The early 2000s are remembered for the dot-com crash, but behind the headlines about Pets.com and internet startups, a more structural failure was unfolding in telecom. Companies like Global Crossing, Qwest, and even Enron created the illusion of growth by engaging in “fiber swaps.” They exchanged capacity with one another—essentially swapping dark fiber or rights to use bandwidth—and then booked those swaps as revenue. On paper, it looked like demand was exploding. In reality, it was a house of cards: the industry was drowning in excess capacity, and the revenue was circular, not real. When the bubble burst, billions in paper profits vanished overnight.
Fast forward to 2025, and the AI industry might be reliving the same pattern under a different disguise.
The Oracle–OpenAI Playbook
Announcement effect
The cycle begins with a press release. Oracle announces a multi-hundred-billion-dollar cloud deal with OpenAI. The market reacts instantly: Oracle’s stock surges, adding over $100B in paper wealth to Larry Ellison. Yet at this point, no product has shipped and no actual revenue has materialized—it’s purely sentiment-driven.
Revenue mismatch
Here’s where the math breaks down. OpenAI’s estimated annual revenue for 2025 hovers around $10B. But the rumored commitment to Oracle is $60B per year—six times its top line. Unless growth accelerates at an unprecedented rate, the numbers simply don’t reconcile.
Circular financing loop
This is where it starts to rhyme with fiber swaps. Oracle invests directly in OpenAI. OpenAI then uses those funds to pay Oracle for GPU compute and cloud services. Oracle books that spend as revenue, reports “record growth,” and the stock jumps again. Net result: Oracle is effectively funding its own revenues by recycling capital through OpenAI.
Economists call this a reflexive loop, a concept made famous by George Soros: perceptions of growth drive capital allocation, which in turn creates temporary growth that validates the original perception. It works beautifully—until reality intrudes.
Speculative infrastructure
Much of the $300B deal is forward-dated, tied to GPUs and data centers that don’t yet exist. Yet Oracle gets credit today. Investors reward the company now for revenues it might collect years in the future.
Ponzi-like dynamics
At least from a skeptical lens, the structure is precarious:
- Customers (like OpenAI) are funded by investors.
- That money is recycled into infrastructure providers (Oracle, Microsoft, CoreWeave).
- Infra providers book it as revenue, stock prices soar.
- Stock gains create new paper wealth that can be reinvested into customers.
As long as the wheel keeps turning, everyone benefits. But the cycle relies not on organic demand, but on financial engineering.
CoreWeave and Nvidia: Same Tune, Different Verse
This dynamic isn’t confined to Oracle. On Monday, CoreWeave shares jumped nearly 8% after announcing a $6.3B order from Nvidia. But hidden in the deal is an unusual twist: Nvidia is obligated to buy back CoreWeave’s unused GPU capacity through 2032.
Here again, we see the circularity:
- Nvidia sells GPUs to CoreWeave.
- CoreWeave rents them to AI customers like OpenAI.
- If demand falls short, Nvidia repurchases the excess.
- Meanwhile, Wall Street applauds the headline numbers and drives up CoreWeave’s market cap (now over $58B).
On paper, CoreWeave’s revenue is exploding—$1.21B last quarter, up 207% year over year—but losses remain steep. Nvidia, its largest supplier, is also its safety net. The feedback loop ensures that “capacity” looks monetized whether or not real market demand exists.
Echoes Across Tech History
The parallels are hard to ignore. Fiber swaps in telecom. Ad-swapping in the dot-com era. Wash trading in crypto markets. In each case, companies built circular revenue streams that inflated growth without creating proportional value.
Academic research into financial bubbles and reflexivity highlights the danger: once perception-driven loops are in motion, they can persist far longer than skeptics expect, because everyone has an incentive to keep playing. The alternative—being the first to stop—invites collapse.
A Bubble Waiting to Pop?
The AI revolution is real at the technological level. Models are improving rapidly, and infrastructure demand is genuine in many areas. But the business models layered on top are showing signs of manufactured demand—capacity booked, revenues recycled, and valuations inflated in ways that echo past bubbles.
When Wall Street’s narratives drift too far from operational reality, corrections follow. The telecom crash of the early 2000s wiped out hundreds of billions in market cap almost overnight. The AI boom, fueled by circular financing loops and speculative infrastructure bets, risks meeting a similar fate if genuine demand fails to catch up with the hype.
For now, the cycle continues—because it doesn’t have to make sense. It just has to last longer than the market’s patience for asking questions.