Update: Hyperscale Pipeline Hits 915 Sites — Texas Ops +71% YoY, Top-3 Cloud Slips to 57%
Versus our Q3 site-ledger print (803 pipeline / Top-3 58%), Synergy’s Aug 19 location rankings raise the known hyperscale pipeline to 915 (+112), put top-20 markets at 60% of capacity, and show Texas ops growing +71% YoY vs +36% world. Big-5 ownership shares still carried.
Loading interactive charts…
What changed since the Q3 site & power print
In the Q3 compute update we promoted Synergy’s hyperscale site-count ledger and Gartner’s US electricity composition: 437 of 803 pipeline sites in the US, Top-3 cloud at 58% of hyperscale capacity, and dedicated AI already about one-third of US data-center power. That post answered whether geography got louder while Big-5 ownership stayed frozen. This August refresh answers the next official vintage the calendar forced: what moved once Synergy restated the world’s leading hyperscale locations on 19 August 2026 — pipeline count, market concentration, and named-market growth?
Three meters rewrite the scoreboard. Known hyperscale pipeline facilities rise from 803 to 915 (+112). Top-3 cloud share of hyperscale capacity slips 58% → 57% (−1 pp). And the location map gets a hard concentration print: just twenty state or metro markets hold 60% of world hyperscale capacity, with Northern Virginia and Greater Beijing alone at 17%, while Texas operational hyperscale capacity grew +71% YoY against a +36% worldwide average. Ownership Big-5 shares remain carried near 71% — Epoch’s Q1/Q2 2026 period print is still open — so the vintage delta is where capacity sits and how fast named markets grow, not a new chip-owner pie.
The dashboard above is built as a Q3→Aug 19 vintage delta: top-20 seat stack, pipeline bars, named-market capacity hints, Texas-vs-world growth, geography concentration deltas, entered/exited rank churn, ownership carry, and tokens-versus-ownership scatter. Use the owner, region, and churn controls to isolate who still owns the silicon versus which markets entered or left the global top-20.
The headline table: Q3 site ledger → Aug 19 location rankings
| Metric | Prior (Q3 2026 update) | New (Synergy 19 Aug 2026) | Δ |
|---|---|---|---|
| World hyperscale pipeline sites | 803 | 915 | +112 |
| Top-3 cloud share of hyperscale | 58% | 57% | −1 pp |
| Top-20 markets capacity share | site counts emphasized | 60% | New concentration print |
| N. Virginia + Greater Beijing | — | 17% | Two metros dominate stock |
| Next-20 markets capacity share | — | 19% | Top-40 ≈ 79% combined |
| US seats in global top-20 | implied rising | 15 of 20 | Non-US seats 5 (was 6 YoY) |
| Texas ops hyperscale growth | Inland corridor narrative | +71% YoY | World avg +36% |
| Big-5 AI chip ownership | ~71.4% carried | ~71.4% carried | 0 pp (Epoch open) |
| US large-DC IT pipeline (GW) | ~45 GW carried | ~45 GW carried | Unchanged Jul print |
That +112 site / −1 pp Top-3 / +71% Texas YoY triad is the clean vintage delta versus our Q3 update. The ownership column is intentionally flat: this release is a location rankings vintage, not an Epoch Chip Owners restatement. For the explorer ownership path see the August update; for the baseline ownership essay see July research.
Pipeline 803 → 915: the site meter keeps climbing
Synergy’s known pipeline of future hyperscale data centers now stands at 915 facilities at various stages of planning, development, or fit-out — up +112 from the 803 hyperscale-only pipeline we carried in the Q3 post. Forecasts lean on that tracked pipeline, so the jump is not a cosmetic revision: it is additional capacity already inside the firm’s visibility cone.
Two definitional reminders still apply. First, the July ~1,500 large-DC worldwide pipeline (almost half in the US, ~45 GW US IT load) is a broader perimeter than Synergy’s hyperscale-operator ledger — do not narrate 915 as a collapse from 1,500. Second, site count is not capacity: new campuses remain larger than the operational average, so each incremental pipeline site can punch above a 2024-era facility. The dashboard’s pipeline panel keeps Q3 and Aug 19 side-by-side so the +112 delta is the object, not a re-baselined stock.
Top-20 markets = 60%: concentration gets a hard print
The Aug 19 release’s sharpest new meter is market concentration. Twenty state or metro markets account for 60% of current world hyperscale capacity. Northern Virginia and Greater Beijing alone make up 17%. The next twenty markets add another 19%, so the top forty already cover roughly four-fifths of the stock. That is the geographic twin of the ownership story we have told all year: a thin set of places and a thin set of firms process most of the world’s AI-ready watts.
Composition of the top-20 seats tilted further toward the United States. Fifteen of the twenty largest markets are now US, with four in APAC and just one in Europe (Dublin). Non-US seats in the top-20 fell to five, from six a year ago and seven two years ago. Tokyo, Sydney, and South Carolina dropped out; Indiana, Tennessee, and Guangdong entered. Power availability and local opposition — the same frictions Synergy flagged when Dublin, Amsterdam, and Singapore ranked down — are rewriting who can still host the next 100 MW campus.
Texas +71% YoY: the growth premium is inland and US-heavy
Texas is Synergy’s loudest named growth print: operational hyperscale capacity in the state grew +71% over the last year, against a +36% worldwide average — a +35 pp premium. Other high-growth markets named in the release include Indiana, Tennessee, and South Virginia in the US, plus Shanghai, Johor (Malaysia), and Jakarta. Northern Virginia remains the densest live hub with Beijing, but Synergy notes it is no longer featuring as highly in new plans as it once did — the same inland shift we tracked when Texas + Midwest jumped from 33% of operational US hyperscale capacity to 53% of the pipeline in the July/August vintage.
That growth map matters for the “who processes / where located” question. Chip ownership is still a Big-5 US story (~71%). Location of new capacity is increasingly a US inland + selective APAC tier-two story. Europe’s sole remaining top-20 seat (Dublin) is a warning light, not a counterweight. Pair the location panels with our global AI data-center build tracker for campus-level context and with the brand token series for the usage side that still does not match ownership.
Top-3 cloud slips to 57% — mild deconcentration, not a regime change
Amazon, Microsoft, and Google still have the broadest footprints and now account for 57% of all hyperscale data-center capacity — one percentage point below the 58% we carried in the Q3 post. They are followed by Meta, Alibaba, Tencent, Oracle, Apple, ByteDance, CoreWeave, and smaller hyperscale operators. Synergy also notes that 62% of hyperscale operators are headquartered in the US, including the four biggest, and that the US accounts for almost half of cloud revenues in several key segments — structural reasons the top-20 seat map keeps US-heavy even when individual metros churn.
A −1 pp Top-3 move is not a breakup of concentration. It is consistent with neoclouds and second-tier hyperscalers adding sites inside a still-tight market. Toggle owners in the dashboard: the Epoch-carried Big-5 chip shares stay flat while Synergy’s facility Top-3 eases a hair — two related but non-identical concentration meters.
Ownership and tokens: still carried, still misaligned
Epoch’s finalized aggregate still anchors on Q4 2025: five hyperscalers own ~71% of global AI compute in H100-equivalents, Google alone near 25%, China as a whole near 5%. We carry those shares again because the Aug 19 Synergy print does not restate chip ownership. The tokens-versus-ownership scatter remains the reminder that usage ≠ silicon: OpenAI and ByteDance can clear large token shares with little or no Big-5-style chip ownership, while Microsoft and Amazon own more than they first-party.
Until Epoch closes a Q1/Q2 2026 period print, treating ownership as the moving vintage would invent a delta. Location rankings did move — pipeline, top-20 composition, Texas growth, rank churn — so that is what this post prices.
Caveats and methodology
- Perimeter mismatch. Synergy’s 915 hyperscale pipeline ≠ the July ~1,500 large-DC worldwide pipeline (colo + enterprise + hyperscale). Compare like with like.
- Site count ≠ MW. New campuses skew larger; a +112 site jump can imply more than +14% capacity if average IT load per site rises.
- Named-market capacity hints in the dashboard are illustrative placements inside disclosed concentration bands (top-2 / top-6 / top-20 / high-growth). Synergy’s public release does not publish a full market-by-market MW table.
- Ownership shares are carried from the Aug explorer / Q3 post. Epoch Q1/Q2 2026 aggregates were not finalized as a period delta at analysis time.
- Top-3 57% is hyperscale facility capacity share (Synergy), not Epoch H100e chip ownership. Do not conflate the two pies.
- Texas +71% is operational hyperscale capacity growth for the state, not a single metro’s IT-MW disclosure.
- Gartner US electricity composition (US ~204 of 565 TWh; ~68 TWh dedicated AI) is carried from the Q3 postno new Gartner vintage in this refresh.
- Rank churn (Indiana / Tennessee / Guangdong in; Tokyo / Sydney / South Carolina out) is as disclosed for the top-20 set; deeper rank movements below the top-20 are out of sample.
Primary sources: Synergy Research hyperscale location rankings (19 Aug 2026); Synergy inland hyperscale investment analysis and Jul 2026 US capacity update (~45 GW carried); Epoch AI Chip Owners / hyperscaler share chart (Q4 2025 anchors, carried); Gartner Forecast: Data Center Power Capacity and Consumption, Worldwide, 2024–2030 (1Q26 / Jun 2026 public release, carried); prior theme posts Q3 site update and Aug explorer update.
The shareable takeaway
Versus Q3, Big-5 chip ownership is unchanged near 71% — the Aug 19 vintage is location rankings and growth. Synergy’s known hyperscale pipeline rises 803 → 915 (+112), Top-3 cloud slips 58 → 57%, top-20 markets lock in 60% of capacity (N. Virginia + Beijing 17%), and Texas ops grow +71% YoY against a +36% world average while US seats fill 15 of 20 largest markets. For the Q3 site-count and power-composition print see the Q3 update; for ownership restatement see the August explorer update; for the baseline essay see July research; for campuses see the build tracker; for tokens see the brand series.