원문 보기
And here we begin again. Look at what Wall Street, CNBC, Silicon Valley, venture capitalists, and hyperscalers are doing right now. Every day you turn on the financial news, and what do you hear? OpenAI is worth $150 billion , $200 billion . Oh, no, no, no, no, no. This is the next trillion-dollar company.
Anthropic—the savior of the corporate sector. SpaceX at 100x forward earnings is cheap. Do you know what they do? They are preparing the public for the biggest liquidity event of all—a mega-IPO in the field of advanced AI . And here's what no one on Wall Street wants to tell you. This is a planned exit.
Venture capitalists , who have invested private funds for years, cannot enter the private markets. Hyperscalers—Microsoft, Amazon, Alphabet, Meta, Oracle— must keep playing, because their entire revenue story depends on it . And now Michael Bury, the man who predicted the subprime market crash before anyone else , has just published an exposé on Substack that reveals “the skeletons in the closet.” In the fourth installment of the Heretic's Guide to the AI Stars series, Bury pulled out an SEC footnote that Wall Street is deliberately ignoring.
And what did he find? Over $3 trillion in hidden off-balance sheet liabilities. Today I break down Bury's numbers, why OpenAI and Anthropic are desperately trying to IPO before open-source models have eroded their pricing advantage, and why retail investors buying index funds will be left with nothing.
Now let's look at the hard numbers. Bury did n't just look at press releases. He reviewed line by line the 10-K and 10-Q reports certified under the responsibility of the CEOs and CFOs of the five largest hyperscalers. And this is what he discovered. First, uncommitted, non-cancellable leases— $1.2 trillion.
You see, according to the ASC 842 accounting standard , if a company signs a 20-year lease for a huge data center, but the building is not yet officially operational, nothing is shown on the balance sheet as debt. This remains completely unnoticed in the notes. So, Microsoft's uncommitted lease liabilities exploded from $92.7 billion to $329.1 billion.
Oracle reports $288 billion out of nowhere. Meta—279 billion as of June plus another 68 billion in July. Alphabet and Amazon bring the total to nearly $1.2 trillion, of which at least $857 billion is strictly non-cancellable. Think about that for a moment. You see, if you rent an office and you don't need it, you can sublease it to an accounting firm or a dentist.
But if you sign a 20- year, liquid- cooled lease that can't be terminated, who are you going to sublease it to? Meta has already had to write off $1.34 billion in 2022 due to the cancellation of data center construction. Second: $1.5 trillion in supply chain procurement commitments. So, in addition to rent, they promised suppliers over $1.5 trillion for chips, cooling systems, and power equipment.
Alphabet alone accounts for 811 billion of this amount. Now add $1.2 trillion in lease payments, $1.5 trillion in procurement obligations, and private SPVs, and Burry has accrued over $3 trillion in off-balance sheet liabilities. And how much did the "big five" earn together last year? Less than 400 billion dollars.
So, they increased their contractual obligations by almost eight times compared to their annual net income for the sake of one technological bet . Burry revealed another dirty secret in the balance sheet—construction in progress (CIP). You see, these five companies now have over $400 billion in construction in progress on their balance sheets .
At Alphabet, it's 122.8 billion or 29% of their gross assets. Meta has $80 billion. You may ask why this is important? Because while the asset is designated as " construction in progress," it is not depreciated under GAAP accounting rules. This is shown as a zero expense in the income statement . Now let's connect the dots.
First, it takes 3 to 5 years to acquire capacity, build substations, and complete a high-density data center. Second, Jensen Huang and Nvidia operate on a yearly production cycle. Blackwell, BG200, BG300, and then Vera Rubin. Third, each generation changes power consumption per rack, cooling density, and engineering networks.
Satya Nadella admitted this on camera. He said he doesn't want to build gigawatts tied to one generation, because by the time the building is ready, the next chip family will require a completely different design. So you have tens of billions of dollars worth of Nvidia chips sitting in warehouses or in semi-finished cases, economically depreciating every month, while GAAP accounting pretends they're not losing a cent of value.
And when these data centers finally launch or close, depreciation and a wave of impairments will crush reported profits. Why do hyperscalers take such crazy risks ? Well, to fuel the narrative that cutting-edge AI labs will generate endless cash flows. Let's look at the mathematics of these labs. OpenAI's internal predictions have been leaked.
They plan to spend $280 billion on computing and operating expenses to reach profitability by 2031 or 2032. 280 billion dollars. This is not a business model. This is a furnace for burning capital. Next, let's look at Anthropic. They appear on financial podcasts and say, "Oh, our operating activities have a positive gross margin." Yeah, if you exclude the costs of computing power and data centers.
It's like an airline saying, "We're running an extremely profitable business if you ignore the cost of jet fuel and planes." So, what is the ultimate goal here? Why are they rushing into an IPO? Because open source AI is breathing down their necks. Deep Seek, Meta's Llama models , open-weight architectures— they are closing the performance gap for a fraction of the cost of inference.
The "moat" around Frontier Labs is evaporating very quickly in real time. So if they wait another 3 years, the market will realize that LLM is mostly a commodity API with rapidly declining margins. They should go IPO right now. They have no choice. Venture capitalists are peaking, hyperscalers are locking in long-term corporate commitments, and insiders are waiting out 6-month stock lock-up periods.
And how does the average person end up paying for it? Well, reports are already circulating that these megacorporations, along with companies like SpaceX, which trade at a forward P/E of 100x, are preparing for direct inclusion in major indices like the Nasdaq 100 and S&P 500. And when a multi-billion dollar loss-making hyper-mechanism hits the major indices, every pension fund and every passive investor automatically buys shares of these companies, whether they like it or not.
And when the bubble bursts, the insiders are already on their yachts, and retail investors will absorb the write-offs from the balance sheet. That's why I don't play Wall Street momentum games . When you buy a trillion-dollar hyperscaler or a bloated lab before an IPO that burns billions, the risk is completely asymmetric against you.
You are taking on a huge balance and duration risk for a potential 20-30% profit if everything goes perfectly. On my YouTube channel MicroCap Explosions, we look for the exact opposite—an asymmetric risk-reward relationship, where the downside risk is somewhat hedged, but the upside potential isn't 20 or 30%.
It's 10x or 100x. So don't let CNBC or the venture cartels sell you their liquidity on the way out. Read Michael Bury's latest article, " Cassandra Unchained." Look at the notes in the 10K reports yourself. The numbers don't lie. So, having said that, thanks for watching and see you next time.
댓글 0