Amsterdam, 1637: A Tulip Bulb Worth Ten Years of a Craftsman's Wages
A Dutch broker trades three hectares of fertile land for a single Semper Augustus bulb. A few weeks later, the market collapses. The bulbs are worthless. The land still produces wheat.
Four centuries later, replace the bulb with a prediction token. The mechanism is identical: a real technology, a speculative frenzy disconnected from fundamental value, then the brutal correction. And here we are.
After 20 years in tech, I've watched this scenario repeat with a regularity that markets stubbornly refuse to acknowledge. 2000, the dot-coms. 2008, the subprimes. 2017, crypto. 2021, NFTs. Every time, the same pattern: a legitimate innovation, collective euphoria, a collapse, then reconstruction, by those who had kept their heads cool.
AI in 2026 checks every box of a major speculative bubble. Here's why, and more importantly, here's what clear-eyed leaders need to do now.
The Numbers That Should Keep You Up at Night
Valuations Disconnected from Any Economic Reality
AI startups are trading today at 25 to 30 times their revenues. The historical norm for a growing technology company is 5 to 8 times. OpenAI is valued at $300 billion with $5 billion in annual losses. Safe Superintelligence raised $32 billion without having a product. Read that sentence again: $32 billion, zero product.
In 2000, Pets.com was worth $300 million without a viable business model. The scale has changed. The irrationality hasn't.
Extreme Market Concentration
The "Magnificent 7" (Apple, Microsoft, Alphabet, Amazon, Nvidia, Meta, Tesla) represent 33% of the S&P 500. In 2000, at the peak of the dot-com bubble, the concentration of technology leaders reached 15%. We're at more than double that.
Nvidia alone now exceeds the GDP of Germany. A graphics chip manufacturer is worth more than the world's fourth-largest economy. When a ratio reaches this level of absurdity, history teaches us that the correction isn't a question of "if," but "when."
The Massive Failure Nobody Wants to See
Here's the number the AI dream-sellers will never put in their pitch deck: 95% of generative AI projects in enterprises fail, according to a 2025 MIT study. For every dollar of value generated, five dollars are spent. And $364 billion has already been burned on AI data centers.
Sam Altman himself acknowledges that "investors are overexcited." Goldman Sachs observes $1 trillion spent for "few tangible results." 57 S&P 500 companies publicly declare they fear never recouping their AI investments.
Anatomy of a Bubble: The Four Invariant Phases
Every speculative bubble follows the same cycle, theorized by economist Hyman Minsky.
Phase 1 : Displacement. A real innovation appears. The internet in 1995. Mortgage securitization in 2003. Blockchain in 2015. Transformers and GPT in 2022. The innovation is genuine, its transformative potential is real. That's what makes the bubble so convincing.
Phase 2 : The Boom. Money floods in. Valuations soar. Media amplifies. "This time it's different" becomes the mantra. The $364 billion in data centers, the fundraising rounds at 30 times revenue : we're right in the middle of this phase.
Phase 3 : Euphoria. Warning signs are ignored. Those who express doubts are sidelined. Regulators hesitate. The last investors, the least informed, enter the market. Trump orchestrates what I call the greatest acceleration of tech capital destruction in history, pushing American companies toward massive AI investments with underpriced risk.
Phase 4 : The Correction. It always comes. And it hits hardest those who invested without a strategy.
We're oscillating between Phase 2 and Phase 3. The window for strategic action is still open, but it's closing.
The European Trap: Between Brain Drain and Underinvestment
Europe watches this bubble from the stands, and that position isn't without danger either.
European VCs invest ten times less than their American counterparts in AI. Our talent leaves the continent for packages our structures can't match. Meanwhile, China, constrained to efficiency by American sanctions, produces remarkable results with a fraction of the budgets : DeepSeek-R1 is the striking proof.
Europe thus faces a dual risk: failing to capture AI's real value (which will exist after the bubble bursts, just as Amazon and Google existed after 2000) while importing the inflated costs of the American bubble through its dependencies on US hyperscalers.
The 200-billion-euro AI Continent Action Plan is a necessary response, but it must be executed with the discipline of a post-bubble investor, not the euphoria of a pre-correction speculator.
Playbook for Clear-Eyed CIOs: Five Decisions to Make Now
When the dot-com bubble burst, the companies that survived, and thrived, weren't those that had invested the most in technology. They were those that had invested in the right fundamentals. The same principle applies today.
1. Audit your foundations before adding AI. Are your business processes documented and under control? Are your data clean, governed, accessible? If your foundations are fragile, AI will only amplify your dysfunctions. This is the most important lesson and the most ignored.
2. Demand measurable ROI at 6 months. Every AI project must demonstrate a quantifiable P&L impact within six months. Not an "interesting proof of concept." Not a "strategic exploration." A financial result. The 5% of projects that succeed according to the MIT study all share this discipline.
3. Prioritize back-office over front-office. The MIT study confirms it: AI's ROI lies in automating internal processes (accounting, compliance, operations), not in marketing chatbots. The back-office is where data is structured, processes are repetitive, and impact measurement is immediate.
4. Negotiate your AI contracts like risk hedges. Avoid long-term commitments to proprietary models with volatile pricing. Diversify your model providers. Include exit and reversibility clauses. When the cost factor between GPT-5.2 Pro and Gemini 3 Pro is 10x for an equivalent task, contractual flexibility isn't a luxury.
5. Invest in skills, not licenses. Train your teams to understand, evaluate, and steer AI. The technology will change, models will be replaced, prices will fluctuate. Your organization's ability to navigate this complexity is the only asset that doesn't depreciate.
After the Bubble, the Value
Let's remember a truth that the current frenzy makes us forget: after every bubble, the underlying technology doesn't disappear. The internet didn't vanish after 2000. Real estate didn't vanish after 2008. And AI won't vanish after the coming correction.
What disappears are the players without fundamentals. The 30x revenue valuations. The projects without ROI. The investments without strategy.
What survives, and thrives, are the organizations that used the technology to solve real problems, with real economic discipline.
The question isn't whether you should invest in AI. The answer is yes. The question is whether you're investing like a tulip speculator in 1637 or like a foundation builder who will weather the storm.
History has already decided.