Capital Flows
· Capital Flows
· July 07, 2026 at 03:13
· ⏱ 2 min read
| Read on Substack ↗
Summary
The article argues that the financialization of AI compute—via futures, perps, and the first compute futures ETFs—is a largely unnoticed but critical market development. It warns that the same mistake of underwriting high-volatility assets with low-vol credit that doomed IBM mainframe lessors in the 1970s is being repeated with GPU financing, and that the yen carry trade providing liquidity to AI names is vulnerable to a sharp unwind when Japanese real rates shift or bond issuance disappoints.
•Corporate credit issuance hit $110 billion in June (strongest month since March, nearly double last June), funding data centers and factories.
•Six ETF issuers have filed preliminary prospectuses for compute futures ETFs before the underlying futures even trade, signaling early-stage market development.
•GPU residual value is the fault line: the closest precedent is IBM mainframes in the 1970s, where lessors who booked aggressive residuals went bankrupt within months of IBM price cuts.
•The yen carry trade keeps exporting liquidity to US AI and compute names; the trade unwinds when Japanese real rates shift or the Ministry of Finance issues below expectations.