IS THIS HOW AN AI PONZI SCHEME GETS BORN?

(NVDA), (AMD), (ORCL), (MSFT), (GOOG), (AMZN), (META)

I bumped into an old trading buddy at the Whole Foods in Mill Valley last month. He runs infrastructure at one of the hyperscalers now, and while we were both reaching for the same overpriced organic coffee, he mentioned something that stopped me cold. 

“We’re not buying AI chips anymore. We’re financing AI chips for companies that promise to buy our cloud services later.” That’s when I knew we’d crossed into dangerous territory.

What he was describing is the financial equivalent of a drug dealer giving away free samples, except the samples cost $100 billion and the addiction might bankrupt entire companies. I’ve been tracking these arrangements through SEC filings that most people never bother reading, and what emerges is a web of circular transactions that would make Enron accountants proud.

Take OpenAI’s celebrated funding round with Nvidia (NVDA). Everyone sees the $100 billion investment and thinks “strategic partnership,” but dig into the details and you’ll find that $85 billion of it boomerangs right back to Nvidia as guaranteed hardware purchases over 3 years. 

It’s vendor financing disguised as equity investment, complete with payment deferrals tied to OpenAI’s ability to actually monetize their models. 

Nvidia books immediate revenue while OpenAI gets the computational power they desperately need, but if OpenAI stumbles, Nvidia suddenly owns a very expensive collection of rapidly depreciating silicon.

The AMD (AMD) deal reveals even more sophisticated financial engineering. OpenAI committed to buying 6 gigawatts worth of chips in exchange for warrants on roughly 10% of the company. 

Think about that structure for a moment. AMD is essentially giving away equity to guarantee purchase orders, with warrant exercise prices tied to OpenAI’s own revenue projections. 

If the AI monetization story works out, those warrants become extremely valuable. If it doesn’t, AMD not only loses their biggest customer but dilutes existing shareholders through below-market warrant exercises.

Oracle’s (ORCL) $300 billion commitment over 5 years represents the logical extreme of this financing evolution. They’re funding this through asset-backed securities tied to projected data center cash flows, essentially asking pension funds and insurance companies to bet on OpenAI’s ability to turn artificial intelligence into actual profits. 

The problem is that unlike the fiber optic cables from the telecom boom that still power today’s internet, GPU clusters become obsolete faster than smartphones. When Nvidia’s current generation chips get replaced in 3 years, they don’t magically transform into useful infrastructure for whoever inherits them.

What fascinates me is watching how the debt markets are finally pricing in this reality. 

Data center REIT yields have started diverging from traditional real estate investments, and smart institutional money is specifically targeting electrical grid upgrades and power generation projects adjacent to AI facilities. Power infrastructure retains value regardless of which chips are humming inside the buildings.

The companies best positioned for this environment are the ones that can fund AI investments from operating cash flow rather than exotic financing structures. 

Microsoft (MSFT), Google (GOOG), Amazon (AMZN), and Meta (META) have diversified revenue streams and fortress balance sheets that let them play with house money. They’re not dependent on circular financing arrangements or vendor partnerships to fund their AI ambitions.

The real danger lies with companies trapped in the middle of the AI stack who lack both user relationships and manufacturing advantages. 

These firms are increasingly reliant on venture debt and revenue-based financing that becomes problematic when growth rates decelerate. When you see partnership announcements where strategic investors also happen to be major suppliers, that’s usually code for limited conventional financing options.

Until recently, AI infrastructure was funded almost entirely by hyperscaler cash flows. Now we’re seeing leverage enter the equation, and that changes everything. What had been a disciplined, cash-funded race is becoming a debt-fueled arms race where someone eventually gets stuck holding worthless paper.

The parallels to previous bubbles are instructive but incomplete. 

During the telecom boom, companies like WorldCom and Global Crossing built physical networks that retained intrinsic value even after the companies collapsed. An AI bubble built on GPUs with 3-year useful lives offers far less durable foundation. However, an AI bubble that spurs massive buildouts in power generation would benefit the economy for decades.

These financing arrangements are either brilliantly prescient or dangerously premature, and we won’t know which until market conditions shift. 

The smart play is staying positioned with companies that have real moats and diversified cash flows while avoiding potential balance sheet accidents among newer entrants who are entirely dependent on AI monetization arriving before the funding dries up.

For now, the momentum can persist as long as the circular financing arrangements hold together. 

But when they start unraveling, it’s going to happen fast, and companies without diversified businesses are going to find themselves in a very uncomfortable position.

So the next time I’m at Whole Foods, I’ll probably buy extra coffee. If my buddy’s right about this financing trend, we’re going to need plenty of caffeine to stay alert when this whole thing starts unraveling. The early warning signs are always worth paying attention to.