Mexico: Energy Infrastructure & Industrial Capabilities

Current Data (refreshed September 7, 2026)

• SENER PRODESEN 2024 State-level installed capacity & clean share from Cuadros A1.8.1–A1.8.8 (Dec 2023 actuals). National total: 88.5 GW, 36.7% clean. Replaces WRI 2016-17 values. Per-fuel state mix is now also PRODESEN Dec 2023 (Cuadros A1.8.1–A1.8.7 per-tech transcription, June 2026).
• Climate TRACE v7 (2025) 166 fossil + biomass facilities, plant-level carbon intensity (g CO₂e/kWh). 2025 total: 140 Mt CO₂e. Plant-emissions layer: circles sized by 2025 emissions, colored by primary fuel. Drives the carbon-intensity choropleth (state aggregation). Does not include zero-emission facilities (solar/wind/hydro/nuclear/geothermal).
• WRI GPPD v1.3 (legacy) 277 plants, 62,366 MW (2016-17 vintage). Kept as the "Power Plants" layer for continuity and its per-fuel breakdown. Tagged status: legacy. Use Climate TRACE 2025 for carbon analysis.
• CENACE PML (MDA) Observed nodal congestion, 2,603 NodosP × 12 months (2025-06–2026-05) + 222 corridor nodes to 2016. Drives the Grid Constraints layer: municipality fill = node aggregate; ranked cells = zona de carga × voltage tier (never state means; see the methodology page). Yucatán peninsula tops the national ranking but sits outside the four corridors.
• CNE permit registry (Feb 2026) Generation permits per state & technology (CC-BY). Powers the new capacity-additions field: 2.4 GW entered operation since Jan 2024, 19.0 GW under construction. Catches post-2023 momentum the plan-grade sources cannot see yet.
• PLADESE 2025-2039 National anchor at Dec 2024: 90.5 GW, 36.9% clean (Anexo Estadístico T 2.1). No per-state tables (state profiles remain the official PRODESEN Dec-2023 set).
• OSM Transmission lines (2,680 segments, 69kV+), Substations (2,487, all with voltage), Generators (41,797 units, 38k solar). No line capacity (MW thermal), no substation MVA, no MW per generator unit. Coverage is sparse in remote regions (Sinaloa interior, Durango, southern Chihuahua, Sierra Madre Occidental show 5-12× lower line density than the Bajío / Northeast corridors). Northern border states better mapped than south. CFE-grade authoritative grid topology not yet integrated.
• GridFinder 69,627 predicted line segments (satellite nighttime-lights + OSM road priors, Arderne et al. 2020, DOI 10.5281/zenodo.3628142, CC-BY 4.0). Off by default: toggle "Predicted Mesh" in the Grid tab. Fills OSM coverage gaps in Sinaloa interior, Durango, Sierra Madre Occidental. Predicted, not surveyed (statistical accuracy ~70-85% precision). Use as complementary backdrop to OSM transmission, not authoritative network. No voltage attribution.
• Global Solar Atlas v2.0 + Global Wind Atlas v3 + CONANP (2026-08-05) Feeds the Resource Beyond the Grid layer: usable solar/wind land classified by distance to the transmission network, on a 500 m equal-area grid. Solar = GSA PVOUT ≥4.5 kWh/kWp/day (Solargis/ESMAP/World Bank, CC BY 4.0; 2019 country package, static since then and older than the GSA live point service; PVOUT is a multi-decade average so this is current, but it is labelled by release year). Wind = GWA capacity factor ≥0.30, IEC3 turbine at 100 m hub (DTU/World Bank, CC BY 4.0). Exclusion mask = 232 CONANP federal protected areas (228,073 km² terrestrial, Libre Uso MX), chosen over WDPA, whose terms restrict commercial use and derivative works. Distance bands replicate AZEL 2023 Scenarios 1 / 3-4 (solar ≤2 km, wind ≤10 km / >20 km). Publishes area (km²), never a technical-potential MW total because that figure is set by an assumed packing density and exclusion set. Solar is not a discriminator in Mexico (76.7% of land clears the threshold, median PVOUT 5.02); wind is (9.1%). Median distance to the RNT is 28.7 km, and 61% of bankable wind / 63% of usable solar sit beyond 20 km. Slope, land cover, urban footprint, water and land tenure are not excluded. Verified four ways (thresholds, endpoints vs CENACE 400/230 kV cells, mid-spans via graph decomposition, and raster coverage), all shipped in the feed’s sensitivity block.
• NREL land use (2026-08-05) Packing densities for the per-park supply sheds on the Industrial Parks layer: utility-scale solar 28 MW/km² (8.9 acres/MWac total footprint), onshore wind 3.0 ± 1.7 MW/km² (project area). Printed at point of use so the reader can redo the arithmetic. The shed answers “could this site power itself” (the private-wire case), not “is clean power available here”: on a synchronous grid a PPA sources from anywhere. Land only; a 24/7 load also needs storage, which is Load Scenarios’ question.
• BNEF Big Green Machine + DataMexico + GripPoint greenfield compilation Investment pipeline: 37 BNEF green-manufacturing projects plus announced FDI / greenfield projects compiled from DataMexico and public announcements. Each greenfield record carries a verification tier (verified / attributed / announced). Investment commitments, not current operating facilities.
• DENUE via DataMexico State x National Industry employment (SCIAN 6-digit). Multi-month aggregate. Mapped to plays via SCIAN 3-digit concordance. SCIAN subsectors are broader than HS6 play products.
• Historical PRI v3.2 (May 2026) This map retains the archived 96 state-play classifications and their original model. They are historical screening results, not current investment recommendations. The Estado de México employment join was incorrect in that model; its weakest-component diagnosis is withheld here. Current LQ observations are shown separately. See the corrected v3.3 research in Methods; these archived scores have not been silently recomputed.
• Relatedness (since June 2026) International product-space proximity (Hidalgo et al. 2007): phi from cross-country co-export, CEPII BACI HS22 pooled 2022-24, 174 exporters; density normalized by expected coverage (state baskets: DataMexico 2024-25, $1M line floor; municipal baskets: HS4, thin, so directional). Replaces the HS-prefix heuristic; rankings shifted materially (Spearman 0.63 / 0.46 / 0.03 vs the old method per play; grid-hardware is essentially uncorrelated with it).
• Five value chains (August 2026) The map covers Auto Supplier, EV Components, Grid Hardware, Heat Pumps and Electric Vehicles. The first three came from the original selection; the latter two were added for coverage. Coverage does not establish a current rank. Consult Product Space for basket validation and scoring status. Heat Pumps carries every layer. Electric Vehicles appears on capability layers (relatedness and DENUE employment) and named plant records. Its subnational trade observations are unavailable because the state and municipal sources do not cover its HS 2022 vehicle codes; the map does not convert that absence to zero. Park readiness retains its published coverage of the original three chains plus Heat Pumps and is a separate model.
• IMSS Jul 2026 + SAT ITG 2024 + state law 2026 Feed the Payroll Tax (ISN) layer and the play cards’ fiscal-footprint block. Two layers with two render rules: the direct layer (IMSS administrative records: registered formal jobs and wage mass, municipality grain, zero suppression) may render spatially; the indirect layer (INEGI MIP multipliers, national, base year 2018 because that is the latest benchmark, published every five years) never does. Every ISN figure is a ceiling on what the state could collect: all 32 state rates were read from primary law and adversarially verified, and 30 of the 32 states can reduce or waive the tax for new investment, 17 of them case by case, so the gap to real collection is unobservable from any published source. The IMSS classification describes the employer’s registered activity and says nothing about the site’s output, and its 2005 catalogue cannot isolate the Heat Pumps or Vehículos Eléctricos chains, so both render as labelled proxy baskets. Fiscal return never enters the Play Readiness Index, the parks composite, or the selector’s scoring.
• CONEVAL 2020 + INEGI ENOE 2024 Feeds the Employment Need layer: municipal poverty rate and Gini (CONEVAL) against state informality (ENOE), crossed with where the five chains actually hold a position. A classification. It carries no score. The components are published with one transparent screening rule (top-tercile municipal poverty AND top-tercile state informality), declared a first pass. There is deliberately no weighted need score: blending need into a readiness number would corrupt both, and need never enters the Play Readiness Index, the parks composite, or the scoring in the play selector. 2020 is the latest municipal poverty measurement that exists, so there is no newer municipal wave to pull: the CONEVAL municipal series runs 2010 / 2015 / 2020 and the 2022 measurement is state-grain only. Informality stays at state grain because that is the grain it is measured at; no municipal informality is synthesised. The published rate is informal / (formal + informal), which reproduces the INEGI national figure of about 54%. The finding: 729 municipalities clear the screen, holding 9.5M people, and none of them holds a position in any of the five chains (published with its sensitivity, since a zero could be an artifact and this one is not: the join is demonstrably live, and high-need municipalities are absent from the municipal trade record at any level, not merely below the advantage threshold).
• Historical sensitivity checks (June 2026): These checks varied model weights; they did not estimate measurement uncertainty or predictive accuracy. Old selection ranks and near-tie claims must not be applied after basket or normalization changes. Current selection sensitivity and validation status are published in Product Space. Park-tier sensitivity belongs to the separate park model; DENUE shrinkage sensitivity does not establish that employment observations are free of sampling or coverage error.
• Data vintage cheat-sheet: Capacity & clean share = PRODESEN 2024 (Dec 2023) · Carbon intensity = Climate TRACE 2025 · Per-state fuel mix = PRODESEN Dec 2023 · Additions = CNE Feb 2026 · Congestion = CENACE MDA 2025-06..2026-05 · Grid topology = OSM (current) · National trade = BACI 2024; state trade/RCA = OEC / DataMexico 2025 (September 8, 2026 revision); municipality trade = DataMexico trade_mun 2023-24. · Solar resource = Global Solar Atlas v2.0 (2019 release, multi-decade average) · Wind resource = Global Wind Atlas v3 · Protected areas = CONANP 2026 · Land-use densities = NREL.

Required for Level 2-3 (source)

• CENACE (cenace.gob.mx): Hourly nodal dispatch, congestion pricing, line flow limits. Required to identify actual grid bottlenecks (vs. inferred from capacity). Public after all: REST API + CSV archives, MDA/MTR nodal prices incl. congestion component, history from 2016. Queued for S2.
• CRE/CNE: Bilateral PPA terms (generation permits integrated June 2026; see CNE registry above). Required for off-grid industrial power analysis.
• CFE / PRODESEN transmission plans: Substation MVA ratings, transformer loading %, 15-year grid expansion plan. Critical for Level 3 scenario validation.
• ERA5 via CDS API (Copernicus): gridded hourly irradiance + wind for capacity-expansion optimization. Closed for the modelled sites 2026-08-05 because long-term averages ship via Global Solar Atlas + Global Wind Atlas, and hourly series now ship via NASA POWER (CC BY 4.0, no key) behind the Load Scenarios clean-build sizing. What remains open is gridded hourly coverage: POWER is a point API at ~55 km, so it sizes a build at a named site but cannot yet drive a national capacity-expansion optimization.
• DENUE / INEGI: Geocoded establishment census. Would provide current operating facilities by SCIAN code. Public API, municipality-level aggregation recommended.

Industrial Parks Layer

• AMPIP via DataMexico 431 industrial parks from the Mexican Association of Industrial Parks registry. Geocoding is municipality-grade for most parks. Only 50 of 431 have a coordinate no other park shares; the rest resolve to one point per municipality, so 431 parks occupy ~160 distinct points. Each park carries geocode_precision, and a 10 km supply shed drawn on a municipality-precision point is a municipality answer. An audit (2026-08-05) also relocated 7 parks whose geocode had landed in a different city, one of which changed readiness tier.
• Regional-context and proximity score = 0.35 × grid proximity + 0.30 × clean energy share + 0.35 × play relevance. Grid proximity: Haversine distance to nearest 230kV+ substation (inverse decay, 200km max). Clean energy: state-level clean share (PRODESEN 2024). Play relevance: state RCA + municipality RCA bonus + DENUE LQ bonus.
• Tiers: High (≥0.65, green), Moderate (0.40-0.64, amber), Low (<0.40, red). Per-play scores available via play selector.
• Limitation: AMPIP registry is a trade association directory, so its coverage is its membership. Parks without AMPIP membership are missing. No attributes for occupancy, capacity, or green certification.