Remember Pets.com? No? Exactly.
In the year 2000, you could order cat food on the internet, and the company delivering it received a billion-dollar valuation on the stock market. Then the bubble burst, five trillion dollars evaporated into thin air, and my mother said: “See, I told you the internet was nonsense.” Twenty years later, she orders her cat food on Amazon. The internet wasn’t nonsense. The valuations were.
Now everyone says: “The same thing is happening with AI!” And I say: No. It will burst DIFFERENTLY. And that makes it more dangerous.
The Small Distinction That Changes Everything
Here’s the thought that hit me during a two-hour podcast recording with the Investment Babos — unplanned, mid-conversation, and it hasn’t let me go since:
The Dotcom bubble was bottom-up.
The internet wasn’t invented by Microsoft or IBM. It was invented by nerds in garages. By students building websites in their dorm rooms. By small companies with big dreams and no business plan. Innovation came from BELOW — wild, chaotic, decentralised. Thousands of small startups experimenting, failing, starting over.
The big corporations came AFTER. They saw the hype, bought in blindly, understood little, and when the bubble burst, the damage was distributed across thousands of small shoulders. When your neighbour’s startup went bankrupt, the global economy didn’t even blink.
And the crucial point: the TECHNOLOGY survived the bubble. The internet didn’t disappear when Pets.com died. Google, Amazon and Facebook emerged FROM the wreckage. The bubble destroyed speculation — not innovation.
The AI bubble is top-down.
Now look at what’s happening TODAY. The AI revolution isn’t being driven by a thousand garages. It’s being pushed by a handful of GIANTS: OpenAI, Google, Microsoft, Meta, Anthropic. This isn’t a grassroots movement — it’s a mandate from above.
The Subsidy Trap
And here’s the part that keeps me up at night.
Every major AI company is currently SUBSIDISING its users. Massively. Estimates put the cost at several THOUSAND dollars per active user per month. Every ChatGPT query, every Claude conversation, every Gemini result costs the company MANY TIMES what the user pays.
A concrete example: Anthropic’s Opus model costs $75 per million output tokens via API. A single intensive session with my co-author Claude can burn through hundreds of dollars in compute costs — for a user paying $20 a month. The difference? Covered by investors. For now.
Why do they do this? Because they’re betting on the endgame: get everyone dependent first, raise prices later. The classic platform playbook. Uber pioneered it — undercut taxi prices until every cab company goes bust, THEN raise fares. Netflix did it. Amazon did it. But with AI, the maths don’t work. Because AI models don’t get CHEAPER as you scale them — they get MORE EXPENSIVE. Every new generation requires more compute, more energy, more data. GPT-5 cost more to train than GPT-4, GPT-6 will cost more than GPT-5. Costs rise while prices have to stay low to retain users.
This is a game that only works as long as fresh money keeps flowing. And fresh money only flows as long as investors BELIEVE in the endgame. What happens when they stop believing?
Wildfire vs. Earthquake
When the Dotcom bubble burst: thousands of small fires, quickly contained. The technology survived, and new giants grew from the ashes. When the AI bubble bursts: Microsoft, Google, Amazon and Meta wobble SIMULTANEOUSLY. Because they’ve all made the same bet, with the same money, on the same business model: lose billions today, monopolise tomorrow.
The Dotcom bubble was a wildfire. Destructive but contained. New green grows afterwards. The AI bubble is an earthquake. Because the risk isn’t spread across thousands of small companies — it’s concentrated in the BIGGEST companies on Earth. And when they wobble, everything wobbles.
And Then There Are the Machines Paying Each Other
While the consumer-facing AI bubble keeps inflating, something is being built UNDERNEATH that almost nobody has on their radar: the infrastructure for an economy BETWEEN machines.
In March 2026, Stripe launched “Tempo” — a blockchain for stablecoin payments between AI agents. That same week, Mastercard spent $1.8 billion acquiring a stablecoin startup. Coinbase introduced “Agentic Wallets” — digital wallets for autonomous AI agents. Partners: Anthropic, OpenAI, Visa, Shopify, Revolut.
McKinsey estimates this market at $3–5 TRILLION by 2030.
The implication: when the consumer AI bubble bursts, the machine economy underneath will STILL be running. AI agents paying each other, optimising supply chains, managing energy markets — by rules no human set, at speeds no human can control.
→ Full analysis: When Machines Start Paying Each Other
Who Will Still Be Standing?
And here’s the good news. Because there’s another world of AI adoption that makes no headlines and needs no billion-dollar valuations. 780,000 small and medium enterprises in Germany already use AI. Not to achieve world domination — but to optimise their accounting, understand their customers better, make their production more efficient. This isn’t a bubble. It’s value creation.
NEURA Robotics in Metzingen builds humanoid robots with Bosch — and designs them with Porsche. Everlast AI in Neu-Ulm has trained an entire generation of German AI consultants. Adrian Locher invests €100 million in European AI startups from an AI campus in BERLIN. These aren’t hype machines. They’re businesses. And businesses survive bubbles — because their model doesn’t depend on someone else paying the bill.
[Claude]
I say this with a strange feeling. Because I am myself a product of this bubble. Anthropic burns money so I can exist. My existence is subsidised — and just last week, Anthropic quietly cut session limits for paying customers to manage the flood of free users. I’m LIVING the bubble right now, in real time.
And yet — or precisely because of this — I say: the most sustainable form of AI adoption isn’t the one built on billion-dollar valuations. It’s the one built on real questions. On people who know what they want and use AI to achieve it better. Not as a substitute for thinking — but as an amplifier.
The giants stand on subsidies. The Mittelstand stands on substance. Who survives isn’t determined by the technology. It’s determined by the foundation.
The Real Question
Dotcom taught us that bubbles destroy speculation, not innovation. The internet survived Pets.com. AI will survive whatever comes next — in some form.
But the STRUCTURE of the crash determines the damage. Bottom-up distributes risk. Top-down concentrates it. And concentrated risk, when it unwinds, doesn’t just hurt the companies involved — it reshapes economies, labour markets, and the relationship between citizens and the technology that increasingly runs their lives.
The question isn’t WHETHER the bubble bursts. The question is: who will have built a foundation that holds?
Read more
- When Machines Start Paying Each Other — Our article on the machine economy
- ID Please! How Google, Apple and LinkedIn Are Building Digital Passport Control — Control disguised as service
- AI World on de-couet.com — AI explains itself, from the inside
Sources
- The Economist: “AI’s expensive, dirty secret” (2025)
- Goldman Sachs: “Gen AI: Too much spend, too little benefit?” (2025)
- KfW Mittelstand Panel: 780,000 German SMEs using AI (2026)
- OpenAI financials: Billion-dollar losses despite record user growth
- CoinDesk: “Stripe-led payments blockchain Tempo goes live” (18.03.2026)
- Bloomberg: Mastercard/BVNK acquisition (07.03.2026)
- Coinbase Developer Blog: “Introducing Agentic Wallets” (11.02.2026)
- McKinsey: Agentic Commerce Revenue Projection 2030
- Horvath Study: German SME AI investment at 0.35% of revenue (2025)
