AI's Dot-Com Moment—And the One Bet That Actually Makes Sense

From Pets.com to ChatGPT

Earlier this week, I wrote to you about the AI bubble—the circular financing propping it up, the staggering amount of revenue AI companies would need to justify their valuations, and Sam Altman openly positioning AI companies as too big to fail.

Like I said several times in that piece, the technology is real. But so is the bubble. Both things are true simultaneously.

But that leaves an obvious question: if AI is genuinely transformative yet wildly overvalued, how should you invest?

Let me show you.

The Dot-Com Lesson Few Remember

Most people instinctively try to pick the winners—figure out which AI companies will become the next trillion-dollar giants and get in early. And sure, it’s probably a little too late for the “getting in early” part… but it’s still tempting. The upside looks massive.

You’ve probably heard the success stories from the dot-com era: Amazon (AMZN), Apple (AAPL), Microsoft (MSFT), and Cisco (CSCO). The companies that survived the crash and went on to become trillion-dollar giants. If you’d invested $1,000 into Amazon in 1997, it would be worth roughly $2.4 million today. A $1,000 position in Apple from that same year would have grown into about $1.34 million.

Those are life-changing returns.

But here’s the problem: picking winners is only easy in retrospect. Doing it in real-time, in the middle of a bubble, is nearly impossible. And history proves it. For every Amazon that survived, there were dozens—maybe hundreds—of companies that went to zero. And at the time, it was nearly impossible to tell which was which.

Take Pets.com, the poster child of dot-com insanity.

In early 2000, Pets.com was everywhere. Its sock puppet mascot starred in a Super Bowl ad watched by 88 million people. The company raised hundreds of millions from respected venture capitalists. It went public on the Nasdaq in February 2000 with the symbol IPET at $11 per share.

Source: Museum of Failure

The business model was simple: sell pet supplies online and deliver them directly to customers’ doors. Convenient. Modern. The future of retail.

There was just one problem: the economics didn’t work. Shipping 40-pound bags of dog food to customers’ doorsteps cost more than the revenue it brought in. The company was selling products at nearly 30% below cost, following the mystifying dot-com logic of “we lose money on every sale but make it up in volume.”

By November 2000—just 268 days after its IPO—Pets.com had collapsed into bankruptcy. The stock that opened at $11 was worth 19 cents. The sock puppet became a punchline.

Now, to drive home my earlier point that AI is real technology—even inside a bubble—consider this: Pets.com wasn’t actually wrong about the future.

Today, a company called Chewy (CHWY) does exactly what Pets.com tried to do. It sells pet supplies online and delivers them to your door. And unlike Pets.com, it actually works. In 2024, Chewy generated about $11.9 billion in revenue doing precisely what Pets.com promised. The business model works. The technology was real. The timing and execution were just catastrophically wrong.

The same pattern played out across the entire dot-com landscape.

Webvan tried to revolutionize grocery delivery. It raised hundreds of millions, built massive infrastructure, went public with an $8.4 billion market cap, and promptly went bankrupt. Today? Amazon and Instacart do exactly what Webvan envisioned, generating billions in revenue.

theGlobe.com was an early social network where users could create profiles, connect with others, and share content. Its stock soared 606% on the first day of trading in 1998. By 2000, it was bankrupt. Facebook later built a $100+ billion-a-year business doing exactly what theGlobe.com tried.

Once again, the technology was real. The vision was correct. But that didn’t save investors.

And as I mentioned earlier, this wasn’t limited to a few companies. From 1995 to 2000, the tech-heavy Nasdaq shot up 372%. But those gains quickly vanished. From the 2000 peak through October 2002, the Nasdaq collapsed by a staggering 78%, wiping out portfolios across the board. It took nearly 15 years for the index to fully recover.

Even the smartest investors got it wrong.

Warren Buffett famously avoided tech stocks during the entire dot-com bubble—not because he didn’t believe in the internet, but because he knew he couldn’t identify the winners in advance.

And that’s the truth—you can’t pick the Amazons while avoiding the Pets.coms.

No one has a crystal ball that accurate. And anyone who claims they can separate the two ahead of time is seriously overestimating their predictive abilities.

Selling the Shovels

So if picking winners is a fool’s errand, what’s the smarter play?

Instead of trying to guess which AI companies will survive the inevitable shakeout—or piling into an overpriced AI-themed ETF stuffed with the same mega-caps everyone already owns—you can focus on what every single one of these companies needs to succeed:

Energy. Massive amounts of it.

Because AI is a power hog.

Take ChatGPT, for example. Every query you make consumes significantly more energy than a simple Google search—estimated at 3 to 30 times more. It’s like every question you ask burns a teaspoon of oil, or something similar in energy terms.

This is why systems like ChatGPT, Google’s Gemini, and Elon Musk’s xAI need massive warehouses of specialized computers—data centers—just to stay alive.

And these data centers aren’t using everyday hardware. They’re running on Nvidia’s AI GPUs, which burn through a ridiculous amount of energy. Ten seconds of physical interaction simulation uses more computing power than the entire Apollo program. Yes, seriously.

So when you stack it all together, the power draw gets insane. Data centers need enormous amounts of electricity just to keep these models running.

In fact, even with AI still in its early stages, the data centers that power the internet already consume about 1% of the world’s electricity. And according to the energy research journal Joule, global AI-related electricity use is projected to jump 56% by 2027—from 85.4 TWh to 134 TWh. That’s more than the annual electricity consumption of Argentina, a country of 45 million people.

The reality is that there won’t be a limit to how much electricity AI will consume—it’ll use everything we can produce.

And that, of course, is where the opportunity comes in.

Why?

Because America has a serious problem. For a superpower trying to compete with another global power in a “national-priority” technology with a bottomless appetite for energy, the U.S. is already playing from behind—and badly. Take a look at this chart showing electricity generation in the U.S. versus China from 2000 to 2024.

As you can see, China overtook the U.S. in total electricity generation back in 2010, and today it produces roughly 2.5 times more electricity than the United States.

When you can produce that much more power than your rival, you can run that many more factories, smelters, and, yes, the power-hungry data centers that drive modern AI.

Now, clearly, President Donald Trump sees that power gap as America’s biggest vulnerability in the AI race.

Last year, during a conversation on X with Elon Musk, he pointed out that the U.S. would need to double its electricity supply to keep up with China in the AI race. Take a listen.

Frankly, I think that might turn out to be a conservative estimate.

Thankfully, the U.S.—unlike energy-import-dependent China—actually has the energy. Oil. Gas. Coal. All of it.

And you can bet that because AI is now framed as strategically critical in America’s competition with China—and with Trump in office—there will be enormous political pressure to put that energy to work.

And this isn’t something that’s “coming soon.” It already started. Earlier this year, just days after taking office, Trump declared a national energy emergency and rolled out his “Unleashing American Energy” order. After that came a whole string of actions to boost production, strengthen baseload power, and finally clear out the regulatory mess choking the grid.

It’s the clearest 180° I’ve seen in a long time. And for resource investors, that’s an opportunity—because this follows the oldest rule in speculative booms: in a gold rush, the smartest move isn’t in backing the prospectors. It’s in selling the shovels. And in today’s AI rush, energy is the shovel. As demand rises across the board, you just need to own the better-positioned providers.

Have a great rest of the weekend,

Lau Vegys

P.S. The coming energy buildout is exactly why a portion of our Crisis Investing portfolio is focused on energy companies positioned to benefit from America’s push to double electricity generation. These picks—many of which Doug himself owns—stand to profit regardless of which AI companies survive the shakeout.