Charted: Top-1 Holds 28% of Big-5 AI Capex — Top-3 Holds 74%

Absolute AI infrastructure dollar totals answer how large is the stack. They do not answer how concentrated is the system at the top of the distribution. That is the capital-markets question this post is built for.

Our theme already maps the perimeter: the July research spend map put Big-5 midpoints near $760B; the late-Aug 202608 update revises that stack to about $858B. Those posts are about levels and vintage deltas. This companion is about shares: top-1, top-3, HHI, cumulative ladders, and who captured the raises as the calendar-year guides climbed.

The interactive dashboard above is a concentration lens, not another KPI strip. Top-1 share is 28% (Amazon, ~$240B). Top-3 share is 74% (Amazon + Alphabet + Microsoft, ~$633B of ~$858B). Approximate Big-5 HHI is ~2,227 — elevated versus a five-way equal split (~2,000). Pair this distribution cut with the intensity / cash-flow frame for affordability, and with the absolute-dollar updates for the vintage path.

The concentration scoreboard

Read the table as a ladder. Absolute dollars rose nearly fourfold from 2024 (~$231B) to late-Aug 2026 (~$858B). Concentration did not equalise. Top-1 share drifted from the low thirties toward the high twenties, then re-concentrated as Amazon’s late-Aug raise reclaimed share. Top-3 stayed in a tight band around the mid-to-high seventies.

Top-1 and top-3: the company ladder

On the late-Aug 202608 vintage the ranked Big-5 ladder is:

1. Amazon — $240B (28.0%)
2. Alphabet — $218B (25.4%)
3. Microsoft — $175B accounting CY (20.4%)
4. Meta — $150B (17.5%)
5. Oracle — $75B net of prepayments (8.7%)

Cumulative share hits 53% after two names and 74% after three. That is the visual the Lorenz panel is built to show: the equal-split diagonal is a straight 20/40/60/80/100 path; the actual path bends hard toward the top-left.

Two definition notes matter immediately. Microsoft’s $175B is the accounting calendar print after lease reclassification — economic campus spend is still commonly modeled nearer ~$190B. Toggle the Microsoft control in the dashboard: under the economic frame, top-1 eases slightly to ~27.5%, top-3 edges to ~74.2%, and the Big-5 sum moves to ~$873B. Oracle’s $75B is net of customer prepayments; gross cash outlays remain nearer $95–100B. Neither caveat erases the concentration story. Both change the second decimal, not the shape of the ladder.

Alphabet’s second-place share is itself a concentration signal. Four guidance raises in a calendar year — from the low-$180Bs toward a $210–225B band with midpoint $218B — mean the number-two program is not a static runner-up. It is a rising second pole. Amazon and Alphabet together hold ~53% of the late-Aug stack before Microsoft is even counted.

Growth is more concentrated than the stock

Stock concentration (who holds today’s dollars) and raise concentration (who captured the Jul→Aug path) are related but not identical. From the July research print (~$760B) to late-Aug (~$858B), the Big-5 midpoint sum rose about +$98B / +12.9% across three refreshes.

Decompose the positive company raises and the picture skews further:

• Amazon +$40B
• Alphabet +$33B
• Oracle +$25B
• Meta +$15B
• Microsoft −$15B (headline onlylease reclass)

Amazon + Alphabet alone account for roughly 65% of positive dollar raises. The donut panel makes that explicit: incremental authorizations are more oligopolistic than the already-concentrated stock. For equity and credit desks that treat “the AI capex cycle” as a diversified industry factor, that raise skew is the practical rebuttal. Factor exposure is not five equal names — it is two programs doing most of the upward rewriting, with Microsoft’s accounting print masking a still-large economic build.

Multi-year path: dollars up, equalisation incomplete

Track top-1 and top-3 across vintages:

Two readings sit in that table. First, as more names authorized mega-programs (Oracle’s step-up; Meta’s band; Alphabet’s serial raises), top-1 share briefly eased even as top-1 dollars exploded — Amazon went from $75B (2024) to $240B (late-Aug) while its percentage of a much larger pie fell then recovered. Second, top-3 never left the mid-to-high seventies. The system did not become a broad industrial spend base; it became a larger oligopoly.

HHI on five Big-5 buckets stays above the equal-split benchmark of 2,000. That is a useful cross-post ranking tool, not a reconstructed global IT-capex census. The dual-axis panel — stack size rising while top-1 share stays sticky — is the capital-markets takeaway: scale and concentration can rise together.

Perimeters: when shares do not travel

Company shares only travel inside a declared perimeter. The dashboard’s perimeter bars put late-Aug Big-5 (~$858B) next to Street (~$845B), CreditSights (~$850B), GS Global Institute all-in AI infra (~$765B), and GS Investment Research hyperscaler (~$815B). Those houses are answering related but non-identical questions.

• Big-5 / Street / Credit are roughly company-gross hyperscaler worldsshare math is meaningful.
• GS GI is an all-in AI infrastructure construct (compute + data center + power). You cannot allocate that total into Amazon/Alphabet/Microsoft shares without inventing a mapping the institute did not publish.
• GS IR is a hyperscaler research path that can sit above or beside GI depending on year and bull/baseanother reminder that “AI capex” is a family of stacks, not one ticker.

Apply the CreditSights-style ~75% AI-attributed haircut and the AI-specific slice of the late-Aug Big-5 stack is about $644B. Toggle it in the ladder view so both perimeters stay visible. Treat the haircut as a convention for comparing AI-heavy programs, not as a disclosed segment line.

What would change the concentration story

Several shifts would rewrite this scoreboard:

1. A sustained non-Amazon top-1Alphabet or Microsoft clearing Amazon on a like-for-like gross (or economic) basis for consecutive vintages.
2. Top-3 share falling through the low-sixties while the Big-5 total still growsevidence that Meta and Oracle (or a sixth hyperscaler-scale program) are absorbing incremental dollars faster than the top three.
3. HHI drifting toward ~2,000 and staying there as the stack scalesequalisation, not just temporary raise noise.
4. Raise concentration cooling so Amazon+Alphabet no longer capture ~two-thirds of positive vintage deltas.
5. Meaningful spend outside the Big-5 perimeter that makes “Big-5 share of AI infra” itself a falling metricsovereign, neocloud, and enterprise build-outs large enough to move the industry distribution, not just the footnotes.

None of those are guaranteed by another earnings-week raise. They are competitive and financing outcomes. Until they arrive, desks should price AI infrastructure risk as concentrated program risk first and diversified sector beta second.

Caveats and reading notes

• Perimeter is Big-5 hyperscaler gross midpoints (with Oracle net-of-prepay as disclosed). This is not a full global AI-infrastructure census and excludes large non-hyperscaler and sovereign programs.
• Microsoft lease reclassification moves dollars off the capex line without canceling servers or campusesuse the accounting/economic toggle rather than treating the headline cut as a build pause.
• Oracle net vs gross and Meta finance-lease principal affect levels; they do not invert the top-1 / top-3 ranking on current vintages.
• HHI is computed on five company buckets inside the Big-5 setuseful for ranking concentration across theme posts, not a plant-level market definition.
• AI-attributed ~75% is a CreditSights-style convention applied uniformly; company-level AI mixes differ and are not fully disclosed.
• Vintage midpoints revise after every call and Street catch-updirectionally the concentration ladder has been stable even when point estimates move $10–20B per name.
• Figures are USD billions; roundings apply. Do not mix GS GI all-in totals with company share percentages in the same sentence.

Bottom line

The AI capex system is a concentration system. On the late-Aug Big-5 stack of ~$858B, top-1 holds 28%, top-3 holds 74%, and HHI sits near 2,227. The Jul→Aug raise path is even more skewed: Amazon and Alphabet captured about two-thirds of positive dollar raises while Microsoft’s headline slipped on lease accounting. Absolute dollars answered how large the cycle is. Shares answer who is writing it — and for now, that answer remains a short list.

Related reading: late-Aug spend update, July spend research map, and hyperscaler capex intensity.