The AI Tulip Mania: How Silicon Valley Reinvented 17th-Century Dutch Financial Stupidity, GPU by GPU

In 1637, a Dutch trader named Jan van Goyen traded his entire house for three tulip bulbs. In 2024, OpenAI traded $10 billion in future revenue, Microsoft’s cloud infrastructure, and whatever dignity it had left for the right to hoard Nvidia H100 GPUs ahead of its competitors. If you’re thinking these two scenarios are separated by four centuries of human progress and enlightenment, you haven’t been paying attention to the financial engineering gymnastics currently performed by Sam Altman and his merry band of Large Language Model enthusiasts.

Welcome to the AI bubble, where history doesn’t repeat itself—it just changes the merchandise from flowers to floating-point operations.

The Evidence: Follow the Silicon, Not the Money

Let’s establish the facts, because unlike most AI hype pieces, we actually care about them.

Between 1634 and 1637, tulip bulb prices in Holland increased by 5,900%. The Dutch government introduced trading restrictions. Sound familiar? Between 2022 and 2025, the price of Nvidia H100 GPUs increased by approximately 400% on secondary markets, and the U.S. government introduced export controls to China. The tulip bulbs, dormant for most of the year, shifted from physical asset trading to paper contracts promising future delivery. Today’s GPU market operates identically—OpenAI, Microsoft, and Meta are signing multi-billion-dollar contracts for chips that won’t be manufactured for another 18 months!

Here’s where it gets delicious: DeepSeek, a Chinese AI lab, recently released models that match GPT-4 performance using a fraction of the compute. They did it with older chips, circumventing U.S. export controls through pure engineering competence. This is the equivalent of a Dutch farmer in 1636 discovering you can grow tulips in regular soil instead of trading your estate for imported bulbs. The response from Silicon Valley? Panic-buying more GPUs, naturally.

The circular financing scheme is the crown jewel of absurdity. OpenAI needs GPUs from Nvidia, and now also from AMD. Nvidia needs cloud infrastructure from Oracle. Oracle needs AI capabilities from OpenAI. So they’ve created a beautiful closed loop of cash, equity, and IOUs that would make any 17th-century Amsterdam merchant blush with envy. OpenAI is simultaneously Nvidia’s customer and investor. Oracle is both OpenAI’s infrastructure provider and strategic partner in securing GPU supply. This isn’t a supply chain; it’s a financial ouroboros eating its own tail while insisting it’s “scaling responsibly.”

Anne Goldgar’s historical research revealed that tulip mania ruined fewer than six people—all wealthy merchants who could afford the loss. The broader Dutch economy was fine. Today’s AI bubble? Same story. When this pops, the casualties will be venture capitalists who threw $300 million at a company with no revenue model, tech stocks overvalued by 400%, and maybe a few C-suite executives who have to settle for only one yacht which will docked at Monaco bays. The rest of us will continue using ChatGPT’s free tier and wondering what all the fuss was about.

The Absurdity: Trading Castles for Computational Futures

The parallels aren’t coincidental—they’re structural. Both bubbles were driven by:

Artificial Scarcity: Tulips could only be cultivated during specific seasons. U.S. export controls artificially restrict GPU supply to create geopolitical leverage. Both turned commodities into currency.

Geographic Competition: Tulips originated in Asia and drove European traders mad trying to secure supply. AI competition with China’s DeepSeek is driving American companies to spend more on GPUs than some countries spend on defense. The anxiety isn’t about capability—it’s about being outcompeted by Asians who figured out how to do more with less.

Futures Trading as Financial Religion: When you can’t get the actual product, you trade promises of future product. The Dutch invented tulip futures. Silicon Valley invented GPU futures, wrapped them in “strategic partnerships,” and called it innovation.

Here is a fictional exchange that’s depressingly plausible:

Visionary CEO (definitely not based on anyone specific): “We’re securing 50,000 H100s for Q3 2026. Yes, I know we haven’t figured out what to do with the 30,000 we already have, but you don’t understand—DeepSeek just released a new model. We need to show the market we’re still in the race and crushing it.”

Desperate CTO: “Sir, DeepSeek’s model runs on chips we can buy right now for a tenth of the cost.”

Visionary CEO: “That’s exactly why we need more expensive ones. It signals confidence.”

Smug VC: “I’ll double our position. This is just like when Sequoia passed on Google. We cannot miss the GPU accumulation phase of the AI revolution.”

This is the groupthink that defines bubbles. Not the technology itself, but the collective delusion that spending more money on the same thing everyone else is buying proves you’re smarter than everyone else.

The Judgment: The Gilded Cage We Built for Ourselves

Here’s the damning verdict: The AI bubble isn’t a failure of technology—it’s a failure of imagination disguised as ambition.

OpenAI raised billions to build AGI. DeepSeek built competitive models for millions. The difference isn’t the chips; it’s the incentive structure. OpenAI needs to justify its valuation by spending conspicuously. DeepSeek needs to justify its existence by building efficiently. One is optimizing for headlines and the next funding round. The other is optimizing for actual results.

The GPU hoarding isn’t a strategy—it’s a financial moat built on US government-enforced scarcity and venture capital’s inability to admit it backed the wrong horse. Every dollar spent on futures contracts for chips that don’t exist yet is a dollar not spent on the unsexy work of making AI actually useful. But “useful” doesn’t raise Series D rounds. “We secured exclusive access to next-generation compute” does.

The export controls were supposed to slow China’s AI development. Instead, they forced Chinese engineers to become better at optimization, while American companies became better at financial engineering. We’ve created a system where efficiency is punished and excess is rewarded. That’s not innovation—that’s decadence.

And when this bubble pops—and it will, because every bubble does—the damage will be contained to the same small network of wealthy players who always survive. The VCs will write off the losses. The founders will pivot to their next venture. The tech giants will absorb the failed startups at fire-sale prices. The only thing that will be “artificial” about this intelligence is the scarcity we created to justify the spending.

The book “The Gilded Cage: How the Quest for Artificial Intelligence Became the Greatest Deception in Human History” is coming soon. It won’t be fiction—it’ll be a receipt.

---

What’s your GPU horror story? Have you watched your company panic-buy compute it doesn’t need because a competitor announced a partnership? Are you an engineer forced to justify billion-dollar chip purchases with PowerPoints about “strategic positioning”? Share your stories below—misery loves company, and so does evidence.