Key Takeaways

  • Hyperscaler and Nvidia debt issuance is tracking near 70% of total US Treasury bond issuance through 2026.
  • John Coogan attributes recent yield spikes to geopolitical shocks like the Iran conflict rather than AI borrowing.
  • Private AI infrastructure cannot serve as liquid collateral, creating a distinct risk profile from sovereign paper.
  • John Arnold warns that synchronized AI data center construction mirrors historical commodity boom-and-bust cycles.

The Disagreement

A massive wave of corporate debt is funding data centers, GPU clusters, and power substations. The sheer scale of borrowing has sparked a debate over whether Silicon Valley is crowding out sovereign debt and pushing Treasury yields upward.

Jordi Hays points to the sheer volume of supply hitting debt markets. He notes that hyperscaler and Nvidia debt issuance equals roughly 70% of total Treasury issuance through 2026. When tech giants issue massive corporate bonds to build physical infrastructure, they compete directly for global capital. As Coogan observes about the gold rush: “why would you invest in literally anything other than data centers when they have such a short payback period and you can throw tens of billions at it.”

Yet Coogan rejects the idea that this capex wave caused the recent spikes in bond yields. He points out that the AI infrastructure push has been building steadily for years without destabilizing rates. In his view, short-term debt market volatility stems from energy shocks and global conflict: “I think that the recent spike is 80% Iran war because I think that the AI buildout has been going on for a while.”

Coogan also draws a sharp line between corporate data center debt and sovereign paper: “Those opportunities are not riskless and importantly you can't use private illiquid AI capex as collateral.”

Who's Right (and When They're Wrong)

Coogan is right on the short-term mechanics. Geopolitical risk and oil price shocks move sovereign debt yields within days. Massive tech bond sales take quarters to price and settle, meaning day-to-day rate spikes reflect wartime oil risk rather than data center construction loans.

Where the crowding-out theory gains teeth is the multi-year horizon. When hundreds of billions in corporate debt lock up long-term credit, yields stay sticky on the way down.

Former hedge fund manager John Arnold adds a structural warning that threatens both sides: oversupply. Tech giants are all reacting to identical market demands at the exact same moment. Arnold explains the trap: “every commodity market, you know, goes through these booms and busts. And that's because you have the producers see the same price signal.”

When every hyperscaler borrows at peak capacity to construct identical compute infrastructure, capacity overshoots actual end-user revenue. If power and GPU capacity flood the market together, payback periods stretch, margins collapse, and the debt load becomes dangerous.

What to Do With This

Stress-test your infrastructure budget for a sudden shift in compute pricing. If you are locking into long-term cloud commitments or hardware leases, negotiate one-year break clauses rather than three-year locked rates. The synchronized buildout will either cause a capacity glut that drops inference prices, or higher capital costs that make fixed debt expensive.