Two enormous borrowing booms are colliding: government debt and AI. I think the combination is setting us up for a crash.
Imagine spending more than you earn, year after year. Your debt grows, your creditworthiness deteriorates, lenders demand higher interest—and now you’re borrowing more just to pay them. Governments have more options than households, but the arithmetic still bites. The US is running a deficit of nearly $2 trillion, with a huge and growing interest bill.
Lenders are demanding more compensation. US 10-year borrowing costs recently reached 5.34%, a 24-year high. France’s borrowing premium over Germany reached a 14-year high. As old debt gets refinanced, higher interest bills feed bigger deficits—and more borrowing.
Now add AI: fantastic technology with a ferocious appetite for chips, memory, electricity and money. Bottlenecks inflate the bill. Micron has reported memory prices rising sharply. Yet the technology giants keep spending because nobody wants to lose the race.
Big Tech capital spending is heading towards $700 billion this year. Cash generation is struggling to keep pace, and borrowing is filling the gap. AI companies are competing with governments for investors’ money, adding pressure to borrowing costs.
But who pays the bill?
Much of the visible profit is in semiconductors and infrastructure—the suppliers benefiting from bottlenecks. Their customers still need to earn a return. A profitable chipmaker doesn’t prove a profitable AI economy.
And now the financing itself is becoming circular. Nvidia, the biggest beneficiary so far, is helping backstop funding for AI infrastructure built around its chips. That financing helps customers buy more of the very systems Nvidia sells. It keeps the spending loop going, but it doesn’t prove that end users can pay for it. It can make demand look stronger while shifting more risk into guarantees and complex financing deals.
Ed Zitron estimates hyperscalers’ annual AI revenue at about $183 billion, but only around $65 billion excluding OpenAI and Anthropic—themselves heavily dependent on funding. That isn’t a direct accounting comparison with capital spending, but it exposes how much future growth the investment assumes.
Anthropic’s IPO disclosures sharpen the picture: rapid revenue growth, but also billions in losses and roughly $500 billion of infrastructure commitments over a decade, most of them binding regardless of usage. Enormous obligations are being locked in against enormous expectations.
My concern is that expensive debt breaks the financing before AI delivers the returns. Projects stop. Assets get written down. Stocks fall. Lenders tighten credit. Suppliers and customers cut jobs. Expensive mortgages meet unemployment, threatening housing. Tax receipts fall while welfare and rescue costs rise—leaving already stretched governments borrowing still more.
Leverage accelerates the damage. South Korea has already shown how leveraged ETFs and borrowed-money bets concentrated in semiconductor stocks can amplify a sell-off.
The escape is AI delivering productivity, profits and GDP growth on a massive scale, quickly. I believe in the technology. It helped me discover a lot of this, connect the dots, plan my finances and even write this post. I'm convinced it will change the world and make our lives and work materially better, by orders of magnitude. But I’m unconvinced the economics will arrive before the financing strains become intolerable.
Eventually, failing credit markets would force emergency liquidity and support; rate cuts when inflation permits. That could stabilise the crisis while leaving more debt and renewed pressure on money’s purchasing power.
I’m shorting QQQ and buying gold, silver and crypto for the liquidity and potential debasement that follow.
The technology is fantastic. The economics haven’t caught up. The debt is the Achilles heel.
An AI ROI miracle could change that. Miracles aren’t a sensible base case.
References / further reading
