The Subprime Compute Crisis

4 min read Original article ↗

If you listen to the self-proclaimed visionaries of Silicon Valley, we are standing on the precipice of an economic miracle driven by artificial intelligence. But if you strip away the utopian narratives and do the actual forensic research on the capital flows, the math tells a much darker story.

We aren’t in a new paradigm. We are watching a masterclass in synthetic demand.

Today’s AI ecosystem is running the exact same structural playbook that Wall Street used to blow up the global economy in 2008. We have simply traded subprime mortgages for subprime compute.

In the mid-2000s, the financial system manufactured demand for housing by issuing NINJA loans—No Income, No Job, No Assets. The goal wasn’t to collect sustainable interest; the goal was to keep the origination fees flowing so Wall Street could bundle the toxic debt and sell it.

Today, the tech industry’s equivalent of the NINJA loan is the AI foundation model startup.

These are entities with astronomical compute requirements, deeply negative gross margins, and virtually no sustainable, organic enterprise cash flow. They aren’t just subprime borrowers; they are generational wealth incinerators. To justify their current infrastructure commitments, these labs would essentially need to capture nearly 100% of all global corporate IT spending just to break even on the depreciation of their GPUs. Yet, they are receiving tens of billions in funding. Anthropic recently closed a round valuing it at $380 billion, while OpenAI is nearing an $850 billion valuation. Their inference costs and infrastructure commitments are so staggering that they are locking into future cloud commitments that vastly exceed their cash generation.

They are kept alive solely by a constant IV drip of venture capital because the underlying unit economics are fundamentally broken.

In 2008, Wall Street hid toxic leverage in complex tranches. Today, Big Tech is hiding the lack of end-user economics through a massive accounting game known as roundtripping.

Here is how the “Infinite Money Loop” works:

  1. The Investment: A hyperscaler or chipmaker invests billions into an AI startup.

  2. The Compute Commitment: As an explicit or implicit condition of that funding, the startup is contractually forced to spend the vast majority of that capital renting GPUs and infrastructure back from the investor’s own cloud division.

  3. The Revenue Recognition: The hyperscaler recognizes that compute usage as top-line cloud growth.

This is circular risk masquerading as innovation. It is closed-loop financing. The demand isn’t real—it’s reflexive. The revenue of one partner is simply the CapEx of another. By “buying” their own revenue, these tech giants are manufacturing the illusion of 30%+ cloud growth to sustain their multi-trillion-dollar market caps. It is a legalized Ponzi scheme of compute.

Venture capital funds aren’t just happily playing along; they are actively architecting the delusion.

When a mega-fund leads a round pushing a startup’s valuation to hundreds of billions of dollars, they aren’t underwriting a viable business. They are validating a fake price tag so they can mark up their prior early-stage investments, show massive paper returns to their LPs, and aggressively extract 2-and-20 management fees on phantom wealth. They are getting rich off the mark-up, not the margin.

The funds mark up the startups, the startups buy cloud compute, the hyperscalers report record revenue, and the hyperscalers’ stock prices go up. Everyone gets rich on paper.

In 2008, the music stopped when the underlying assets—the subprime homeowners—finally defaulted, taking the entire collateralized debt structure down with them.

The AI bubble will burst the exact same way. Eventually, the VC dry powder will run out. The moment a major lab misses a funding milestone and can’t pay its multi-billion-dollar infrastructure bill, the reflexive demand loop shatters.

Hyperscalers will suddenly report massive misses on their cloud revenue guidance, the private valuations of these AI darlings will crater, and the entire ecosystem will face a margin call. Stop praising the visionaries. They are just arsonists in Patagonia vests, burning billions of dollars of LP capital in a closed-loop furnace to prop up their own equity.

Discussion about this post

Ready for more?