The Debt That's Making Itself More Expensive

The Federal Reserve held interest rates steady, and the 30-year Treasury yield still climbed to 5.28% — its highest level since 2007 ^[1]. Market analyst Casey Sprake put the oddity plainly.

the bond yield curve is actually doing the the work for the Fed in that — Casey Sprake

The force analysts most often named wasn't monetary policy. It was supply — specifically, the flood of corporate bonds the big tech companies are issuing to build data centers. Hyperscaler investment-grade debt rose about 35% over the past year and now accounts for 30% of all net investment-grade issuance, with forecasts pointing toward $1 trillion ^[2]. Pimco's Marc Seidner calls the pace "too much, too fast," arguing it is crowding out government funding and pushing the 10-year yield toward 4.75% ^[3]. TD Securities and MRB Partners each reach the same conclusion independently: AI issuance, not lost faith in the Fed, is what's lifting long-term yields ^[4][5]. Then the loop turns. Higher Treasury yields set the benchmark for corporate borrowing, so the same debt that pushed yields up now makes the next round of data-center financing more expensive. Morgan Stanley's arithmetic shows how little room there is: a fully optimized data center costs $25 billion a year to rent and generates only $23 billion in output, and the price of AI output is still falling ^[6]. The cash flows are already straining — Alphabet posted its first negative free cash flow while raising capital spending guidance to $205 billion, and Meta's free cash flow fell 91% ^[7]. The government sits in the same loop. The national debt passed $40 trillion this month, with annual interest above $1 trillion ^[8]. Treasury Secretary Scott Bessent's answer is that AI-driven growth will let the country outgrow the bill.

There’s nothing magic about the $40 trillion number, and we can grow our way out of that. — Scott Bessent

But the AI debt that's supposed to produce that growth is the same supply pushing up the yields that make servicing $40 trillion costlier. The counter deserves its weight. JPMorgan's Kelsey Berro notes that record supply has met record demand, and the hyperscalers remain solid credits ^[9]. For now, that holds. But the strain is already visible at the edges: Apollo's Torsten Slok calculates a $1 trillion gap between what AI needs and what public bond markets can absorb by 2030 ^[10], and roughly $100 billion in investment-grade bonds already trade at spreads wider than the junk curve ^[11]. The government's cure and the disease are the same debt. For now the loop has no obvious exit that doesn't pass through higher yields — unless demand grows fast enough to absorb the supply, or AI output starts paying for itself.