The AI infrastructure boom now runs on bonds, leases, and private capital. Which means the thing powering your chatbot is financed like a strip mall in Tucson.
Leveraged investors pile into this market because nothing says "transformative technology" like a financing structure borrowed from commercial real estate. Server farms that train large language models get built the same way someone funds a Cheesecake Factory. Borrow big. Lease the equipment. Hope the revenue shows up before the interest payments do.
This creates layers of market risk that nobody can track. The headline admits as much. Harder to track means the exposure is distributed across enough counterparties that when it blows up, everyone gets to pretend they didn't see it coming.
Retail traders hear "AI infrastructure" and think they're investing in the future. They're actually buying exposure to lease obligations on hardware that depreciates faster than a Ford F-150. The servers obsolete themselves every eighteen months. The debt does not.
Private capital floods in because public markets ask too many questions. Questions like "What's the cash flow?" and "Who's buying this capacity?" Private investors skip that part. They just wire the money and hope Nvidia keeps printing.
Bonds tied to data centers sound sophisticated until you remember bonds are just IOUs with extra steps. Someone borrows money to build a building full of GPUs. That someone promises to pay you back. The promise is only good if AI companies keep renting compute at current rates. Current rates assume infinite demand growth. Infinite demand growth assumes everyone on Earth needs to generate images of cats wearing hats forever.
The financing stack now looks like 2007 but with worse collateral. At least houses stood still. These data centers burn capital on cooling costs alone.
Leveraged bets on AI infrastructure: because apparently we needed subprime mortgages but for robots.
Photo by Google DeepMind on Unsplash
Leave a Comment