AI Borrowers Now Hold 48% of Long-Dated U.S. Investment-Grade Issuance

AI-related borrowers now account for nearly half of long-dated U.S. investment-grade corporate bond issuance, according to internal portfolio analysis released by PIMCO this week. The shift leaves 48% of new 20-year-plus duration supply concentrated in technology infrastructure credits—cloud operators, data center REITs, and semiconductor manufacturers—up from 19% in early 2023. The firm's strategists noted the change forces portfolio managers into either accepting single-sector concentration risk or forfeiting benchmark-relative yield.

The rebalancing began quietly in mid-2024 when Microsoft, Amazon Web Services, and Google Cloud floated a combined $47 billion in 30-year paper to fund compute capacity expansions. That wave continued through 2025 with secondary issuers: CoreWeave issued $3.2 billion in 25-year notes in March, Digital Realty followed with $2.1 billion in April, and Nvidia's financing arm placed $8.5 billion across the curve in September. Traditional long-duration issuers—utilities, pipelines, and telecoms—collectively reduced gross issuance by $22 billion year-over-year as AI credits absorbed investor demand. The Bloomberg Barclays U.S. Corporate Long Index now tilts 11.4 percentage points heavier toward technology than it did eighteen months ago.

The composition shift creates two problems for allocators. First, duration and sector risk now move together: a 100-basis-point move in tech credit spreads delivers the same portfolio volatility as a 60-basis-point parallel shift across the entire curve, eroding the diversification benefit long bonds historically provided. Second, the credits themselves carry execution risk tied to capital spending cycles that can reverse without warning. PIMCO's credit team pointed to the $140 billion in planned AI infrastructure spending announced since January 2025, much of it financed with 20-to-30-year paper, as a potential source of refinancing pressure if utilization rates disappoint or energy costs spike. The firm's quantitative models now treat long-tech duration as a distinct risk factor, separate from both interest-rate beta and traditional credit beta.

Allocators face limited alternatives. Stepping down the curve into 10-year corporates sacrifices 80 to 110 basis points of yield depending on the credit. Moving into structured credit—CLOs, CMBS—introduces different but equally concentrated exposures, particularly as commercial real estate and leveraged lending markets digest their own post-rate-hike adjustments. A handful of portfolio managers have rotated into European long-duration corporates, where utilities and industrials still dominate issuance, but foreign exchange hedging costs reduce the effective spread pickup to 30 basis points or less. The more disciplined approach, per PIMCO's positioning, involves underweighting the long end of U.S. corporate indices entirely and reallocating that duration budget to sovereign curves or inflation-linked securities.

Watch for refinancing announcements from the 2024-2025 cohort of AI infrastructure issuers starting in Q2 2027, when early call provisions begin to unlock. If utilization data from hyperscale operators softens or energy input costs rise faster than modeled, expect spread volatility to migrate from the equity volatility surface into long-duration credit. The next inflection point arrives in six to nine months, when the first wave of AI-funded data centers reaches operational status and actual cash-on-cash returns become observable. Until then, the index composition itself remains the risk.

PIMCO's own book reduced long-tech duration exposure by $4.7 billion notional across separately managed accounts in October, the largest single-month sector rotation the firm has executed since the regional banking stress of March 2023.