The past few weeks I’ve spent a fair amount of time sourcing startups. And honestly? The AI hype is even bigger than I expected.

Quick stat to set the tone: in 2020, there were ~40k .ai domains. Today? 859k. That’s a 2,000% jump in just four years. Accelerators now have batches where 90%+ of the startups call themselves “AI-first.” A growing chunk of VCs are AI-only.

Take Y Combinator. Scroll through the last few cohorts and it’s the same soundtrack: “AI this, AI that.” If you’re really lucky, you’ll get the extra spice of “Lovable for X” or “Duolingo for Y.” (Yes, sarcasm intended.).

To make sure I wasn’t just imagining things, I checked manually. I went through ~300 signals from new companies and stealth founders. And yeah — apart from a tiny handful, every single one branded itself as an “AI startup.” Sometimes it’s real tech. Most of the time? Just an API wrapper, a shiny landing page, and a very confident founder thread on X. If at all.

So I asked myself: have we ever seen hype this concentrated in one sub-sector before? And if so, when? The answer was obvious: the Dotcom Bubble in the late 90s / early 2000s. And the parallels are… hard to ignore.

Quick Disclaimer: This isn’t meant to be a Michael-Burry-style crash article. No “everything will collapse” sermon. Rather, a (nearly) sober look at where opportunities and risks lie – and where I clearly see a bubble. If you think I’m off on something (or have a completely different take) drop it in the comments. Always happy to hear other perspectives.

If you lived through it, skip this. If not, here is the quick refresher:

In the late ’90s, the internet was new, exciting, and mysterious. Everyone knew: something big was happening. But no one could say exactly what.

So companies were founded at breakneck speed. All you needed was a .com domain and a half-decent idea. Selling shoes online? Revolutionary. Cat food by mail? Jackpot. Business plans? Optional. Profits? Irrelevant.

Capital flowed like Red Bull at a nerdy startup party – fast, sweet, and with no concern for side effects. IPOs went through the roof. Billion-dollar valuations were the norm, even when revenues barely exceeded the local kebab shop.

The motto: “Doesn’t matter if it makes money – as long as it’s on the internet!” Growth at all costs, OG edition.

Then came the crash in 2000. Rates ticked up, a few analysts started asking uncomfortable questions, and suddenly investors went: “Wait… are these companies ever going to make money?” Spoiler: most weren’t (wow, surprise…).

The Nasdaq dropped 75% in two years. Pets.com became the synonym for “burning investor money.” A few survived – Amazon, eBay, Google. But 90% went under.

Sounds funny in hindsight. Yeah. And it doesn’t seem like it has much to do with today, right? But if you look closely, there are uncomfortable parallels with the current AI hype. Way too many, honestly.

To illustrate those parallels, I dug a bit deeper and decided to analyze some data. Here’s an overview of the main issues I found:

1. Domains exploding2020: ~40k .ai domains.2025: ~890k .ai domains.That’s 2,000%+ growth.

Looks wild? The exact same thing happened with .com. 1992: fewer than 15k domains. 1997: over 1M. That’s 6,600% growth in five years. The investor logic back then: “.com = money.” Today’s version: “.ai = money.”

2. Billions for hot airPets.com had a $300M market cap while losing $10M a month because shipping cost more than the product.

Today? Thinking Machines Lab, the new startup from ex-OpenAI CTO Mira Murati. Raised $2B at a $10–12B valuation.

Product? None.Revenue? Zero.

That’s Pets.com on mega steroids — with two extra zeros!!

And no, it’s not “just VC money.” VCs don’t print money. Their LPs are pension funds, state funds, institutions. In Europe, a good example is the EIF, which ultimately means taxpayer money.

Congratulations, you’re now co-funding castles in the air.

3. Paper millionaires 2.0In 2000, thousands of devs thought they were millionaires overnight — thanks to stock options at Yahoo!, Netscape, etc. Then the Nasdaq dropped 75%, and most of those fortunes vanished. Some even owed taxes on money they never actually saw. Brutal.

Fast-forward: same game today. AI startups hand out insane option packages. Employees get starry-eyed. But here’s the catch: liquidation waterfall. Investors get their money back first. Often with 2x prefs. Employees? Often nothing.

Example: You own 1% of a company worth $100M in 2019 (that means $1M on paper). The founders hold 40% ($40M on paper). Then the company raises $10M in 2020, another $15M in 2022 (2x liq pref). By 2023 the company’s worth $35M because of bad execution. Because VCs have liquidation preferences, they claim their money back first before anyone else gets paid. Investor B cashes out $30M (2x 15). Investor A gets $5M (a 50% haircut). Founders and employees: zero. Welcome to the Paper Millionaire Club.

4. Absurd salariesAnd then there’s the real money. Zuck’s “AI Avengers.” He’s throwing billions at recruiting top talent. Some of those researchers are reportedly making hundreds of millions each — in just 2–3 years. That’s not tech salaries anymore, that’s NBA contracts.

5. Accelerators going all-in on AIYC Spring 2025: 144 startups. 67 of them pure AI-agent plays. That’s 46%. And if you count the rest of the AI-first companies, it’s basically the whole batch.

And it’s not just YC. Every accelerator, every incubator, every early-stage program looks the same right now: AI, AI, AI. Substance? Usually missing. Valuations north of $30M without a real product? Definitely not missing.

Sure, some of them already show solid revenues. Fair point. And yes — technically, valuations are always “fair,” since supply and demand set the price. But let’s be honest: early ARR is super fragile. Where does it usually come from? Exactly — the classic FFF loop: family, friends, fools. In other words, distribution-by-network, not market validation.

True, durable ARR only starts showing up much later. Until then, most of these “revenue numbers” are closer to structured sales tricks than proof of product–market fit.

6. Blue: the demo problemMy favorite example: Blue.

Pitch: “The first voice assistant that can control every app on your iPhone.” Basically Make.com for mobile, powered by AI. One of the key features: speed.

And what happens in the official demo video? At the end, the disclaimer:

Yep. One of the most important feature, fast execution, doesn’t work. So they sped up the video. Valuation? Probably still somewhere between $20M and $40M.

Not picking on Blue here — maybe they’ll turn out to be the unicorn I completely missed. But they’re just one datapoint in a bigger pattern: half-baked (a few months old) products, demos with disclaimers, wrappers dressed up as “deep tech,” and valuations starting at $20M+. That’s pure bubble DNA.

Of course not. AI is real. OpenAI, Anthropic, Grok, Google – they’re building real substance. The technology saves time and transforms markets.

The problem: behind these few players are thousands of startups that are basically just wrappers on top of existing LLMs. Many never get beyond a demo that had to be “just sped up” to look impressive. Yet they fetch multi-million-dollar valuations.

And every time OpenAI or Google releases a new feature, another ten of those companies die. Products, dreams, careers – gone.

Fun fact if you like numbers: MIT recently reported that 95% of corporate AI projects fail. In other words: 19 out of 20 corporate AI projects are Pets.com. One might be Amazon.

Just like during the Dotcom Bubble, there will be winners. Amazon, Google, eBay survived back then. OpenAI, Google DeepMind, Anthropic & Co. will survive now.

But: the bulk of AI startups – inflated valuations, VC money without grounding, option packages made of air – will not.

For investors, employees, and yes, taxpayers, that means caution. Sure, seize opportunities. But the parallels to the Dotcom Bubble are too obvious to just wave the hype through.

So what does it mean? AI isn’t a fad. But today’s AI ecosystem is a bubble. And like every bubble, the technology survives – but 90% of the startups don’t. Whether this bubble bursts next year or in five, nobody knows. Until then, plenty of investors and founders will strike gold. But the collapse will come. And when it does, only a handful of real players will remain — and most others will be remembered as footnotes in the AI hype cycle.

Cheers 🥂