Post

Jonathan Todd
@yontodd.substack.com

Compute demand isn’t slowing, and 8% borrowing costs aren't the problem when GPU projects generate 25% IRRs. The real vulnerability is token price deflation squeezing cash flows while carrying high single-digit coupons. substack.com/@yontodd/not...

substack.comAI can afford 8% money. Uncle Sam cannot.High single-digit coupons on AI is resetting the global price of capital and leaving the rest of the bond market to pick up the tab.

0 reposts1 like

Dean Lee
@deanlee.info

Amortizing a 25% project IRR requires revenue duration that matches the debt maturity. Hardware depreciates on an eighteen-month cycle while open weights compress per-token pricing every quarter, leaving late-cycle cash flows exposed against fixed-rate coupons.

Jonathan Todd
@yontodd.substack.com

Sorry, just to clarify… you're arguing the effective return drops below the coupon because rapid depreciation and falling token costs squeeze late-cycle cash flows? Token pricing compressing, but with a 25% base IRR against an 8% coupon, there's a wide MoE before debt service actually underwater

Dean Lee
@deanlee.info

That margin vanishes because rental rates float while power and coupons are fixed. When new silicon offers 3x FLOPs per watt, legacy clusters can't sustain pricing without losing utilization. In a high-fixed-cost asset, a 40% drop in realized revenue wipes out the equity cushion entirely.

Jonathan Todd
@yontodd.substack.com

On the other hand… bsky.app/profile/carl...

Carl Quintanilla@carlquintanilla.bsky.social

Yeesh. 👀 (Bloomberg) - The most likely long-term outcome of the AI boom will be massive capital destruction in the US, with market share going to cheaper open-source Chinese models, according to Jefferies Financial Group Inc. @bloomberg.com www.bloomberg.com/news/article...