A screening tool is only worth the data behind it. Here is exactly what we pull live per coordinate, what is regional, how the call is made, and where it's reliable.
Every decisive figure in a verdict is a clickable evidence object. Open it to see the publisher, the date, an honesty label, a direct link to the original source, and a short "how to verify" note telling you exactly where to find the cited figure there. Live measured signals open the map or calculator at your coordinate so you can reproduce the number yourself; article and filing pages also carry an archived copy in case the page moves. We store facts - never a copy of the source - so each citation stays inside fair-use and public-domain reuse.
Three honesty labels, shown on every line:
Measured - pulled live from the source this run (NREL solar, OSM gas/substation distance).
Modeled - our screening estimate or synthesis of filings; planning-grade, ±40%.
Claimed - stated by a third party; we cite it, we don't independently assert it.
Source-trust tiers, highest to lowest. When sources conflict we show both and take the higher tier; newer always wins over older.
| Tier | What qualifies | Re-review |
|---|---|---|
| Live | Measured live this run - NREL PVWatts, OpenStreetMap / Overpass, Nominatim | Every screening |
| Tier 1 | Government / regulator / ISO-RTO / national lab - NREL, EIA, FERC, state PUCs, LBNL, PJM / ERCOT / MISO queues | Quarterly |
| Tier 2 | Primary company disclosure - SEC filings, official investor-relations / press releases | Quarterly |
| Tier 3 | Established trade press - Data Center Dynamics, Datacenter Frontier, Utility Dive, Reuters | Monthly |
| Tier 4 | General news | Monthly |
| Tier 5 | Blog / forum / social - used only as a lead, never as the basis for a verdict | Monthly |
Sources older than their cadence are flagged for re-review; conflicts are surfaced, not hidden. This page is also designed to be machine-readable so AI assistants can cite it accurately.
When you screen a site, three pulls run against public APIs for that exact point - no canned answers:
| Signal | Live source | How it drives the verdict |
|---|---|---|
| Solar capacity factor | NREL PVWatts v8 | Sizes and firms the solar / solar-+-BESS pathways for that latitude. |
| Gas pipeline distance | OpenStreetMap / Overpass | Nearest gas main. On-site (≤5 mi) → fast gas; 5-18 mi → lateral + months; none in range → gas paths ruled out. |
| Substation / plant distance | OpenStreetMap / Overpass | Grid-edge and brownfield proximity for interconnect and repower potential. |
| Place & state | Nominatim (OSM) | Maps the coordinate to its ISO/market for queue and PPA context. |
Two inputs are sourced regionally and clearly labeled as estimates in every result. These are the next data layers to make per-coordinate.
| Input | Source |
|---|---|
| Interconnection-queue year | LBNL "Queued Up", ISO/RTO public queues (PJM, MISO, ERCOT, CAISO) |
| Nuclear-PPA depth | SEC EDGAR filings - Constellation, Vistra, Talen |
| Tech lead time / cost | NREL ATB, Lazard LCOE, GE Vernova / Wärtsilä disclosures |
LBNL covers generation queues - a proxy. For the markets where our users actually build, we add a curated large-LOAD connect timeline: the real wait for new data-center load, hand-built from utility / ISO / PUC disclosures and refreshed as they change. Where a coordinate falls in a covered county, the grid timing uses this load estimate (labeled with the utility) instead of the generation proxy.
| Market | Utility / ISO · what drives the wait |
|---|---|
| Northern Virginia | Dominion - most congested DC market; queue cited up to ~15 yr |
| Dallas-Fort Worth | ERCOT / Oncor - SB6 (2025) large-load study + post-2025 curtailment |
| Columbus, OH | AEP Ohio - 2023 moratorium lifted; 2025 data-center tariff + study |
| Atlanta | Georgia Power - >100MW tariff, 15-yr contract + PSC review |
| Chicago | ComEd / PJM - ~8-yr app→COD; ~160-wk transformer lead |
| Quincy, WA | Grant County PUD - load caps until transmission expansion (7-10 yr) |
| Phoenix | APS / SRP - subscription model; new gen 2031-33; water-constrained |
| Abilene, TX | ERCOT / AEP Texas - ERCOT West; SB6 large-load study; transmission-gated (Stargate cluster) |
| Central Louisiana | Entergy (MISO South) - utility-built gen 2028-29 (Franklin Farms 1.5GW) + $1.2bn 500kV line; LPSC “Lightning” fast-track |
| Memphis, TN | MLGW / TVA - TVA board approval + substation-gated; on-site gas bridged xAI Colossus |
Curated, screening-grade, as of 2026. Outside these counties, grid timing falls back to the LBNL/ISO generation-queue proxy.
The amber markers on the map are real campuses that run on on-site or behind-the-meter generation - the exact strategy this tool screens for. Every figure below is sourced to a primary or industry record. If one is wrong, tell us and we'll correct it.
| Campus · operator | Location | On-site power (verified) | Source |
|---|---|---|---|
| Stargate Site 1 · Crusoe (OpenAI / Oracle) | Abilene, TX · ERCOT | On-site natural-gas turbines (captive): 360 MW Unit 1 (online 2026) scaling to ~1.4 GW Unit 2, +191 MW diesel backup; campus 1.2 GW IT | GEM / TCEQ permits |
| xAI Colossus · xAI | Memphis, TN · TVA / MLGW | On-site natural-gas turbines, behind-the-meter: ~420 MW deployed (≈35 units); 247 MW (15 units) air-permitted in 2025 | CNBC |
| Homer City Energy Campus · Homer City Redevelopment / Kiewit | Homer City, PA · PJM | On-site natural gas: 7× GE Vernova 7HA turbines, up to ~4.5 GW dedicated; built on a retired coal-plant site; power ~2027 | DCD |
| Project Kati · Soluna | Willacy County, TX · ERCOT | Behind-the-meter wind: up to 166 MW under a PPA with the co-located 273 MW Las Majadas wind farm; Phase 1 (83 MW) ~2026 | POWER |
| Scrubgrass · Stronghold Digital Mining | Venango County, PA · PJM | On-site coal-refuse (waste-coal) generation, behind-the-meter: 83.5 MW Scrubgrass + 80 MW Panther Creek ≈ 165 MW owned, with co-located compute | Stronghold IR |
| HyperGrid (Project Matador) · Fermi America | Amarillo, TX · private grid | Behind-the-meter "HyperGrid": on-site nuclear (Westinghouse AP1000, 4 GW planned) + natural gas (600+ MW GE/Siemens turbines acquired; first 6 GW gas state-approved) + solar. 11 GW campus, first ~1 GW targeted late 2026. | DCD |
| WY-1 · Prometheus Hyperscale | Evanston, WY · off-grid | Behind-the-meter natural gas (PureWest low-carbon supply) + carbon capture; 1.2 GW campus, Phase 1 ~150 MW (~2026-27). | DCD |
Verified 2026 from the cited sources. Markers locate each campus for discovery; exact coordinates may be approximate. Grid-connected campuses are intentionally excluded - e.g. Hut 8 River Bend (Entergy grid), TeraWulf Lake Mariner (reused grid transmission), and AWS-Talen Susquehanna (shifted to front-of-the-meter after FERC's 2024 rejection of its behind-the-meter pact).
You give a coordinate, a firm-power need, and a timeline. We pull the live signals above, then score ten power pathways - four single-tech baselines, three hybrids, three nuclear PPAs - on the same yardstick: time-to-power, firm capacity, and rough cost. A pathway "meets" if it arrives on time at ≥90% firm capacity. If at least one does, the site is ADVANCE; a near-miss is CAUTION; none is KILL.
Good at: killing obviously-infeasible parcels fast, ranking candidate sites on a common yardstick, and giving you a defensible first-pass story before you spend on engineering.
Not good at: bankable numbers, exact interconnection study outcomes, site-specific permitting and water rights, or final OEM pricing. Anything past the screen needs a real feasibility study.
We log every material correction publicly. Accuracy is the product; the track record of being right is the moat.
| Date | Correction |
|---|---|
| 2026-Q2 | Moved the verdict from a regional bucket to live per-coordinate pulls (NREL solar + OSM gas/substation distance). |
| 2026-Q2 | Mapped every U.S. state to its ISO/RTO for queue and nuclear-PPA depth - any coordinate now resolves to a real market instead of a generic default. |
| 2026-Q2 | Added a curated large-LOAD connect layer for 7 hot zones (NoVA, DFW, Columbus, Atlanta, Chicago, Quincy WA, Phoenix); grid timing there reflects real data-center load waits, not the generation proxy. |
| 2026-Q2 | Expanded the curated large-LOAD layer to 10 hot zones - added Abilene TX (ERCOT / AEP Texas), Central Louisiana (Entergy), and Memphis TN (MLGW / TVA). Each verdict now shows the on-site vs grid path with its source and a confidence rating. |
| 2026-Q2 | Gas pathways now ruled out automatically when no pipeline is found within ~7 mi of the coordinate. |
| 2026-Q2 | Raised simple-cycle gas turbine lead time to ~28 months to reflect OEM backlog. |
| 2026-Q2 | Marked solar-only / solar+BESS as non-firm by default; they cannot meet a 24/7 data-center load alone. |
| 2026-Q2 | Plotted real on-site / behind-the-meter powered data centers on the map - now seven (Crusoe/Stargate Abilene, xAI Colossus, Homer City, Soluna Project Kati, Stronghold Scrubgrass, Fermi HyperGrid Amarillo, Prometheus WY-1 Evanston), each with a primary source. Verified and excluded grid-connected campuses (Hut 8 River Bend on Entergy, TeraWulf Lake Mariner, AWS-Talen Susquehanna after its FERC-driven shift to front-of-the-meter). |
| 2026-Q2 | Added a ground-truth layer: screening a coordinate that lands on one of these verified campuses now returns the real, sourced on-site solution (a "Proven" result) instead of a blind public-data verdict - public maps can't see private on-site generation. |
| 2026-Q2 | Shipped visible provenance: every decisive number in a verdict is now a clickable evidence object with a source-trust tier (1-5), an honesty label (measured / modeled / claimed), publication and "as-of" dates, a deep link that highlights the cited line, and an archived snapshot. Added the source-tier rubric above and a scheduled re-review cadence per tier. |