Update: AI Debt Hits $489B Mid-Year — Hyperscalers Are Only 40% of the Credit Wave

What changed since the July research vintage

In late July we published the theme’s baseline: hyperscalers funding roughly a third of AI capex with IG bonds, with Goldman’s path pointing to ~$250B of five-name issuance in 2026 (~33% of capex) and ~$400B / ~35% in 2027. That post answered the issuer question — how do Amazon, Alphabet, Meta, Microsoft, and Oracle fund the build? This update answers the market question credit desks actually trade on next: what moved in the newest official vintage, and how large is the theme once you leave the five-name perimeter?

One Goldman Sachs credit refresh forces the rewrite. On the Exchanges desk conversation recorded in early August 2026, Amanda Lynam and Zach Ablon put a new perimeter around the story: nearly $500 billion of AI-related debt issuance year-to-date — reported in market coverage as about $489 billion by mid-year — of which the hyperscalers are only about 40%. The July YTD print of ~$194B for the five names still stands. What changed is the frame. The binding constraint is no longer “can five cash-rich tech issuers clear the IG calendar?” It is “can the entire AI credit complex — utilities, industrials, data-center JVs, high-yield wrappers, and project finance — clear without saturating indexes and concentration limits?”

The dashboard above is built as a vintage delta: debt-stack pie, supply-share path, spread dumbbells, duration ranks, and overflow capacity when US IG saturates.

The headline table: Jul research vs Aug 2026 update

| Metric | Jul research vintage | Aug 2026 update | Δ | |--------|---------------------:|----------------:|--:| | Hyperscaler IG YTD | ~$194B | ~$194B | Unchanged | | Hyperscaler FY26 IG path | ~$250B | ~$250B | Path held | | Debt share of HS capex (2026) | ~33% | ~33% | Path held | | AI-related debt YTD | Not scoped | ~$489B | New perimeter | | Hyperscaler share of AI debt | Implied ~100% of lens | ~40% | Rest = ecosystem | | AI share of US IG supply | ~7% (2025) | ~18% (2026 YTD) | +11 pp | | Non-USD share of HS supply | ~14% (2025) | ~33% (2026 YTD) | +19 pp | | Long-end (≥15y) AI share | Not highlighted | ~40% of 15y+ IG | Concentration |

Two readings matter. First, the July hyperscaler spine did not break — YTD and full-year paths are intact. Second, the scope expansion is the real vintage delta: roughly $295B of AI-tied debt sits outside the five names, and the theme’s weight inside US IG gross supply has more than doubled versus the 2025 calendar share.

The $489B stack: hyperscalers are a slice, not the wall

Toggle the dashboard’s Debt stack lens. The donut is the point of the update. Goldman’s mid-2026 tally puts AI-related debt near $489B, with hyperscalers contributing about $194B (~40%). The residual ~$295B is the broader ecosystem — power utilities funding AI load, industrials and equipment names, data-center joint ventures, high-yield and leveraged-loan prints, and other credit wrappers that sit next to — but not inside — the Amazon–Alphabet–Meta–Microsoft–Oracle IG calendar.

That is why July’s “hyperscaler debt binge” headlines understate what passive IG and HY portfolios actually absorb. Concentration risk is thematic, not just name-specific. Desk commentary puts AI-related issuance at about 18% of US IG year-to-date supply, up from roughly 7% in 2025 and about 1% in 2024. High yield shows a similar ~18% theme share. Investors who thought they were “underweight Big Tech bonds” can still be overweight the AI financing theme through utilities, JVs, and project vehicles.

Spreads and duration: indigestion is now measurable

The July post documented median new-issue spread widening for Amazon, Alphabet, Meta, and Oracle across the curve — 2–4y from ~30 to 40 bp, 5–7y from ~50 to 60, 20y+ from ~108.5 to 118. That still holds. The August desk vintage adds the portfolio-level meter: Goldman’s AI leader basket moved from tights near 74 bp to nearly twice that (~148 bp) over twelve months. Insurance and real-money order books that once put fifteen $50M+ tickets into the 30-year part of large AI financings in Q1 were roughly half as large by late Q2.

Duration concentration is the second stress vector. Amazon’s duration weight in the US IG index jumped from roughly #20 last year to #1; Alphabet moved from about #86 to #18. Roughly 40% of 15-year-plus IG issuance this year is AI companies or AI-funding names. Long-end paper that looked like a gift to liability-driven buyers in early 2025 is now a concentration and curve-shape problem. New-issue concessions that used to clear at 2–3 bp have printed as wide as ~20 bp on mega hyperscaler deals.

Pair that with the capex intensity update: every major builder’s free-cash-flow cushion is thinner as intensity rises. Credit is not pricing a sudden default wave. It is pricing multi-year releveraging without a clean end date.

FX mix: the IG book is already globalizing

A quieter vintage delta sits in currency mix. Non-dollar bonds accounted for about 33% of hyperscaler supply in 2026 YTD, up from roughly 14% in 2025. CAD, sterling, Swiss franc (including Alphabet’s record CHF print), euro, Aussie dollar, and yen packages are no longer exotic diversions — they are overflow valves when US dollar books thicken. TwentyFour and other credit houses flagged the same pattern: Amazon’s later ~$25B multi-tranche deal cleared at ~1.6× cover versus ~$119B books / much higher cover in March, with concessions of 12–22 bp. Issuers remain opportunistic; investors are selective.

European AI-related supply remains under-contributing relative to US IG in Goldman’s reading — which means EUR credit is still an open channel, not a saturated one. The waterfall Amanda Lynam describes is literal: exhaust depth in one market, open the next.

Overflow: private markets, project finance, and bank-comparable capacity

When the dashboard switches to Overflow channels, the scale changes. Goldman’s case study: if hyperscalers approached the index weights of the largest US bank issuers, incremental US IG capacity alone is on the order of ~$510B — meaningful runway, but not infinite once duration skew and equity-side AI exposure are netted by real-money committees. Against that, private markets dry powder across private credit, infrastructure, real estate, and private equity is framed near $4.5 trillion. Project and data-center finance is sketched at roughly $300B in 2027 above and beyond direct hyperscaler IG. ABS and structured credit enter once campuses are completed and cash flows stabilize.

Access to capital for the AI theme is not the binding constraint in this vintage. Where and at what price are. That is the same conclusion as July — restated with a larger perimeter and harder concentration math.

Equity and ETF capacity still fund the narrative, not the PP&E

The equity channel from the July post still matters as secondary capacity and primary-raise context. FactSet’s 2025 US thematic ETF inflows near $43.5B, with QQQ alone at about $21.7B, remain the public-markets sentiment sleeve. Alphabet’s ~$85B June 2026 equity raise sits beside the bond calendar as proof that issuers will reopen equity when credit books thicken. ETF flows do not fund data-center concrete directly — they price the willingness of public capital to underwrite the theme while primary debt does the heavy lifting.

Caveats and methodology

1. Perimeter differs by source. July’s post used Goldman’s five-hyperscaler global IG path plus Reuters/LSEG deal tallies. August’s ~$489B is Goldman’s broader AI-related debt estimate (hyperscalers ~40%). Do not subtract or annualize across perimeters without labels.
2. YTD is not a run-rate. Mid-year AI debt near $489B does not imply ~$978B for the calendar; front-loading and summer quiet periods matter. Hyperscaler FY path remains ~$250B, not a restatement.
3. Debt share of capex (~33%) is still a hyperscaler research ratio (issuance ÷ cohort capex), not the share of the $489B stack.
4. Supply-share percentages (~18% of US IG) are desk estimates of theme-tagged issuance, not a single official SEC series. High-yield theme share is described as similar, not identical.
5. Spread basket vs new-issue medians measure different things. The AI leader basket is a traded portfolio construct; tenor medians are deal-print statistics from the July Reuters/LSEG read.
6. Overflow capacities ($510B bank-comparable room; $4.5T private powder; $300B project finance) are research case studies and aggregatesdirectional funding maps, not committed pipelines.
7. ETF flows remain secondary-market ownership metrics, carried for narrative capacitynot issuer proceeds.

Primary sources: Goldman Sachs Exchanges — How AI Debt Is Reshaping Credit Markets (recorded Aug 3, published Aug 5, 2026); The Global Treasurer synthesis of the $489B mid-year tally (Aug 18, 2026); CNA/Reuters–LSEG hyperscaler YTD and spread context (late July 2026); prior theme baseline in AI financing research. Capex intensity context from the August intensity update.

What to watch into year-end 2026

Three coincident signals will tell you whether the expanded vintage stabilizes or tightens further: (1) full-year hyperscaler IG versus the ~$250B path — and whether Microsoft stays light relative to peers; (2) AI’s share of US IG / HY supply — if the theme stays near high teens into Q4 while concessions refuse to shrink, saturation is binding even with private-market overflow; (3) project-finance and JV secondary performance — desk notes that 17 of 23 data-center JV deals were already trading wide to issue yields as indigestion spilled from IG into leveraged structures. Until those three clear, the live number is not July’s five-name story alone. It is ~$489B of AI-related debt — with hyperscalers as the visible 40%, and the other 60% reshaping credit benchmarks whether or not the next mega tech print makes the front page.