infrastructure they enabled created trillions in economic value. But the stock still crashed 80% because the company was priced for perfection. When growth rates normalized from extraordinary to merely good, the multiple collapsed.

Cui bono if Nvidia follows Cisco's trajectory? Short sellers, obviously. But also late-cycle investors who avoid the peak. The company itself will likely survive and thrive long-term, just like Cisco did. The question is whether today's valuation makes sense.

The numbers don't lie. Nvidia trades at 35x forward earnings with expectations for continued 50% revenue growth baked in. When that growth moderates, as it inevitably will, the stock reprices violently. History doesn't repeat, but it often rhymes, and right now it's rhyming with 1999.