Buildout patterns & signals

Name: Buildout patterns & signals | Data Center Watch
Creator: Data Center Watch

Commonalities across the site’s sourced data, and the directional predictions they support. Two lanes: Observed (primary-sourced facts) and Derived (arithmetic on those facts). Every derived figure shows its formula, the exact source numbers, and a defensibility tier — grounded (a ratio of two sourced figures) or directional (a trend to read as “at the recent observed pace”). Nothing is estimated to fill a gap.

Observed — the ERCOT large-load funnel

438.0 GWRequested (queue)

9.0 GWApproved to energize

3.9 GWObserved operating peak

ERCOT large-load interconnection queue, snapshot 2026-03-31. Source: ERCOT LLI.

ERCOT metered system peak91.1 GW

ERCOT firm dispatchable capacity (gas+coal+nuclear+oil+hydro)87.1 GW

ERCOT total nameplate capacity (all fuels)174.9 GW

Derived — what the funnel implies

2.1%

grounded

Approval rate — share of the requested queue cleared to energize

approved_to_energize ÷ requested = 9,042 ÷ 438,000 MW. The queue is a demand signal, not a delivery schedule.

Inputs: ERCOT LLI

43%

grounded

Realization rate — approved capacity that is actually drawing power

observed_peak ÷ approved_to_energize = 3,883 ÷ 9,042 MW.

Inputs: ERCOT LLI

0.89%

grounded

Realized-to-queue — real operating load vs the headline queue

observed_peak ÷ requested = 3,883 ÷ 438,000 MW. Under 1% of the headline queue is metered load today.

Inputs: ERCOT LLI

4.8×

grounded

“Paper gigawatts” multiple — requested queue vs the whole metered grid

requested_queue ÷ metered_system_peak = 438,000 ÷ 91,075 MW. The requested queue exceeds ERCOT's entire current peak by this multiple — full realization would require the grid to grow several-fold.

Inputs: ERCOT LLI · EIA-930

-3,967 MW

directional

Firm-capacity headroom before any new large load lands

firm_dispatchable − metered_peak = 87,108 − 91,075 MW. A thin cushion on the dispatchable fleet alone; intermittent nameplate is excluded. Point-in-time, not a forecast.

Inputs: EIA-860M · EIA-930

flat vs surging

directional

Approvals have stalled while requests surged

approved_to_energize sits at 9,042 MW at the latest snapshot with a value, while requested has climbed toward 438,000 MW — each incremental GW requested converts to a smaller share of real approval. Extrapolating headline queue growth to future load is unreliable.

Inputs: ERCOT LLI

Operators by attributed capacity

Sum of per-facility disclosed/attributed power (gross MW, falling back to IT MW) by operator. Disclosure-biased: only 636 of 27,647 tracked facilities state any power figure, and rows mix operational, announced, and third-party-estimated capacity — read as a coverage-weighted ranking, not an operational-MW census.

Operator	Facilities	Attributed capacity
Amazon	15	17.1 GW
Crusoe	3	12.5 GW
Softbank/Open AI (unconfirmed)	1	9.2 GW
TransGas Development, LLC	2	4.8 GW
Homer City Redevelopment (with Kiewit as contractor/partner)	1	4.5 GW
Meta	20	4.4 GW
Google	24	4.1 GW
Balico LLC	1	3.5 GW
Open AI	1	3.2 GW
Vermaland LLC	1	3.0 GW
Stak	1	3.0 GW
IREN	2	2.8 GW
Microsoft	14	2.5 GW
VALCO HANOVER COUNTY	1	2.4 GW
Hut 8	4	2.1 GW
Tract	2	2.1 GW
Rubix	1	2.0 GW
Poolside/CoreWeave	1	2.0 GW
Granite Renewables	1	2.0 GW
Eneus Energy	1	2.0 GW

Counties by tracked capacity

Where disclosed capacity concentrates — a directional “next-hotspot” read, same disclosure caveat as above.

County	Attributed capacity
Laramie, WY, United States	10.2 GW
Pike, OH, United States	9.2 GW
Box Elder, UT, United States	9.0 GW
Lea County, NM, United States	7.0 GW
Pecos, TX, United States	7.0 GW
Mingo, WV, United States	4.8 GW
Indiana, PA, United States	4.5 GW
Lackawanna, PA, United States	4.3 GW
Hanover, VA, United States	4.2 GW
Pittsylvania, VA, United States	3.8 GW
Maricopa, AZ, United States	3.8 GW
Effingham, GA, United States	3.2 GW
North Slope, AK, United States	3.1 GW
Lake, IN, United States	3.1 GW
Pinal, AZ, United States	3.0 GW

Derived — connectivity & measurement presence

Cross-referencing the facility registry against 1,065 RIPE Atlas anchors (measurement servers hosted inside data centres) locates where the buildout has confirmed, measured internet presence — and where it doesn't yet. Source: RIPE Atlas / RIPE NCC.

741 data centers

grounded

Data centers with confirmed measurement presence

943 of 1,065 RIPE anchors (88.5%) geo-match a tracked facility within 8 km (haversine spatial join). A hosted anchor is hard evidence of live, carrier-grade connectivity on-site — a signal a greenfield announcement can't fake.

Inputs: RIPE Atlas

Cross-source accuracy check: only 4 of the 943 matched anchors disagree on country between RIPE and our registry — independent corroboration that both geolocations are sound.

DNS root-server latency (global reachability)

DNS root	Probes	Median RTT (ms)	p10	p90
A-root	13,944	19.69	2.55	117.08
B-root	2,000	35.71	13.99	171.5
C-root	4,247	44.47	15.48	191.49
D-root	13,944	6.33	1.04	23.79
E-root	13,944	17.12	1.75	143.24
F-root	13,944	28.87	8.16	170.47
G-root	2,035	32.17	1.49	245.97
H-root	13,945	26.37	5.5	106.8
I-root	13,946	21.41	4.53	134.56
J-root	13,946	26.93	8.08	120.18
K-root	13,947	19.78	5.15	114.74
L-root	13,945	10.63	1.14	65.41
M-root	13,945	6.77	0.97	28.24

Networks hosting the most measurement infrastructure

Network (AS holder)	Anchors hosted
DigiCert	21
Vultr LLC	18
G-Core Labs S.A.	16
Console Connect	13
Misaka Network, Inc.	12
Tele2 Kazakhstan	12
DigitalOcean	8
NetActuate	7
OVHcloud	7
marbis GmbH	6

These are analytical readings, not investment advice or a delivery forecast. ERCOT is the deepest-covered region today; the same funnel math generalizes to PJM/MISO/CAISO and other ISOs as their queues are ingested (each with its own source). Derived figures update every build as the underlying sourced data refreshes.