Q3 Geographic Split: US Takes 57% of Big-5 AI Capex — Northern Virginia Alone ~11%

Our research ledger sized the multi-house AI infrastructure spend universe — Big-5 guidance midpoints, CreditSights aggregates, GS Investment Research paths, and GS Global Institute all-in AI infra. The mid-Q3 spend update then restated the 2026 perimeter to ~$835B gross Big-5 midpoints (~$626B AI-attributed at the CreditSights ~75% convention) after the Jul→Q3 cumulative raise of ~$75B. The Q3 concentration companion answered a different desk question: how thick is the company tip inside that perimeter (Amazon ~27.5% top-1, top-3 ~73%, HHI ~2,211). The late-Aug geography companion already mapped the hotter ~$858B tape. This post locks the mid-Q3 guide vintage and asks the map question on that cooler stack: where do those dollars land geographically — by region, by US metro corridor, and by scenario scope?

The interactive dashboard above is built as a Q3 geography lens. Toggle Regions, US metros, Scenarios, and Risk map. On regions flip share versus dollars and gross versus AI slice; on metros flip global share, US share, or YoY growth; on risk filter by region. The punchline is deliberately multi-map. On regional facility geography, the United States takes about 57% of the mid-Q3 Big-5 perimeter — one percentage point softer than the late-Aug ~58% print. On metro densification, Northern Virginia alone is roughly 10.8% of global Big-5 dollars. On raise geography, about 60% of the Jul→Q3 cumulative dollar raise still lands in US facility tips. And on interconnect risk, the same corridors that print the largest absolute shares — NoVA, Ireland/Dublin, Singapore, AZ/NV — also sit at the top of the ordinal power-risk ladder.

The headline map: four meters, four geographies

Read the table as a family of maps, not one slogan. Regional geography tells you which continents absorb hyperscaler facility dollars. Metro geography tells you which power and interconnect corridors densify inside the US tip. Scenario geography tells you how the US share moves when you switch from Big-5 gross to GS GI all-in AI infra. Risk geography tells you where a grid or interconnect shock becomes a capacity event. Averaging these rows into “AI spend is an American story” or “Virginia is the AI story” is a category error — the US is a regional story; NoVA is a corridor story; GS GI’s slightly lower US weight is a scope story.

Regional facility dollars: US majority, APAC a touch thicker

Filter Regions. Against the mid-Q3 Big-5 gross perimeter of ~$835B, the United States rolls to about $476B, or ~57%. Europe follows near 18% (~$150B) on Ireland, Nordics, Frankfurt/NL, and UK campus tips. Asia-Pacific sits near 17% (~$142B) on Japan, Singapore, India, and Australia ramps — about +1 pp versus the late-Aug geography companion as Japan and India campus tips convert from announcement to spend. Middle East & Africa and Latin America remain mid-single-digit and low-single-digit slices; an analytical residual closes the perimeter.

The regional pie is therefore US-majority without being a monopoly. A system where one country-bloc holds a little under three-fifths of recorded Big-5 facility geography still leaves Europe and APAC with roughly another third combined. That is the geographic twin of the Q3 company concentration ladder: Amazon remains the largest single company; rolling campus densification onto a US regional sheet makes the United States the thickest facility region without claiming that “most AI compute is American” in a planetary sense once you leave the Big-5 perimeter.

Toggle Dollars if you want the absolute ladder rather than shares, and flip AI slice to apply the ~75% CreditSights AI-attribution factor (US ~$357B of ~$626B). Ordering is stable; the visual weight of the US tip becomes clearer beside Europe’s still-material $150B print. The stacked area path for 2024–2027 keeps the same geometry at larger scale: US dollars grow fastest in absolute terms, while APAC and Europe gain share only at the margin as sovereign and enterprise campus pipelines mature.

US metros: Northern Virginia as the global corridor tip

Open US metros. Inside the US slice (~$476B), Northern Virginia leads at about 19% of US facility dollars and ~10.8% of the global Big-5 perimeter (~$90B). Texas (Dallas / Central) follows near 8.4% global; the Midwest power corridor (Ohio / Iowa / Illinois) is close behind at ~7.9% with the fastest YoY growth among large corridors (~49% desk). Pacific Northwest, Southwest (Arizona / Nevada), and Southeast (Georgia / Carolinas) fill the mid-single-digit global bands; a long “other US” residual closes the domestic map.

The metro ladder is the post’s first hinge: country majority ≠ corridor monopoly. The US can be ~57% of regional dollars while NoVA alone is only ~10.8% of the global perimeter — still the thickest single corridor, still not “half the AI build.” Toggle YoY growth to see why Midwest and Texas matter for forward geography even when NoVA still wins the stock tip: power-available inland corridors are pulling incremental campus announcements even as coastal densification remains the absolute-dollar champion.

The companion scatter (global share × interconnect risk) puts NoVA in the upper-right quadrant — large share, high ordinal power risk — with Texas and Midwest slightly lower on risk but still material on dollars. That geometry foreshadows the Risk map view: dollar geography and constraint geography are related but not identical.

Scenario geography: scope changes the US weight

Switch to Scenarios. Stacked bars apply regional weights to four different perimeters. Mid-Q3 Big-5 gross (~$835B) and the AI slice (~$626B) share the same ~57% / 18% / 17% US–Europe–APAC weights by construction. GS Global Institute’s all-in AI infra scenario (~$765B in the Tracking Trillions 2026 frame) is slightly less US-skewed at about 52% US — because the all-in perimeter pulls in more non-hyperscaler compute, power, and international enterprise build. GS Investment Research’s 2027 hyperscaler base (~$1.14T) eases the US share toward ~55% as Europe and APAC ramps compound.

Do not mix scopes when reading absolute stacked heights. A GS GI bar that looks “smaller” than a Big-5 bar is not a forecast downgrade — it is a different universe (all-in AI infra versus company gross capex). The useful comparison is within a house: how the US share moves when the perimeter widens beyond the five hyperscalers, and how 2027 paths redistribute continental weights even if the US remains the majority tip. Pair this panel with the research ledger’s scenario table when you need annual dollar paths rather than geographic weights.

Risk map: dollar giants and interconnect giants overlap

Open Risk map. The spend-share × interconnect-risk scatter puts Northern Virginia at the tip of both axes among large corridors — high global share (~10.8%) and the highest ordinal power-risk score (91). Ireland / Dublin and Singapore sit high on risk with mid-single-digit global shares; Nordics print lower risk with still-material Europe dollars; UAE sits lower on both axes but is the MEA tip. Filter to Europe or APAC to isolate regional clusters without the US corridor weight dominating the visual.

The second hinge is geometric: risk geography is not the inverse of dollar geography. The corridors that attract the most hyperscaler dollars are often the ones with the tightest interconnect queues and the loudest local grid politics — not empty deserts waiting for load. Midwest growth is partly a response to that overlap: desks chase power availability inland while NoVA and Dublin still clear the largest disclosed campus tips. For power-system companions keep the AI power-grid concentration series open; this lens only ranks ordinal interconnect risk against facility-dollar shares inside the Big-5 perimeter.

How this complements concentration, spend updates, and late-Aug geography

Pair this map with the Q3 concentration vintage when you need top-1 / top-3 / HHI among Amazon, Alphabet, Microsoft, Meta, and Oracle, and with the mid-Q3 spend update when you need the Jul→Q3 dollar path and house-by-house midpoint deltas. Geography does not restate company guidance; it asks whether Amazon’s top-1 company rank is better understood as a US facility-region story — which it partly is — and whether the ~$75B cumulative raise is better understood as a US raise-geography story (~60% of the lift) — which the meter strip shows explicitly.

The late-Aug geography companion answers the same map question on the hotter ~$858B / ~58% US tape. Read the two vintages together: absolute dollars stepped up ~$23B from mid-Q3 to late-Aug while the US regional share tightened by about 1 pp — a hotter stack that is also slightly more US-skewed, not a rewrite of the continental map. The intensity companion (Q3 intensity update) answers how hard companies are spending relative to revenue and cash flow. This lens answers where the physical dollars land on the mid-guide tape.

Caveats and what this map is not

• Regional and metro shares are desk reconstructions from disclosed campus / interconnect tips plus Synergy-style capacity geographythey are not company-reported geographic segment breakouts in 10-K footnotes.
• AI-slice dollars apply a flat ~75% CreditSights AI-attribution factor to gross regional weights; true AI intensity varies by company and campus.
• Scenario geography applies regional weights to GS GI and GS IR perimeters for map readability; those houses do not publish identical continental splits, and absolute stacked heights must not be mixed across scopes.
• Interconnect-risk scores are ordinal desk ranks (0–100), not ISO queue extracts or utility interconnection study results.
• Jul→Q3 raise geography (~60% US) attributes the cumulative guidance lift by facility tip, not by HQ domicileOracle’s net-of-prepay print and Microsoft’s lease reclass still sit inside the company perimeter from the Q3 spend update.
• Residual / other US buckets close the $835B and $476B perimeters; they are accounting closures, not claims about a specific metro.
• Vintage pairing: mid-Q3 (~$835B, US ~57%) versus late-Aug (~$858B, US ~58%) are adjacent desk stepsdo not average them into a single “2026 geography” without labeling the tape.

What to watch next

Watch three geography meters, not one. First, whether the US regional share holds near ~57% if Europe and APAC sovereign AI campus pipelines convert from announcement to spend faster than the 2027 path implies — or whether the late-Aug ~58% tip becomes the new mid-guide baseline. Second, whether Midwest and Texas growth rates continue to outpace NoVA even while NoVA remains the absolute-dollar tip — a signal that power availability is rewriting incremental geography. Third, whether GS GI-style all-in perimeters keep printing a softer US weight (~52%) than Big-5 facility maps as non-hyperscaler and power-side dollars grow. For company tip thickness keep the Q3 concentration series open; for absolute dollar vintages keep the mid-Q3 update open; for the hotter map keep the late-Aug geography companion open; for the multi-house scenario spine keep the research ledger open.

Related reading: Mid-Q3 spend update · Q3 concentration · Late-Aug geography · Research ledger · Q3 intensity.