The AI Ponzi Scheme’s Final Act: When the House of Cards Runs Out of Cards

Welcome to the AI endgame, where the tech industry’s most expensive game of musical chairs is about to run out of music—and seats. While Silicon Valley’s philosopher-kings have been busy promising us AGI, immortality, and a new industrial revolution, a small problem has emerged: the money is running out, the hardware is rotting faster than a dot-com business plan, and the entire ecosystem is being held together by the financial engineering equivalent of duct tape and prayer.​

This isn’t your grandfather’s bubble. It’s far more spectacular.

The Evidence: Follow the Money (Before It Disappears)

Let’s start with the numbers that should make any rational investor reach for the Pepto-Bismol. OpenAI, the crown jewel of the AI revolution and humanity’s supposed savior from tedium, plans to build 26 gigawatts of data centers in the coming years. Each gigawatt costs approximately $60 billion. That’s $1.56 trillion—more than the combined five-year free cash flow of Amazon, Google, Meta, Microsoft, and Apple during a pandemic boom when tech usage soared.​

Let that sink in. OpenAI needs to raise more capital than Big Tech’s entire profit engine produced in five years. And for what? A company projected to generate $15-20 billion in revenue this year while losing $9 billion, with losses ballooning to $47 billion by 2028. Meanwhile, xAI is burning through $1 billion per month. This isn’t a business model; it’s a black hole with a ChatGPT interface.​

The Bank of England issued a warning this week that tech stock prices, inflated by AI optimism, face “heightened risk of a sharp market correction”. The IMF’s Kristalina Georgieva echoed the concern, noting that while stocks have soared on “optimism about productivity-enhancing potential,” financial conditions could “shift suddenly”. Even Jamie Dimon is worried. When the suits who survived 2008 start sweating, it’s time to pay attention.​

Here’s where it gets deliciously absurd: We’ve officially run out of “organic capital”—a fancy term for actual money from actual investors who expect actual returns. The classical VC funding ladder (big VCs at $1-10 billion, SoftBank at $10-30 billion, then IPO) has collapsed because AI labs refuse to go public. Why? Because an IPO means analysts would dissect their business model and discover there isn’t one. Even if they wanted to IPO, it wouldn’t raise nearly enough capital, since AI labs now require over $100 billion in new investments annually.​

The GPU Death Spiral: Your Billion-Dollar Asset Is Already Obsolete

Now for the technical bombshell that should terrify every CFO in Silicon Valley: GPUs are depreciating faster than a new car driven off the lot—if that car also caught fire and exploded.​

Nvidia has moved to a one-year upgrade cycle. Jensen Huang himself admitted that between the Hopper and Blackwell generations, they’re driving token costs down 10-20x—while Moore’s Law would have achieved only 20%. This is technological progress on steroids, which sounds great until you realize it means every GPU you buy is obsolete before you finish installing it.​

Jonathan Ross, CEO of Groq and one of the founders of Google’s TPUs, uses a one-year depreciation cycle because, in his words, people using 3-5 year cycles “are wrong”. Yet hyper-scalers like Microsoft and Google use 3-4 year depreciation schedules, while neo-clouds like CoreWeave stretch it to six years. Meta reported annualized failure rates of H100 GPUs at around 9%—meaning over one in four are dead after three years.​

By the time a GPU reaches the end of a traditional three-year depreciation schedule, it’s three generations old. At six years? Six generations behind. Microsoft’s Satya Nadella confirmed this nightmare, noting they see “more than 2x price performance gain for every hardware generation and more than 10x for every model generation”. Economic obsolescence is arriving long before physical failure.​

The accounting fraud—sorry, “creative depreciation”—is breathtaking. CoreWeave simply extended its GPU depreciation from four years to six in January 2023. Problem solved! Except the losses would be “much, much bigger” if they used the correct 1-2 year cycle. Oracle is already losing $100 million per quarter renting data centers primarily to OpenAI, though they’re calling it a “timing issue”. Nothing says “sustainable business model” like immediately losing money on your biggest deal.​

The Circular Financing Charade: Nvidia Becomes the Bank of Last Resort

Here’s where the farce reaches its apex. With organic capital exhausted and free cash flow depleted from hyper-scalers buying Nvidia chips, who’s left to finance the next trillion-dollar data center? Nvidia itself.​

Nvidia structured a potential $100 billion investment in OpenAI at $10 billion for each gigawatt of power OpenAI brings online. Let that logic marinate: Nvidia is financing its own customer’s ability to buy Nvidia products. This is the corporate equivalent of a drug dealer giving his best customer a loan to buy more drugs, then calling it “strategic investment.”​

Why would Nvidia do this? Because OpenAI and Anthropic are currently the end buyers of one-third of all Nvidia GPUs. If OpenAI collapses, Nvidia’s entire demand story evaporates. The deal isn’t confidence; it’s desperation dressed up as vision. Nvidia also invested in xAI’s recent round. When your supplier becomes your creditor, you’re not in a boom—you’re in a Ponzi scheme’s final act.​

Morgan Stanley’s Lisa Shalett observed that “the guy at the epicenter is basically starting to do what all ultimate bad actors do in the final inning”. She’s referring to Jensen Huang, who now occupies the delicious position of being both the primary beneficiary of AI hype and its primary financier. OpenAI signed a $300 billion deal with Oracle over five years—$60 billion annually. When Oracle announced the deal, its shares soared 40%, adding nearly one-third of a trillion dollars to its market value in a single day. OpenAI’s valuation jumped from $300 billion to $500 billion in less than a year.​

This is circular financing masquerading as validation. Tech companies are now 40% of the S&P 500. AI capital expenditures surpassed U.S. consumer spending as the primary driver of economic growth in the first half of 2025. The entire U.S. economy is being propped up by the promise of AI productivity gains that remain stubbornly theoretical. One analyst at Yale noted that the “dependence among these major AI players could trigger a devastating chain reaction” akin to the 2008 financial crisis.​

The Judgment: We’ve Seen This Movie Before

History doesn’t repeat, but it rhymes—and right now it’s rhyming in Dutch tulips, dot-com flameouts, and 2008 subprime mortgages. We have unsustainable valuations predicated on infinite future growth. We have fast-depreciating physical assets being accounted for as long-term investments. We have circular financing schemes where suppliers fund customers to buy their own products. We have concentration risk where a handful of companies control the entire ecosystem. And we have a complete disconnect between revenue reality and capital requirements.​

The bull case for Nvidia rests on the assumption that customer money is infinite. It’s not. The bull case for AI labs rests on achieving AGI before bankruptcy. The odds aren’t great. Adam Slater, lead economist at Oxford Economics, noted that indicators of a bubble include “a prevailing sense of extreme optimism regarding the underlying technology, despite significant uncertainties about its ultimate outcomes”. Check, check, and check!​

When the music stops—and it will—the contagion will be swift and brutal. As one analyst at The Conversation warned, “bubbles are extremely disruptive and affect people in very real ways. Stocks fall, pensions suffer, unemployment rises”. But at least we’ll have learned an expensive lesson about confusing technological promise with business fundamentals. Again.​

The Aftermath

So, dear reader, as you watch this slow-motion trainwreck unfold: Have you trimmed your AI positions yet, or are you riding this rocket all the way into the ground? What’s your GPU depreciation schedule—one year of honesty or six years of hopium? And when Nvidia becomes the Federal Reserve of AI, what could possibly go wrong?

---

This article is just a tremor. The earthquake is coming.

The patterns of hype, deception, and greed laid bare here are part of a much larger story. They are the evidence file for the great deception of our time.

The full, unvarnished truth is detailed in the forthcoming book from our founder, Simba Mudonzvo:

The Gilded Cage: How the Quest for Artificial Intelligence Became the Greatest Deception in Human History.

Stay tuned. The reformation is coming.