Charted: The US Absorbs ~72% of Funded AI Credit — HQ, Facilities, and ETF Maps Disagree

Our theme already maps who funds the AI build-out. The July research put hyperscalers on a Goldman path toward ~$250B / ~33% of AI capex funded with IG bonds. The mid-year update opened a ~$489B AI-related debt perimeter. The August stock-map update replaced flow-only tallies with a Booth/Hepp funded channel stock near $1.065T. The August 202608 concentration companion then scored distribution: HS senior unsecured alone ~49%, top-3 channels ~91%, Amazon ~41% of the five-name IG spine. This post answers the map question desks trade next: where does that activity, risk, and capacity sit geographically — and do HQ, collateral, books, and ETF domiciles agree?

The interactive dashboard above is built as a geography lens, not another Top-k issuer ladder. Toggle Regions, Facilities, Books & ETFs, and HQ vs assets. The punchline is deliberately multi-map. On funded credit stock, the United States absorbs about 72% of ~$1.065T. On hyperscaler IG issuer HQ, the US share is 100%. On project / data-centre collateral, the US share falls to about 52%. On currency books, USD still clears ~78% of the five-name YTD spine. On ETF listing domicile, US listings take ~91% of 2025 thematic flows. Same theme, five different geographic tips.

The geography scoreboard

Read the table as a family of maps, not one slogan. Funded-stock geography says the US is the modal home of outstanding AI-infra claims. Issuer HQ geography says the borrowers who print the thickest sleeve are entirely US-domiciled. Facility geography says the collateral is more plural — Europe and Asia-Pacific together clear roughly 42% of project/DC finance. Book geography says most of those US HQ notes still clear in USD. ETF geography says public-market ownership demand is even more US-listed than the credit stock. Averaging these rows into “AI financing is American” is true at the tip and false as a single map.

Funded stock: the US is ~72% of outstanding claims

Open Regions. On the Booth/Hepp funded map (~$1.065T), the United States holds about $767B / 72%. Europe is ~13%, Asia-Pacific ~9%, and MEA + LatAm + residual close the last 6%. Top-3 regions (US · Europe · APAC) clear about 94%.

That US tip is not the same as “all AI debt is US IG.” Roughly $520B of the US slice is hyperscaler senior unsecured attributed to US issuer HQ. The rest is project finance, private credit, ABS, and GPU-secured claims that sit where facilities and borrowers sit. Flip the metric to Dollars and the absolute gap is obvious: the US alone is larger than Europe + APAC + MEA + LatAm combined.

The stacked path underneath (2024 → 2026 YTD) shows the US dollar stack thickening fastest in absolute terms even when share stays sticky in the low-to-mid 70s. Geography concentration can rise in levels without every share meter moving the same way — the same hinge the concentration companion showed for channels versus issuers.

Facilities: collateral is more plural than HQ

Toggle Facilities. Project / data-centre finance (~$250B) is the facility-tied sleeve of the funded map. Here the US share is about 52% — twenty percentage points below the funded-stock tip and forty-eight points below HS IG HQ. Europe takes ~24% (Dublin / Nordics / Frankfurt corridors); Asia-Pacific ~18% (Tokyo / Singapore / Mumbai tips); MEA + LatAm + residual ~6%.

Inside the US project slice (~$130B), Northern Virginia remains the densification tip at roughly 25% of US project credit and ~13% of the global project perimeter. Texas and the Midwest print smaller absolute shares but faster YoY growth — the same inland shift our AI capex spend geography post tracks on the spend side. Flip corridors to YoY growth or Risk score and NoVA still leads interconnect risk while Texas leads growth. Credit capacity and growth are two different US maps.

This is the operational hinge for lenders. Desks that underwrite “AI credit = US Mag-7 bonds” are reading the HQ map. Desks that underwrite project finance are reading a collateral map where Europe and APAC together are almost as large as the US tip. Both sit inside the same $1.065T stock.

Books and ETFs: USD and US listings thicken further

Toggle Books & ETFs. On the five-name hyperscaler IG YTD spine (~$194B), USD books clear about 78%, EUR about 18%, and other FX about 4%. Amazon and Meta EUR megadeals widened the European book sleeve without relocating issuer HQ. Currency geography is a placement map: where the paper clears, not where the servers sit.

Flip the lens to ETF domicile. FactSet’s 2025 thematic + mega-tech inflow sleeve was about $43.5B. US listings absorb ~91%; UCITS Europe ~6%; APAC residual. QQQ alone is ~$21.7B of the US tip — the same public-markets concentration the August stock-map update and concentration companion already flagged. Equity flow geography is even more US-centric than funded credit stock.

Pair the two panels: USD books and US ETF listings are the public-markets plumbing around a credit stock that is already US-heavy. Europe’s larger role is in facility collateral and EUR placement, not in issuer HQ or ETF domicile.

HQ vs assets: the mismatch meter

Toggle HQ vs assets. For each channel, compare US funding HQ share with US asset / collateral share. Hyperscaler senior unsecured shows the extreme gap: 100% US HQ versus roughly 58% US facility attribution on the spend map — a ~42 pp mismatch. Private credit is next: GP HQ ~84% US versus asset geography ~48% US (~36 pp). Project finance is nearly aligned (HQ ~58% / assets ~52%). GPU-secured SPVs again print a thick HQ tip against more global fleets.

The scatter beside the gap bars puts corridor project credit dollars against interconnect risk. NoVA sits upper-right: large credit tip, highest risk score. Texas sits mid-credit with extreme growth. Dublin and Singapore show how European and APAC corridors can print meaningful credit without matching US HQ dominance. Filter the scatter to Europe or APAC to see the non-US tips without the US visual field dominating.

This mismatch is why a single choropleth fails. Capital formation (who issues, which GPs raise, which listings clear creations) is more US-concentrated than physical collateral. Stress that starts in a NoVA interconnect queue is a facility story; stress that starts in Mag-7 IG technicals is an HQ / book story. The maps disagree on purpose.

Who is exposed — and what would change the story

Exposed: Global IG portfolios that treat “AI credit geography” as diversified across regions when ~72% of funded stock and 100% of HS IG HQ sit in the US; European project lenders who correctly own facility risk but incorrectly assume their book diversifies the theme’s HQ concentration; ETF allocators who read UCITS AI wrappers as a geographic hedge when ~91% of thematic flow still clears in US listings; private-credit LPs who underwrite US GP platforms without noticing ~half of assets can sit outside the US; risk books that conflate USD book share (~78%) with facility share (~52%).

Relative winners under current rules: US HQ issuers that can still clear multi-tranche USD and EUR books without relocating domicile; US GP platforms that raise against a global DC asset map; NoVA / Texas corridor lenders at the densification tip; US-listed mega-tech ETF complexes that capture ownership demand regardless of where the next campus interconnect clears.

What would change the story: a sustained rise in non-US issuer HQ so HS IG HQ share falls materially below 100%; project/DC collateral where Europe + APAC together exceed the US tip for several vintages; USD book share falling below ~60% as EUR / local books absorb Mag-7 AI notes; thematic ETF flows dispersing so US listings hold less than ~70% of the sleeve; private-credit GP HQ diversifying so the US GP share falls toward the ~48% US asset share. None of those are the central print on this geography vintage.

Caveats and methodology

• Funded-stock geography (~$1.065T) attributes HS senior unsecured to issuer HQ and project / private / ABS / GPU sleeves to facility or borrower geographya desk reconstruction, not a regulator segment filing.
• Regional shares are estimated from disclosed HQ domicile, FactSet currency tags, and Synergy/CBRE-style capacity roll-ups; treat top-1 / top-3 as order-of-magnitude maps.
• Project corridor dollars inside the US (~$130B) are capacity-weighted tips, not loan-tape extracts; interconnect risk scores are ordinal desk ranks.
• Currency book shares cover the five-name IG YTD spine (~$194B), not the full funded stock.
• ETF domicile shares use 2025 FactSet thematic listing tallies (flows, not AUM); QQQ is a Mag-7 proxy, not a pure AI product.
• HQ vs asset gaps compare two different attribution rules on purposedo not average them into one “true” US share.
• Lease overhang (~$675B) remains outside funded totals; the ~68% US lease tip is a commitment geography aside, not part of the 72% funded headline.
• Cross-border SPVs in the residual slice can sit in multiple jurisdictions; the residual is a closing plug, not a geographic claim.

The shareable takeaway

AI financing geography is US-heavy — but which US share depends on the map. On funded credit stock, the United States is about 72% of ~$1.065T. On hyperscaler IG issuer HQ, the US share is 100%. On project / DC collateral, the US share is only about 52%. On currency books, USD clears ~78%. On ETF listings, the US takes ~91% of thematic flows. The build-out is funded in credit and public markets through a system that looks globally diversified in campus press releases and concentrated once you rank HQ, stock, books, and listings — three of which tip harder toward the United States than the facility map does.

Related reading: Aug 202608 concentration companion · August stock-map update · AI capex spend geography · AI financing research 2026.