AI companies spent the last two years borrowing money to build data centers the size of small nations. Worked great when Treasury yields sat at nothing. Now the 10-year is spiking and suddenly the infrastructure buildout costs actual money.
These companies don't generate profit. They generate press releases about how they'll generate profit once they figure out how to monetize chatbots that hallucinate legal advice. The business model is "borrow now, revenue later, way later, like maybe never but definitely later." Wall Street called this innovation until the bond market started asking for its money back with interest attached.
The debt wasn't a problem when rates were low because nothing is a problem when rates are low. You could finance a casino on the moon when borrowing was free. But Treasury yields don't care about your AI roadmap. They care about math. And the math says lending money to companies that burn through billions building server farms to teach computers how to write poetry costs more now.
Retail traders bought AI stocks because they saw charts going up. They didn't read the balance sheets. They don't know what a balance sheet is. They think EBITDA is a Swedish furniture brand. Now those same traders will watch their holdings get crushed by the cost of debt service while CNBC explains that rising yields are actually bullish if you squint hard enough and ignore basic arithmetic.
The AI buildout continues anyway because stopping would mean admitting the whole thing might have been expensive performance art. So they'll keep borrowing at higher rates and building more data centers and praying that somehow this ends in profit instead of bankruptcy.
Turns out the future costs more when the present wants its cut.
Photo by Tyler Prahm on Unsplash
Leave a Comment