What the Early Web Teaches Us About AI Hype Cycles

If you were around for the late 1990s internet boom, the current AI landscape probably feels eerily familiar. The breathless predictions. The valuations untethered from revenue. The absolute certainty that this time, everything changes overnight. And the quiet, creeping suspicion that some of this might not end well.

Here's the thing: the internet did change everything. Just not on the timeline or in the ways most people predicted in 1999. And AI is going to change everything too. But the gap between "this technology is transformative" and "every company riding this wave will succeed" is where fortunes get lost.

Let's talk about what the early web actually teaches us about this moment.

The Pattern Is Older Than the Internet

Technology hype cycles aren't new. Railroads in the 1840s. Radio in the 1920s. Personal computers in the early 1980s. The internet in the late 1990s. Each one followed a remarkably similar arc: genuine technological breakthrough, followed by a flood of investment, followed by wild speculation, followed by a painful correction, followed by the technology quietly becoming foundational infrastructure.

Carlota Perez documented this pattern decades ago. Every major technological revolution goes through an installation phase (where the technology gets built and speculation runs wild) and a deployment phase (where it actually gets integrated into the real economy). The transition between those two phases is usually messy.

AI is following the same script. We're deep in the installation phase right now. The infrastructure is being built. The investment is pouring in. And a lot of companies exist primarily because they have "AI" in their pitch deck, not because they've found a sustainable business model.

What the Dot-Com Boom Got Right

Before we talk about what went wrong, let's acknowledge what the dot-com era got right. The core thesis was correct: the internet would fundamentally reshape commerce, communication, media, and daily life. The people who said "this changes everything" were not wrong. They were early, and they were imprecise about the details, but the broad strokes were accurate.

Amazon was founded in 1994. Google in 1998. Both survived the bust and became among the most valuable companies in human history. The technology was real. The transformation was real. The opportunity was real.

The AI thesis is also correct. Machine learning, large language models, and generative AI represent genuine technological capability that didn't exist five years ago. The ability to generate text, code, images, and analysis at scale is not a parlor trick. It has real applications. It creates real value. The people saying "this changes everything" are, broadly speaking, right.

But being right about the technology and being right about the business are two very different things.

What the Dot-Com Boom Got Wrong

The dot-com era produced Pets.com, Webvan, Kozmo.com, and hundreds of other companies that burned through billions of dollars and left nothing behind. The failure mode wasn't that the internet was fake. It was that people confused "the technology works" with "any business built on this technology will work."

Here's what actually went wrong:

Mistaking access for advantage. Everyone could build a website. That wasn't a moat. The companies that survived had something beyond just being on the internet. They had network effects, logistics advantages, data flywheels, or genuine technical innovation. Being online wasn't enough.

Confusing growth with sustainability. Dot-com companies measured eyeballs, page views, and user growth. Revenue was an afterthought. "Get big fast" was the strategy, with monetization deferred to some vague future date. For most, that date never came.

Overestimating adoption speed. In 1999, people predicted e-commerce would dominate retail within a few years. It took closer to two decades. Technology adoption follows S-curves, and the middle of the S is a lot slower than the beginning suggests.

Ignoring unit economics. Kozmo.com delivered snacks to your door in under an hour. The technology worked. The economics didn't. Delivering a $5 bag of chips with a human courier was never going to be profitable, no matter how much venture capital you threw at it.

Sound familiar?

The AI Parallels Are Hard to Ignore

Look at the current AI landscape through that lens:

Mistaking access for advantage. Everyone can call the OpenAI API. That's not a moat. Wrapping a thin layer around a foundation model and calling it an AI product is the 2024 equivalent of putting up a website in 1999 and calling it an e-commerce company. The companies that will survive need something beyond API access: proprietary data, deep domain expertise, genuine workflow integration, or network effects.

Confusing growth with sustainability. AI companies are growing users fast. But many are subsidizing usage with venture capital. The actual cost of running large language models is enormous. When the subsidy ends, retention will tell the real story.

Overestimating adoption speed. Enterprise AI adoption is slower than the headlines suggest. Most companies are still running pilots. Regulatory uncertainty is real. Integration with existing systems is hard. The people who think every company will be "AI-first" by 2027 are making the same timing error as the people who thought everyone would shop online by 2002.

Ignoring unit economics. Training and running large models costs real money. A lot of it. If your AI feature costs $0.50 per query to run but your user pays you $10 per month and makes 200 queries, the math doesn't work. And "the models will get cheaper" is a hope, not a business plan.

What Survived the Dot-Com Bust (and Why)

The companies that survived the dot-com crash shared certain characteristics:

Real revenue from real customers. Amazon was actually selling things. eBay was facilitating real transactions. Google was selling ads against genuine search intent. The revenue was real, even if the valuations were stretched.

Genuine competitive advantages. Amazon had logistics. Google had PageRank. eBay had network effects. These weren't just "internet companies." They were companies that used the internet to do something specific better than anyone else.

Willingness to be boring. The survivors weren't chasing the flashiest ideas. They were solving real problems, often mundane ones, and doing it well. Selling books. Connecting buyers and sellers. Finding information. Not glamorous, but valuable.

Capital discipline. Not all of them, but the survivors generally had a plan for how to become profitable. They might have been burning cash, but they understood unit economics and had a path to sustainability.

What Will Survive the AI Shakeout

Apply those same filters to the current AI landscape:

The companies that survive will have real revenue from customers who would genuinely miss the product if it disappeared. Not free tier users. Not people experimenting. Customers who have integrated the product into their workflow and would be willing to pay more to keep it.

They'll have competitive advantages beyond API access. Proprietary data, deep domain expertise, workflow lock-in, network effects, or genuine technical innovation at the model level. "We use AI" is not an advantage. "We use AI to do this specific thing better than anyone else because of our unique data and domain knowledge" might be.

They'll be willing to be boring. The AI companies solving mundane business problems, like making accounting faster, improving customer support, or automating repetitive data entry, will outlast the ones promising artificial general intelligence by next quarter.

And they'll have a credible path to sustainable economics. Not "we'll figure out monetization later." Not "the models will get cheap enough eventually." A real plan, with real numbers, for how the business works when the venture capital stops flowing.

The Good News

Here's what the dot-com parallel should make you optimistic about: the technology is real, and the long-term transformation is real. The internet did change everything. It just took longer than people thought, happened differently than people predicted, and the winners weren't always who people expected.

AI will follow the same pattern. Five years from now, we'll use AI tools every day in ways we take for granted, just like we use the internet every day without thinking about it. The transformation will be genuine. But a lot of today's AI darlings won't be around to see it.

The dot-com bust didn't kill the internet. It killed the companies that were cosplaying as internet businesses without having real businesses underneath. The AI correction, whenever it comes, will do the same thing.

What This Means for You

If you're building on AI, ask yourself the dot-com survivor questions: Do I have real customers? Is my advantage more than API access? Am I solving a boring, real problem? Do my unit economics work?

If you're investing in AI (your money or your career), look for the same signals. The company that's quietly automating insurance claims processing with AI and has 200 paying enterprise customers is a better bet than the one with a slick demo and 10 million free users.

If you're adopting AI tools, don't bet your workflow on any single vendor. The landscape is going to consolidate. Some of your current tools won't exist in three years. Build flexibility into your stack.

And if you're watching the hype and feeling skeptical, you're not wrong to be skeptical. But don't confuse "the hype is overblown" with "the technology doesn't matter." The internet skeptics of 1999 were right that the bubble was going to pop. They were wrong if they concluded the internet was a fad.

AI is the internet of this generation. The hype is real. The technology is also real. Learning to separate those two things is the most valuable skill you can develop right now.

The dot-com bust hurt a lot of people. The companies that failed took real money, real jobs, and real careers with them. The AI shakeout will do the same. But on the other side of that correction, the technology will be more useful, more accessible, and more integrated into daily life than most people can imagine today.

That's the lesson of every technology revolution. The hype misleads. The correction hurts. And the transformation happens anyway.