Alpha Exchange — Amanda Lynam: GS credit on AI capex financing

Opening

Goldman Sachs Chief Credit Strategist Amanda Lynam maps the AI financing cycle as a multi-year IG supply wave — hyperscaler global IG ~$250bn in 2026 and ~$400bn in 2027, plus ~$300bn of separate data-center/chip project finance in 2027 — against a still-constructive credit backdrop where yield-based buyers (insurers, pensions, foreign) keep spread-widening short-lived and crowding-out is not yet visible. Alpha Exchange (~50m) with Dean Curnutt; return visit after Oct 2023. Ground covered: credit constructive frame and left-tail risks; US IG index AI share vs banks; why debt is raised ahead of any cashflow gap; absorb capacity and private-markets handoff post-~2028; diversification into banks/energy/healthcare/F&B; foreign demand; software 2028–29 maturity wall. Whisper asr from Simplecast audio (YT captions HTTP 429). Prefer Watch. Figures marked (asr).

Key takes

Constructive corporate credit: "good enough" growth, solid leverage/coverage for the broad market, and elevated all-in yields pulling insurers, pensions, and foreign buyers — so spread-widening episodes have been elusive and short-lived; GS still forecasts only modest widening, not historically wide. Expected widening framed as technical indigestion of elevated supply, not broad fundamental deterioration. Left-tail watch: CCC US/EU HY and small private-credit vintages from ultra-low-rate eras (coverage near/below 1). Past peak defaults in leveraged finance via distressed exchanges/formal defaults. Key non-AI risk: rates vol tipping the rates backdrop from tailwind to headwind for credit. [asr]

"Quality/defensiveness" inverted vs years ago — moving up to AA (IG) / BB (HY) hits concentrated tech/project-finance supply and thinner spread cushions vs elevated yields; GS prefers selective move down: overweight BBB vs higher IG, overweight single-B vs BB in US HY (very selective in CCC). DC project finance called out as often BB-rated in HY. [asr]

US IG (Bloomberg index): tech ~10%; AI-related index-eligible notional ~12%; banks still largest at ~21–22%. Path: US IG does more heavy lifting in 2027–28 while runway remains before concentration norms bind; from ~2028 private markets do more — binding constraint is issuer/theme concentration conventions, not IG ratings headroom. Credit's downside focus = higher hurdle for outsized theme weight than equities. [asr]

Debt raised ahead of any CFO–capex gap: gap math would have implied little/no need in 2025–26, yet hyperscalers issued $108bn global IG in 2025 and $229bn YTD by mid-Sep 2026; full-year 2026 ~$250bn; 2027 ~$400bn (+~60%); separate DC/chip financing another ~$300bn in 2027. Debt-financed share of capex assumed to peak at 35% in 2027 (3-month to 2-year monetization lag). Equity-research hyperscaler capex 2026–2030 upwards of $6tn (larger if include 2025 and push to 2031). Hyperscalers only ~40% of total AI-related gross issuance — theme broader (project finance ~22–24%; tech-adjacent software/memory/equipment; utilities/machinery 30% haircut into AI basket). [asr]

Broader US IG gross issuance: 2026 raised to $2.3tn (from $2.1tn; summer slowdown never came); 2027 $2.4tn — pandemic-era record if unadjusted for inflation. Little evidence of crowding-out of other corporates or Treasuries from hyperscaler supply; rates colleagues link some UST upward momentum more to commodities; mechanical Treasury selling from duration hedges on 30y corporates exists but is not AI-specific allocation crowding-out. Non-AI IG spreads contained vs episodic AI-ecosystem indigestion. Banks, energy (geo hedge), healthcare, food & beverage bid as diversifiers as multi-year AI credit exposure grows; allocators also counting total AI factor across equity/converts/private/syndicated. Europe expected to contribute more AI supply into late 2026–27 after under-contributing. [asr]

Foreign demand floor: Fed ~29% foreign ownership of US corporates; H1 2026 foreign net buys $251bn — on pace to beat 2025 full-year record despite USD strength and hedge-cost swings; Europe <½ US market size limits alternatives; hedging costs below 2022–23 peaks. BoJ FSR (Apr): Japanese banks tilting to CLOs/alts/PE vs IG/HY corporates — marginal, >½ foreign flow still Europe. All-in yields: only ~15–20% from credit spread, rest risk-free — IG behaves more like a rates product than 2010–21. HY structure: avg new issue ~$900m (IG >$1.5bn); smaller issuers → private credit; software 2028–29 maturity wall in BSL/private credit the refinancing overhang to watch (AI disruption narrative hard to disprove; early refinancings "gone well"). Real assets (50/30/20 framing) as diversifiers. [asr]

Key math

Hyperscaler capex 2026–2030 upwards of $6tn (asr — GS GIR equity) — multi-year build size. [asr]

Hyperscalers ~40% of total AI-related gross issuance (asr) — theme broader than Mag7 debt. [asr]

Hyperscaler global IG: $108bn 2025; $229bn YTD mid-Sep 2026; ~$250bn FY26; ~$400bn FY27 (+~60%) (asr) — proactive issuance path. [asr]

Separate DC/chip project finance ~$300bn in 2027 (asr) — non-hyperscaler AI credit. [asr]

Debt-financed share of capex peaks ~35% in 2027 (asr — GS assumption) — monetization lag 3m–2y. [asr]

US IG gross issuance $2.3tn 2026 (raised from $2.1tn); $2.4tn 2027 (asr) — market capacity context. [asr]

US IG index: tech ~10%; AI-related ~12% notional; banks ~21–22% (asr — Bloomberg index) — concentration not yet max. [asr]

Foreign ownership ~29%; H1'26 foreign net $251bn US corporate credit (asr — Fed / Treasury TIC) — demand technical. [asr]

Credit spread ~15–20% of IG all-in yield (asr) — rates dominate carry. [asr]

Quotes

"That CAPEX estimate is upwards of $6 trillion." — Amanda Lynam [asr]

"The hyperscalers are only 40% of total AI-related gross issuance. So the theme is actually much broader." — Amanda Lynam [asr]

"What we quickly realized… there was not a need to raise debt in 2025… What instead we saw… hyperscalers globally issued 108 billion… [and] 229 billion [YTD]." — Amanda Lynam [asr]

"We actually see very little evidence of a crowding out as it relates to the hyperscaler issuance." — Amanda Lynam [asr]

"The binding constraint in this financing cycle is more likely to lie in… issuer concentration… and less about how much debt could these firms issue and stay comfortably within IG territory." — Amanda Lynam [asr]

"Is there a tipping point where… rates become a headwind for credit?" — Amanda Lynam [asr]

Variant perception

Priced in — Hyperscalers are heavy IG issuers; AI capex is multi-trillion; IG spreads tight on yield demand; foreign buyers matter for USD credit; private credit absorbs smaller/bespoke borrowers; software AI disruption overhang is a known credit narrative.

What's new — Explicit GS path: $250bn→$400bn hyperscaler IG plus ~$300bn DC/chip in 2027; debt issued ahead of CFO–capex gap (not residual financing); 35% debt-financed capex peak year; hyperscalers only 40% of AI-related issuance; US IG AI share still ~12% vs banks ~22%; little crowding-out with non-AI spreads contained; concentration conventions (not ratings room) as binding constraint → private markets post-~2028; quality/defensiveness inverted toward BBB/single-B; software 2028–29 wall as the maturity focus; 30% haircut on utilities/machinery into AI basket.

Bear case — Forecasts are GS model outputs that have already been raised ($2.1tn→$2.3tn); "little crowding-out" could reverse if AI supply accelerates further or risk-off hits; rates-vol tipping point is acknowledged but un-timed; software disruption could reprice HY/private credit faster than "nuanced" selection assumes; foreign-demand "floor" fails if Japan/Europe repatriate harder than TIC shows.

Discount — GS Global Investment Research credit strategist talking GS forecasts and index work — institutional franchise, not a sell-side pitch for a single deal, but still house-view marketing. Host is markets-pod framing; equity-research capex $6tn is a colleague input she cites, not her primary book.

Positioning

AI capex durability — STRENGTHENS. Multi-year proactive debt issuance ($250bn→$400bn hyperscaler IG; separate $300bn project finance), $6tn 2026–30 capex stack, and debt raised before a cashflow gap all read as financing that assumes the build continues — with IG absorb capacity still open through 2027–28 before concentration norms push more into private markets.

Inference margin inversion — NEUTRAL. Episode is credit/financing microstructure, not token economics; software 2028–29 refinancing overhang is a credit-selection risk, not a lab gross-margin claim.