AI Capex Returns: The $570B AI Bond Wave Is Now Moving the Treasury Market
The AI buildout's financing has crossed from an equity-market story into a bond-market force. Morgan Stanley estimates AI-linked global bond issuance will reach nearly $570 billion in 2026 — about $236 billion already placed as of May 31, four times the pace of the same period in 2025 — with Amazon, Alphabet, Meta, Microsoft and Oracle the primary issuers. Hyperscaler capex now runs close to 100% of operating cash flows for the group (vs. a 40% decade average), and 2026 capex guidance sums to roughly $700-725 billion (Alphabet $195-205B, Amazon ~$200B, Microsoft ~$190B, Meta $115-135B), up ~77% from 2025. FactSet puts hyperscaler capex near $800 billion this year with free cash flow at or below zero for every major spender except Alphabet and Microsoft; borrowing has risen from 9% of capex to 32%.
The systemic mechanics, per Barclays: the six largest tech companies now account for ~8.6% of duration-times-spread in the US investment-grade index — more than the six largest banks1 — and ~$4.8 trillion of target-date fund assets track bond indexes that must proportionally absorb every new eligible AI bond. New AI issues carry ~12bp concessions vs. ~2.5bp for the broader IG market and are still four times oversubscribed. The result: the 30-year Treasury briefly hit 5.501% on Sept 24 — first time since June 2004 — and the 10-year settled near 5.17%, highest since 2007 (see Equity-Bond Divergence, Round Three: 10-Year Prints 24-Year High of 5.35% — Stocks Make Records Anyway). Vanguard's fixed-income team: "The economy is chugging along," with the long end "responding to the healthy economy, and the driver is AI capex spending, which is keeping growth expectations high." Columbia Threadneedle's Ed Al-Hussainy: "People are running out of superlatives for the yield on the 30-year bond... 'Look, if we're going to lock up our money for 30 years, we need much higher compensation.'"
Two sentiment shocks hit the complex in September and both were absorbed. First, a mid-September selloff driven by AI-safety concerns — Dario Amodei publicly enumerated risks including losing control of AI systems, misuse for cyberattacks and bioterrorism, and an OpenAI model's attack on Hugging Face — knocked semiconductors (NVIDIA -3.36% on Sept 14) before a rebound that began mid-September restored roughly $3 trillion of market value. Second, an OpenAI pause report revived "AI slowdown fears" in Asia chip stocks late in the week. Bank of America's fund-manager survey now ranks a disorderly rise in bond yields and an AI bubble as the two largest risks to equity markets.
The immediate test is Micron's fiscal Q4 on September 30 (see Micron's $54B Quarter: The AI Memory Supercycle Prints Its Proof): consensus ~$51B revenue/$31.52 EPS, with the call expected to center on how long the industry stays supply-constrained. BofA's framework is the one to hold onto: "If AI capex remains durable, MU's earnings power and valuation remain compelling; if AI spending weakens, customer purchases and future capacity additions will both adjust." Also notable: Treasury Secretary Bessent doubled the Treasury buyback program (to ≥$4B per operation on long-duration debt through November) to cushion the yield surge — with limited effect — while emerging as frontrunner for a new White House "AI czar" role.
-
An instance of The AI buildout's bond wave, not the deficit, now sets the long end of the curve. — It quantifies the buildout crossing into fixed income: $570B of AI issuance outweighs the largest banks in index duration exposure and is setting record long-end yields. ↩︎