Almost Half of New Long-Dated Investment Grade Is AI Paper — Owning the Index Is Now Owning AI Capex

• Financial issuers are excluded, so the concentration is a corporate-capex story rather than a bank-funding story.
• The figure measures gross issuance, not index weight: hyperscalers are still only 4.7% of the FTSE US Broad Corporate index by market value, so the flow has run far ahead of the stock of AI paper investors already hold.
• The stock of AI-tagged debt is not small either: hyperscaler investment-grade debt across FTSE's index family reached $500bn at end-August 2026, with Amazon at $144.8bn, Alphabet at $120.4bn, Oracle at $117.2bn, Meta at $81.2bn and Microsoft at $36.5bn.
• Measurement choice matters: on the broader ICE US Corporate index Amazon carries 1.10% of the market versus 1.39% on Bloomberg, because the ICE benchmark holds 11,755 issues against Bloomberg's 8,909, including more privately placed paper.

• Amazon raised long-term borrowings by $107.5bn in nine months, from $65.6bn at Dec. 31, 2025 to $173.1bn at Sept. 30, 2026, lifting total debt including $98.8bn of lease obligations to $271.9bn while net debt more than tripled to $216.2bn.
• Meta went from $48.0bn of net debt at the end of 2025 to $100.9bn at Sept. 30, 2026, even as trailing free cash flow to the firm stayed positive at $30.6bn.
• Oracle held borrowings at $125.3bn against $36.4bn of cash at Aug. 31, 2026, and capital spending ran at 82.6% of revenue in fiscal 2026 and 105.4% on a trailing basis.
• Microsoft and Alphabet still generate cash and carry under 0.6x net leverage, which is why their paper prices inside the cohort rather than at its edge.

• Duration contribution has reordered the index: Amazon climbed from 20th to 2nd among contributors to the US index's duration and Meta from 31st to 8th, while Alphabet is now the largest single contributor to the pan-European index's duration.
• Insurers roughly halved their large orders in 30-year AI deals between the first and second quarters of 2026, and new-issue concessions widened to as much as 20 bps — the tell that the marginal buyer is stepping back, not the marginal issuer.
• European exposure is far more concentrated than the US: hyperscalers hold 1.52% of the European index's market value but 2.6% of its duration, and Alphabet alone is the largest duration contributor there.
• Bonds maturing beyond 20 years make up 0.6% of the euro corporate index against 16% in the US, so an identical AI supply wave would hit European duration far harder than it has hit Wall Street.

• Agency MBS is the substitute that needs no argument about AI capex at all: PIMCO's carry test, the 30-year FNCL par coupon index yield less the average of 5- and 10-year Treasury yields, stays positive against the broader investment-grade universe, and the pickup over US IG corporate spreads stood at 33 bps on Nov. 5, 2026.
• AAA non-qualified mortgage paper pays investors to hold extension risk that most buyers refuse, and issuers have begun funding senior coupon step-ups out of interest paid by lower tranches, which accelerates the paydown the senior holders care about.
• Dollar emerging-market investment-grade paper offered a median spread pickup of about 40 bps over risk-matched comparables in 2026, and a hypothetical shift into emerging-market external and local debt raised ex-post portfolio Sharpe ratios by up to 0.13 over three years.
• The diversifiers work because they deliver duration without the narrative. Each one is a way to own long credit while owning nothing of the AI capex cycle that now dominates the long end of US investment grade.

The repricing in AAA non-QM creates an intriguing opportunity.

• Voya Financial's published guidance for longer-dated AI debt is to choose carefully and stick with quality: most of the long-dated AI-related private placements fund data center projects, which it calls an uncomfortable hybrid of commercial real estate and infrastructure, and it advises avoiding medium- and long-duration below-investment-grade data center debt and data center CMBS despite the headline spread.
• Capital rules push insurers out of exactly the paper that pays best: the proposed life risk-based capital factors raise the BBB- charge by 51% and the BB charge by 454%, while the single-A tranche is the position insurers actually prefer and they have broadly resisted spreadier BBB tranches.
• Dealer desks are the other side of the same trade, and the reported statistics understate their book: private placements that migrate into the index through bank paper or yield-to-investment-grade conversions push AI credit into index-eligible supply after the fact, so published AI shares are a floor rather than a ceiling.
• Pension and insurance duration programs that benchmark to an investment-grade index have absorbed the same exposure without an allocation decision, and the hyperscaler cohort's own five-year volatility of 9.49% against 7.66% for the index is the risk they were not compensated for taking.

• Next two quarters: the trade is dispersion inside the cohort, not default. With hyperscaler spreads at 119 bps against 86 bps for the index and single-name CDS ranging from 56 bps to 331 bps, quality selection pays long before any credit event does.
• Next two quarters: concession, not spread, absorbs supply. New-issue concessions have already widened to as much as 20 bps, and with 30-year AI orders from insurers roughly halved since the first quarter, primary-market concessions are the fastest-moving price in the complex.
• Next two quarters: the $220bn-$300bn of hyperscaler issuance forecast for 2027 versus $279bn in 2026 means the calendar, not the macro backdrop, sets the marginal price of AI credit.
• Through 2030: hyperscalers could approach 10% of the US investment-grade market on JPMorgan Chase research estimates, at which point index duration and index spread are decided by five corporate boards rather than by the credit market.
• Through 2030: the counterweight is real. The U.S. Treasury Borrowing Advisory Committee found no clear evidence of tech issuance crowding out Treasuries or non-tech credit as of Nov. 4, 2026, even though fixed-income duration supply rose to $5.1tn in 2026 from $4.0tn in 2025 with tech investment-grade and mortgage-backed securities accounting for the entire increase.
• What breaks the thesis: spreads track hyperscaler free cash flow, not ratings. Amazon already swung to -$43.4bn of trailing free cash flow to the firm, so a second and third quarter of deteriorating cash generation would reprice the long end with no downgrade anywhere in the cohort.