Update: AI Infra Credit Stock Hits ~$1.07T — Plus $675B in Uncommenced Leases

What changed since the Q3 flow vintage

In our Q3 financing update the live numbers were about theme weight inside credit supply: AI-related debt still near $489B year-to-date, hyperscalers still ~40% of that stack, and desk synthesis lifting the theme’s share of USD investment-grade issuance from ~18% to ~23% (HY to ~20%). Private data-centre transactions since early 2025 printed ~$200B, and Meta/Amazon long paper kept pricing below official AA. That post answered the absorption question — is the theme still claiming more of the primary calendar?

This August 202608 vintage answers a different question credit and risk desks trade next: where does the financed stock actually sit once you leave the flow perimeter? Stefan Hepp’s Chicago Booth Review synthesis (Aug 7, 2026), drawing on The AI Infrastructure Debt Complex, maps funded channels at roughly $520B hyperscaler senior unsecured, $250B project and data-centre finance, $60B infrastructure/ABS, $200B private credit (estimate), and $35B GPU-secured specialist debt — about $1.07T in total. On top of that funded map, S&P Global Ratings has identified about $675B of signed but not-yet-commenced hyperscaler lease obligations that never appear in the funded-debt totals.

The dashboard above is built as a vintage delta: flow-vs-stock bars, channel stock map, hyperscaler issuance path, new Aug disclosures, equity→credit stress scatter, first-loss pie, named SPV/GPU structures, and carried ETF sentiment capacity.

The headline table: Q3 flow print vs Aug 202608 stock map

| Metric | Q3 2026 update | Aug 202608 stock map | Δ | |--------|---------------:|---------------------:|--:| | AI-related debt YTD (flow) | ~$489B | ~$489B | Perimeter carried | | AI share of USD IG / HY supply | ~23% / ~20% | ~23% / ~20% | Share carried | | Private DC deals (since early ’25) | ~$200B | ~$200B | Carried as flow | | HS senior unsecured stock | Flow lens only | ~$520B | New stock | | Project / DC finance stock | 2027 path ~$300B | ~$250B outstanding | Stock lens | | Infra / ABS stock | Not scoped | ~$60B | New | | Private credit stock (est.) | Deals ~$200B | ~$200B stock | Stock restatement | | GPU-secured specialist debt | Not scoped | ~$35B | New | | Funded channel sum | n/a | ~$1.07T | Headline stock | | Uncommenced leases (S&P) | Not scoped | ~$675B | Overhang | | Stress credit-loss band | Not published | ~$60–140B | New |

Two readings matter. First, the Q3 dollar flow perimeter and IG share did not break — we carry $489B and ~23% deliberately so this update does not pretend Booth rewrote Goldman’s mid-year tally. Second, the stock map is additive: ~$1.07T funded channels plus ~$675B lease overhang is a different object from year-to-date issuance. Do not “correct” $489B upward to $1.07T; they measure different things.

Funded stock ≈ $1.07T — about 2.2× the Q3 flow perimeter

Toggle the dashboard’s Stock map lens. The left panel puts three bars side by side: the Q3 $489B flow perimeter, the Aug $1.07T funded stock, and the $675B lease overhang. That visual is the point of the vintage. Flow tells you how much AI-tagged paper cleared the calendar this year. Stock tells you how much institutional capital already sits against AI infrastructure cash flows — including channels that never show up as hyperscaler IG bonds.

The vertical channel panel decomposes the funded map. ~$520B of senior unsecured hyperscaler bonds sit with bond funds, insurers, and pensions at the lowest-risk end of the spectrum. ~$250B of project and data-centre finance is facility-tied, shorter-maturity, and dependent on contracted tenant cash flows. ~$60B of infrastructure and ABS is the structured sleeve. Private credit’s ~$200B is an author estimate — hard to observe, overlapping the Q3 private DC deal tally without being identical to it. GPU-secured specialist compute adds another ~$35B, with chips themselves as collateral on a two-to-three-year technology cycle.

Use Focus → Funded only to strip the lease bar, or Lease overhang to isolate the S&P commitment. Junior / GPU highlights the sleeves where residual-value and collateral risk concentrate — private credit and chip-secured lending.

The lease overhang is the silent vintage number

Reported corporate debt is an incomplete measure of AI infrastructure commitments. Hepp cites S&P’s identification of about $675B of signed but not-yet-commenced lease obligations across the hyperscalers — excluded from the funded-debt totals in the stock map. That figure sits beside the ~$1.07T funded channels, not inside them.

For capital-markets readers this is the hinge between the capex spend update and the financing theme. When Big-5 midpoints climb toward ~$858B, the residual is not only senior notes and project loans; it is also multi-year lease capacity that will eventually print as PP&E, rent, or off-balance financing. An institution can underwrite “manageable hyperscaler leverage” on the bond stack and still inherit lease-linked exposure through infrastructure funds, real-estate debt, and ABS.

The Q3 post’s private DC ~$200B channel was the overflow flow. The August lease print is the overflow commitment. Both matter; they are not substitutes.

Issuance path: H1’26 already above full-year 2025

Booth’s frame on the hyperscaler bond calendar is blunt. The five names issued about $120B of corporate bonds in 2025 versus an average of roughly $28B annually between 2020 and 2024. Through the first half of 2026, issuance had already exceeded the full-year 2025 amount — consistent with the Goldman / LSEG spine we carried in Q3 (~$194B YTD toward a ~$250B full-year path, ~33% of capex).

The dashboard’s issuance line makes the slope visible without restating the IG supply-share story. Theme weight inside USD IG supply can sit at ~23% and the absolute hyperscaler calendar can still be accelerating; those are compatible. Saturation shows up first as cover-ratio compression and spread concession, then as migration into project finance, private credit, multi-currency packages, and structured SPVs — exactly the channels the stock map is trying to count.

Stress: $10–14T equity shock maps to $60–140B credit loss

Switch to Stress & loss. Hepp’s illustrative calibration is deliberately not a base-case forecast: a severe debt re-rating could cut AI-linked equity value by about $10–14 trillion, while realized credit losses in the accompanying band sit nearer $60–140 billion. Equity loss does not become credit loss one-for-one. Strong investment-grade names first reprice via wider spreads; project and asset-backed sleeves react through tenant credit, covenants, advance rates, and refinancing windows.

The more important structural claim is distributional. Under the paper’s assumptions, first-loss and junior positions sit mainly outside the regulated banking system — with private-credit limited partners (pensions, endowments), private-equity-owned insurance platforms, and, to a lesser extent, BDC investors. Banks remain important providers of project and senior secured financing but tend to sit higher in the capital structure. That is why the dashboard’s loss-bearer pie is directional, not a regulatory capital model: it shows who marks the junior risk, not a prediction of bank failure.

July’s market episode — a freely available Chinese model approaching leading US systems, plus regulatory proposals that could constrain the most powerful models — illustrated how quickly demand assumptions can wobble. The August stock map does not say a correction is inevitable. It says the financing system now has enough layered claims on AI cash flows that diversification-by-asset-class can still leave portfolios exposed to a common repayment story.

Structures: $35B Anthropic SPV and the GPU collateral problem

Toggle Structures / ETFs. Named deals illustrate how debt leaves the hyperscaler balance sheet without leaving the AI cash-flow complex.

In June 2026, Apollo and Blackstone helped raise about $35B of debt for Anthropic’s computing capacity through a special-purpose vehicle: the money bought chips and leased them back, keeping the debt off Anthropic’s books. Broadcom’s residual promise let senior lenders price closer to investment-grade — until the conditions that trigger the guarantee are exactly the conditions that weaken it. Senior notes become a claim on a manufacturer’s promise; junior notes become a claim on used chips whose secondary market has never been stress-tested at scale.

That $35B SPV sits beside the broader ~$35B GPU-secured specialist book in the Booth map. Chip collateral depreciates on a two-to-three-year cycle. Vendor financing — including Nvidia’s September 2025 agreement to take about $6.3B of CoreWeave unsold capacity through early 2032 — converts hardware residual value into bankable support that is hard to observe in public tallies. These structures are why a clean “hyperscaler IG only” screen understates theme risk.

The equity/ETF panel is carried on purpose. FactSet’s 2025 US thematic ETF inflows near $43.5B, with QQQ alone around $21.7B, remain a sentiment capacity meter — secondary ownership, not issuer proceeds. Pair that sleeve with the mid-year financing update if you need the $489B flow perimeter and hyperscaler 40% share in one place.

Who is exposed under the August stock map

Exposed: real-money books that assumed underweight Big Tech bonds meant underweight AI infrastructure credit; institutions holding “diversified” mixes of tech equity, data-centre funds, private credit, and ABS that share the same tenant or demand assumptions; LPs in junior AI-infra credit vehicles that sit first-loss under Booth’s stress band; portfolios that treat uncommenced leases as someone else’s problem until commencement hits the income statement.

Relative winners under current rules: desks that can map claims by cash-flow source rather than legal label; senior holders higher in project and ABS stacks; multi-currency and private channels that clear when USD IG cover ratios compress; investors who keep the Q3 ~23% supply-share print and the Aug stock map in the same risk packet.

What would change the story: a restated Booth/S&P perimeter that collapses funded stock well below $1T or lease overhang well below $500B; a Q4 print that pulls AI’s USD IG share back into the mid-teens while concessions compress; or a broad rating migration that turns today’s technical AA–BBB gaps into cohort downgrades. None of those reverse signals are in the August stock map.

Caveats and methodology

1. Flow ≠ stock. The ~$489B AI-related debt YTD figure is carried from Goldman’s August mid-year estimate (hyperscalers ~40%). The ~$1.07T funded sum is Hepp’s channel stock map. Do not add them or annualize either into the other.
2. Lease overhang is off-balance by design. The ~$675B S&P signed-but-not-commenced figure is excluded from funded totals in the source analysisand in our dashboard’s funded sum.
3. Private credit ~$200B is an estimate. Hepp flags measurement difficulty. It overlaps conceptually with Q3’s private DC deal tally without a perfect identity.
4. Stress bands are illustrative. The $10–14T equity and $60–140B credit-loss figures are stress calibrations, not forecasts of overvaluation or default probability.
5. Loss-bearer pie shares are directional viz aids for first-loss geography under the paper’s assumptionsnot audited capital allocations.
6. IG/HY supply shares (~23% / ~20%) are carried from the Q3 desk refresh; this vintage does not re-estimate them.
7. ETF flows remain secondary-market ownership metrics.
8. Named deals (Anthropic SPV ~$35B, Nvidia–CoreWeave ~$6.3B) are structure illustrations, not a complete census of vendor or residual support.

Primary sources: Chicago Booth Review — How Worried Should We Be About AI Debt? (Stefan Hepp, Aug 7, 2026); Hepp working paper The AI Infrastructure Debt Complex (Jul 2026); Goldman Sachs Exchanges — How AI Debt Is Reshaping Credit Markets (Aug 2026) for the carried flow spine; prior theme posts Q3 financing update and mid-year $489B update. Capex context from the late-Aug spend vintage.

What to watch into year-end 2026

Three coincident signals will tell you whether the August stock map tightens or expands: (1) whether funded-channel estimates keep climbing past ~$1.07T as project finance, ABS, and private credit clear more of the residual — if stock rises while USD IG theme share stays in the low-to-mid twenties, saturation is migrating off the public calendar rather than disappearing; (2) commencement of the ~$675B lease overhang — conversion into funded debt, rent, or PP&E is the real-time meter of how incomplete the bond stack remains; (3) junior and GPU-secured secondary performance — if chip-collateral and SPV residuals trade wide to issue while senior HS paper only widens modestly, first-loss is already behaving like a separate asset class. Until those three clear, the live pair is not Q3’s $489B / 23% alone. It is ~$1.07T funded stock + ~$675B lease overhang — with stress losses still concentrated outside the regulated banking core.