Total Economic Simulation
About this map
Where big companies physically exist: 49,000+ data centers, plants, mines, warehouses and headquarters of the S&P Composite 1500, about 3,000 large companies elsewhere, and the suppliers behind them, in 180+ countries. Pick a company to see its whole network, or filter by type, sector and country.
The first layer of a model of the real economy: where big companies physically exist. The data centers, plants, warehouses and headquarters of the S&P Composite 1500 (the S&P 500, MidCap 400 and SmallCap 600), of about 3,000 large companies outside it (from Saudi Aramco and Samsung to Coal India and the Chinese power groups), plus the suppliers they depend on and the suppliers behind those. Under it all, a grid of where every country’s GDP and people are. Goods flows, money flows and what-if shocks get built on top of this map. Pick a company to see its whole network, or filter by sector, facility type and country. Every point says where it came from and how sure we are about it. Full notes on sources are below the table.
Shortcuts: / search · Esc close a panel · F full-screen map
| Facility | Company | Type | Location | Run by | Status | Confidence | Source |
|---|
How this is built
Each company has its own data file listing its sites, and every non-S&P supplier (TSMC, Pou Chen, the big Coca-Cola bottlers) has one too. Links record who supplies whom. Tier 1 means a direct supplier or partner of the S&P company, and tier 2 means a supplier to one of those suppliers.
Confidence. reported means the company itself says so. third party means press coverage, another company's filing, or a public reference. estimated is our own best identification, and goods routes between specific plants are almost always estimated. Hollow rings are sites still under construction.
Precision. Some points mark the actual building, while most mark the town or region. Points in the same town are spread slightly apart so they don't sit on top of each other. Open a point to see which kind it is.
Nike's factories come from Nike's own Manufacturing Map (data as of July 2026): 712 contract factories with their workforce figures. Nike's list had a few location errors. Two had flipped coordinates and 14 Vietnamese factories were placed at the center of the country, so those were moved to their industrial zone or district from the address, and their notes say so.
Oil and gas. Eleven S&P 500 energy companies: the majors and producers (ExxonMobil, Chevron, ConocoPhillips, Occidental, EOG), the refiners (Marathon Petroleum, Phillips 66, Valero) and the pipeline companies (Kinder Morgan, Williams, ONEOK). Refinery capacities are rounded from EIA’s refinery capacity report. Oil and gas fields are drawn at basin or block level, and pipelines are drawn through their main corridor points, not surveyed routes. Crude and LNG shipments by sea follow real shipping lanes. Which refinery a given cargo goes to is usually our illustration, and those are marked estimated.
Infrastructure. Turn on layers in the panel. Highways and railways outside North America come from Natural Earth. North American rail is the federal North American Rail Network (FRA/BTS) main lines. Amtrak and VIA routes are traced along that network between their stations, and high-speed lines elsewhere are drawn station to station. Shipping lanes are Eurostat’s MARNET network, and ports come from Natural Earth. Solar farms are every solar power plant mapped in OpenStreetMap, plus older plants from WRI’s Global Power Plant Database where OpenStreetMap has none; dot size follows capacity. Biolabs are the world’s BSL-4 (maximum containment) laboratories as listed by the Global BioLabs project, placed at town level.
Data centers and licence plate cameras. Data centers are every site mapped in OpenStreetMap as a data center (or named like one), Wikidata items OpenStreetMap lacks, the site lists twelve operators publish (Equinix, Digital Realty, NTT, Vantage, QTS, STACK, EdgeConneX, Flexential, DataBank, Cologix, CoreSite, TierPoint), and the frontier AI data centers tracked by Epoch AI (CC BY); a listed site within 400 m of a mapped one is merged into it, and a site known only to the town is added only when that operator has nothing mapped nearby. Trackers count about 11,000 to 12,000 data centers worldwide, so some are still missing, mostly small ones and those in China, Russia and the Gulf. Licence plate cameras are every automatic licence plate reader (ALPR) mapped in OpenStreetMap (surveillance:type=ALPR), most of them found and mapped by volunteers of the DeFlock project; maker and operator are as tagged, and many cameras have not been mapped yet. Data © OpenStreetMap contributors, ODbL.
S&P MidCap 400 and SmallCap 600. These 1,000 companies come from Wikipedia’s lists of the two indexes, matched to SEC filers by CIK. Each one’s headquarters is the business address on its SEC filings, placed on the building with OpenStreetMap’s geocoder (or the town centre when the street can’t be matched). Its other sites come from three places: the towns named in the Properties section (Item 2) of its latest 10-K, typed from the surrounding words and drawn at the town centre, not the site; sites Wikidata lists as owned or run by the company or its subsidiaries; and sites OpenStreetMap mappers tagged with the company as operator or owner. Coverage is thinner than for the S&P 500, because many smaller companies describe their properties only in general terms. Use the Stock index filter to see one index at a time.
Large companies in Russia and Africa. US-listed companies have few sites in Russia or Africa, so this group adds about 1,000 companies from Wikidata: every business Wikidata lists as operating or owning a mapped site in Russia or an African country (RusHydro, Gazprom, Sonatrach, Sonelgaz, KenGen, AngloGold Ashanti, ALROSA and so on), plus the listed, state-owned and 1,000+ employee companies registered there. Each gets its headquarters, the power plants, mines, oil fields, plants and offices Wikidata places on the map, and its subsidiaries’ headquarters. For foreign groups (Heineken, Shell, TotalEnergies...) only their sites in Russia and Africa are drawn. Railway stations, hotels and airports are left out. Coverage follows what Wikidata volunteers have entered.
Large companies worldwide. About 1,100 more companies come from Wikidata (CC0): every business there with 15,000+ employees, a stock listing or a business industry that is not in the S&P 1500 or the Russia and Africa group (Toyota, Siemens, Nestlé, Samsung, TSMC, Saudi Aramco, SABIC, Iran Khodro, Reliance, Petrobras…). Each gets its headquarters, the sites Wikidata says it or a subsidiary owns or operates (two levels of subsidiaries), and its subsidiaries’ headquarters. Bank branches, post offices, stations, hotels and other non-sites are left out. Sectors are mapped from Wikidata’s industry to the S&P sector names, so treat them as approximate.
The biggest companies outside the US. About 190 of the largest companies by revenue that are not American (the Fortune Global 500 kind of list: State Grid, Saudi Aramco, Sinopec, CNPC, Volkswagen, Toyota, Samsung, Shell, Vitol, Glencore, Allianz, ICBC, Tata, Reliance, ADNOC…) are always included, even where the name rules above would have folded them into another company. Their footprint goes deeper: every subsidiary Wikidata lists (parent organization and subsidiary, two levels down) and every OpenStreetMap feature whose operator, owner or brand is tagged with one of those companies or subsidiaries (power plants, works, depots, data centers, rail yards, named offices). Insurance agencies, energy and phone shops and other customer fronts are left out. Many Chinese state groups still show fewer sites than they run, because neither Wikidata nor OpenStreetMap links most of their plants to them.
Company groups. The sources often name a different legal entity for the same group: Global Energy Monitor credits China Huaneng’s plants to Huaneng Power International and a dozen provincial companies, and CHN Energy’s to China Shenhua Energy. Where GEM’s ownership tracker shows a stake of 50% or more, or Wikidata lists one company on the map as the parent organization of another, the two are tied into a group. GEM names many Chinese state power owners (Huaneng Renewables, Datang Shaanxi, State Grid Xinyuan…) without any ownership record, so for the big state energy groups a power, energy or materials company that carries the group’s name joins it. That makes about 90 groups with 180 member companies: China Huaneng goes from 103 sites to about 400, CHN Energy from 1 to 69, State Grid from 18 to 60. Picking the parent shows the whole group, its card adds up the group’s plants and emissions and lists the members, and each member links back to its parent. Stakes that do not mean control, and businesses sold since Wikidata was updated (Brenntag, Daimler Truck, Covestro and others), were taken out by hand. Revenue figures stay the parent’s own, since a parent’s annual report already includes its subsidiaries.
Energy and industry owners. Whole economies were nearly empty because their big companies are state-owned or unlisted: Saudi Arabia, Iran, Iraq, the Gulf, much of China’s power and coal sector. Global Energy Monitor’s trackers name the owner and parent of every plant, mine and field, so every parent that runs at least 1.5 GW of operating power plants or five operating industrial sites became a company (about 940 of them: Coal India, NNPC, National Iranian Oil, Saudi Electricity, China Huadian, QatarEnergy…) with its operating plants of 20 MW or more and its industrial sites. Governments, ministries and individual shareholders are not companies and are skipped. Companies already on the map got the GEM sites they were missing.
Economy grid. The Economy switches draw Kummu et al.’s 2025 gridded GDP (Scientific Data, CC BY 4.0, Zenodo 16741980): each country’s 2022 GDP (PPP, constant 2017 dollars) spread over 5-arc-minute cells using its regions’ GDP per person, summed here to quarter-degree squares. People per square km is that GDP divided by the GDP per person. Hover for the half-degree square and the country’s totals, with World Bank GDP at market exchange rates for comparison.
Schools, universities and prisons. Schools are primary and secondary schools: every US public school in the NCES Common Core of Data 2024-25 and every private school in the NCES Private School Universe Survey 2023-24 (both public domain), and France’s Annuaire de l’éducation (Licence Ouverte 2.0); every other country comes from OpenStreetMap amenity=school (ODbL, pulled in October 2026 through overpass-api.de and Geofabrik’s Postpass), 1.39 million schools in all. OpenStreetMap coverage follows how much mappers have done: dense in Europe, Latin America, East Africa and South Asia, thinner in China and parts of the Gulf. Regions load as you zoom in; names and details load when you click a school. Universities and colleges: the US from IPEDS 2025 (directory and 12-month enrollment, public domain), elsewhere Wikidata (CC0) and ROR (CC0); most points outside the US sit at the city, not the campus. Prisons and detention: HIFLD Prison Boundaries for the US (public domain), OpenStreetMap amenity=prison everywhere (ODbL), ICE detention statistics for FY26, the CoreCivic, GEO Group and MTC facility lists, and Wikipedia’s lists of UK and Australian prisons (CC BY-SA). “Run by a private company” means a company runs the site day to day; where nothing says who runs it the site is marked “operator not known”.
Data centers and hospitals worldwide. These layers are everything OpenStreetMap mappers have tagged as a data center (plus untagged buildings named like one in 12 languages, and Wikidata’s data center items) and as a hospital, pulled in October 2026 (ODbL). Who runs each one is the operator tag; data centers are grouped into cloud, colocation, telecom, government and other by operator name, and hospitals into public, private, non-profit, religious and military by the operator:type tag or, where that is missing, by name patterns like “District Hospital” or “人民医院”. Coverage follows the mappers: very good in Europe, North America and Japan, thinner in China, Russia, Africa and the Gulf, and some countries tag large clinics as hospitals. A company’s sites in these layers are matched by operator name and shown on its card.
Layers. The map opens on the Layers tab, which now holds everything: picking a company, searching its sites and the company-site filters (type, sector, country, index, who runs it, status, confidence) sit under Company sites at the top. Type in the box at the top to find a layer, and the ones that are on are listed under it with an × to switch them off. Every layer gets its own filters once it is on (fuel, status and size for power plants; status, type, size or a name search for pipelines, LNG terminals, solar farms, labs, cables and ports; ship traffic for lanes; road and line types for roads and rail). Where a worldwide layer already draws a site, the company’s own record of the same plant, mine, data center, hospital or warehouse (within a couple of km) is hidden while that layer is on, so each one shows once, in the layer’s colors. The Company sites switch at the top hides every company site and supply link, and Turn all layers off gives a blank map. Under the layers, each layer that is on can be downloaded: a CSV of exactly what the map shows after your filters, or the whole source file.
Equal-area view. The Display tab has Mercator, Equal area and Globe. Equal area is the Behrmann projection (cylindrical, true scale at 30° north and south), so every country is drawn at its true size relative to the others. Satellite and street tiles only exist in Mercator, so they are off in that view.
Kraft Heinz comes from a sourced research file. It covers the 40 sites named in public sources (Kraft Heinz runs 69 plants but doesn’t publish a list), its named co-manufacturers and tomato sourcing, plus plants it has closed.
Power plants, every fuel. The power plant layer is Global Energy Monitor’s Global Integrated Power Tracker (map file of September 2026, CC BY 4.0): every coal, gas and oil, nuclear, hydro, wind, bioenergy, geothermal and utility-scale solar unit GEM tracks, merged into about 136,000 plants (solar loads separately because it is 91,500 of them). Operating plants are filled dots, plants being built or planned are rings, retired and mothballed ones are grey; cancelled and shelved projects are left out. Owners and parent companies come from GEM’s ownership data and are matched to companies on this map (by SEC CIK, by name, and up the chain of parent companies), which links about 4,200 plants and 1,300 GW to their owners’ networks. Generation figures are WRI’s (Global Power Plant Database, to 2019, partly estimated) where a WRI plant of the same fuel sits within 3 km.
Heavy industry. Steel plants, cement plants, coal mines, iron ore mines, oil and gas fields, coal terminals and large chemical plants come from GEM’s trackers of each industry (2024–2026, CC BY 4.0), with status, capacity, owners and parents, matched to companies the same way.
Emissions. The emissions layer is Climate TRACE’s estimate of 2024 greenhouse gas emissions (tonnes of CO2-equivalent, 100-year warming potential) for the biggest emitting sites in each source: power plants, oil and gas production and refining, coal mines, steel, cement, chemicals, airports, landfills, cattle feedlots and more. About 29,000 sites, 28 billion tonnes in all, roughly half of the world’s emissions. Climate TRACE models these from satellite observations, sensors and activity data; they are estimates, not company-reported figures. Owners use the same GEM company ids and are matched to companies on this map.
Money. Revenue, net income, operating income, capital spending, R&D, total assets, plant and equipment, equity and cash come from the XBRL data in each company’s SEC filings (EDGAR frames API, the fiscal year ending in 2025 where filed, else 2024). Banks that report no revenue total show interest and dividend income instead. Companies that do not file with the SEC get the latest figures from their English Wikipedia infobox where one gives them (most of the biggest non-US companies, quoting the annual report), else from Wikidata’s revenue, profit, assets, equity and employee statements, converted to dollars at the World Bank’s average rate for that year. Those are entered by volunteers, so they can be a year or two old, and values that are clearly in the wrong unit (more than $3 million of revenue per employee outside oil, power, finance and trading) are dropped.
Trade by product. Type any good in plain words (coffee, laptops, electric cars, shoes, lithium) in the Trade layer’s search, or pick a popular one, and the map draws who sells it to whom. Up to eight products can be drawn at once, each in its own colour, alone or for one country. The search covers the 1,228 four-digit HS headings and 96 HS chapters (“all of it”, such as all coffee, tea and spices) under shortened names, the official heading text and everyday words, so nobody needs to know an HS code. Product figures are CEPII’s BACI database (HS 2022, version 202601, released January 2026; Etalab Open Licence 2.0), which reconciles what the exporter and the importer each reported to UN Comtrade into one value per flow, free on board; 2024 values with 2023 for comparison. For each product we keep its 300 biggest country-to-country flows, its top 15 sellers and buyers, and for each country its top 40 products each way with their 5 biggest partners. BACI figures differ a little from the Comtrade totals in the main trade layer because of the reconciliation, 2024 is preliminary, services are not included, and goods passing through hubs such as the Netherlands, Singapore or Hong Kong count as their trade.
Railways worldwide. Outside North America the main lines are Natural Earth’s 10m railroads, now colored by the railway that runs them. The operator comes from OpenStreetMap: every railway line tagged with an operator is counted in half-degree squares, and each Natural Earth line takes the operator with the most tagged track in its square (DB InfraGO, SNCF Réseau, Russian Railways, China Railway, Indian Railways, JR East…, about 190 in all). It is an approximation: a short private or industrial line inside a national network’s square takes the national operator, and India and much of Africa have few operator tags. Pick a railway company under the layer to draw only its lines.
Live ships. Positions of ships right now, from the AIS radios every larger ship (and many fishing boats and yachts) carries. Worldwide positions come from aisstream.io, which relays shore-based receivers run by volunteers, so busy coasts, ports and straits are well covered and the open ocean is mostly empty (satellite AIS is a paid service). aisstream.io does not allow web pages to connect directly, so this site runs a small relay that keeps one connection and hands every visitor the ships heard in the last hour. The northern Baltic also comes straight from Finland’s coastal stations (Fintraffic / digitraffic.fi, licence CC 4.0 BY), and that part still works if the relay is down. Both refresh every minute while the layer is on; arrows point the way a ship is heading, dots are ships at anchor, moored or stopped, and colour is the type the ship reports.
Satellites. Every working satellite in CelesTrak’s catalogue (about 16,000), placed where it is right now. CelesTrak publishes each satellite’s orbit (the GP element sets, updated through the day from US Space Force tracking); your browser works out every position once a second with the standard SGP4 model (satellite.js). Owners and launch dates are from CelesTrak’s SATCAT. Kinds come from CelesTrak’s groups (navigation, weather, Earth resources, science, space stations) and the names of the big constellations; everything else, mostly communications and military satellites, is “Other”. Click a satellite to draw one lap of its orbit. On the globe, heights up to 2,000 km (every low orbit, Starlink at about 550 km, the space stations at about 420 km) are to scale; higher ones are drawn at a quarter of their real height so the navigation belt (about 20,000 km) and the geostationary ring (35,786 km) stay on screen. CelesTrak lets each address download the list once every 2 hours, so the site relays it and your browser keeps a copy.
Trade between countries. The Trade switch draws goods exports from each country to each partner as reported to UN Comtrade (annual, 2023, or 2022 where 2023 is not out; read from the Comtrade API and, for the countries past its daily limit, from the World Bank’s WITS service, which republishes the same figures), every flow of $100 million or more, with each country’s top exports (by HS chapter, or by HS section where only that is available). Click a country for its GDP, people, top buyers and suppliers and top exports. Countries that report nothing to Comtrade only show up as buyers, services are not included, and re-exports through Hong Kong, Singapore and the Netherlands count as their exports.
Warehouses and logistics worldwide. The Warehouses & logistics hubs switch is every OpenStreetMap warehouse, depot, sorting center and logistics yard that has a name or an operator (October 2026, ODbL), up to the 20,000 biggest, grouped by name and operator into parcel, freight, e-commerce, retail, cold storage and other. Sites run by a company on this map are matched by operator name and shown on its card.
Places. The Places switches draw every location that brands publish in their own store finders, collected by All The Places (CC0), plus OpenStreetMap where those don't reach: fast food, coffee shops, hotels, rental cars, bank and credit union branches, car dealerships, grocery stores, Walmart, Target, Costco and other warehouse clubs, post offices, DMVs, city halls, courthouses and embassies. Bank headquarters come from Wikidata. All The Places only covers about 20 national postal services and mostly Western banks, so post offices and bank branches everywhere else (China, India, the Gulf, Africa, Latin America) are filled from OpenStreetMap’s amenity=post_office and amenity=bank, grouped by brand or operator; unbranded OSM banks are only kept in countries All The Places barely covers. Food distribution centers (Sysco, US Foods, McLane, Martin Brower, HAVI and others) and Amazon warehouses come from OpenStreetMap and are far from complete. ChangoMâs (Argentina, 89 stores) comes straight from its own store locator at masonline.com.ar, which All The Places does not cover. Each switch loads its own data the first time you turn it on.
Distribution networks. Warehouses, fulfillment centers, depots and hubs for Amazon, Walmart, Kroger, Costco, Home Depot, UPS, FedEx, CVS, TJX, Walgreens, Starbucks, Berkshire Hathaway’s industrial companies and McDonald’s supplier Martin Brower come from the distribution maps published by the On the Seams newsletter (2025–2026). Sysco’s 143 sites come from its own locations list.
Pipelines and LNG terminals. Gas pipelines come from Global Energy Monitor’s Global Gas Infrastructure Tracker (November 2025 release) and oil and natural gas liquids pipelines from its Global Oil Infrastructure Tracker (map file of June 2026), both CC BY 4.0: about 3,200 gas, 1,200 oil and 230 NGL pipelines or sections that are operating, being built, proposed or idle (cancelled, shelved and retired projects are left out). Dashed lines are being built or proposed, or for gas, drawn by GEM between known end points rather than traced. Oil capacity is GEM’s, in barrels of oil equivalent a day; the oil file has no route accuracy, so its lines are drawn solid. LNG terminals are GEM’s September 2025 release (its November 2025 map file lists the same terminals), with five units that started up since then marked operating from company and US EIA reports: Golden Pass Train 1 (first cargo April 2026), Corpus Christi Stage 3 (all seven trains complete August 2026), Energía Costa Azul (first cargo July 2026), Eni’s Nguya FLNG off Congo (February 2026) and Plaquemines Phase 2 (producing since July 2025 while commissioning). Qatar’s North Field East and Woodside’s Pluto Train 2 are still shown as being built, as their first cargoes had not shipped by October 2026.
Mines and refineries. The Mines and refineries switch draws about 7,000 mines, smelters, refineries and processing plants for metals and minerals: every facility in the USGS Minerals Yearbook 2024 country chapters (volume III, public domain) and the USGS database of US mines, smelters and refineries (2026), with what each produces and its capacity, its operator and owners. The yearbook has no coordinates, so sites are placed from the USGS regional mineral GIS (Africa, China, Southwest Asia, Indo-Pacific, Latin America), Jasansky et al. 2023 (CC BY 4.0), Geoscience Australia, Natural Resources Canada and Wikidata where the names match, and otherwise at the town or province named in the yearbook; each card says which. Coal, iron ore, steel and cement are in the GEM industry layer instead. The yearbook lists producing and idle sites, so 20 projects it misses are added by hand from news and company reports checked in October 2026, each with its source: rare earths in Jamaica (a closed 2013 red mud pilot and an exploration area), Greenland (Tanbreez), Brazil (Serra Verde), Malawi, Angola, Uganda, Tanzania, South Africa, Sweden, Australia and Estonia, lithium in Nevada, Serbia and Ukraine, copper at Reko Diq and Resolution, and seabed nodules in the Pacific. Being built or explored is not producing: most of these will take years, and some may never open.
Who runs each country. Click a country on the flat map (or the “Who runs” link on a What if country card) for its head of state and government, cabinet ministers, central bank governor and envoys: the CIA’s World Leaders directory (public domain) for 199 countries, saved on 18 October 2026, with each page’s own update date shown (most are from 2025 or 2026, a few older). The CIA lists foreign governments only, so the US cabinet and senior White House staff come from Wikidata (current office holders, October 2026), with this year’s changes (attorney general, homeland security, labor, intelligence, the Fed chair, press secretary) checked against news reports.
Checks. Every lab is tested against identities it must satisfy and against figures published by the OECD, IMF, CBO, HMRC, the CEQ Institute and the World Bank; the results are rebuilt from the test scripts each time the site updates: how the simulation is checked.
Debt and deficits. What if > Tax policy. Source: International Monetary Fund, World Economic Outlook database (general government gross debt, primary net lending, interest as primary minus overall net lending, nominal and real GDP growth, 2024-2031, and GDP in US dollars for 2026); 177 countries with every number. Transformed: the lab rebuilds each path in money (debt plus interest minus the primary balance plus the residual other flows), holds the 2031 primary balance, growth and interest rate on debt to 2045, and re-runs it with a fiscal multiplier (fading linearly), the budget’s automatic response to GDP, interest rates passed through as debt is refinanced or newly issued, and growth and inflation changes.
Income tax in 131 countries. What if > Tax policy. OECD Taxing Wages 2025 country tables for a single worker at 67%, 100% and 167% of the average wage (allowances, taxable income, central tax, credits, local tax, employee social security), the OECD Tax Database central government brackets for 2025 (Germany: the EStG section 32a formula with the solidarity surcharge), OECD average and marginal rates at 133% (used only to check the lab), OECD decile ratios of full-time gross earnings, and personal income tax revenue from OECD Revenue Statistics; all CC BY 4.0. Earnings follow a two-piece lognormal fitted to the decile ratios with a Pareto tail (alpha 2.5) above the 90th percentile; revenue changes scale actual income tax revenue by the modelled change; static, no behaviour. Argentina, Barbados, Brazil, the Dominican Republic, Ecuador, El Salvador, Guatemala, Guyana, Honduras, Nicaragua, Panama, Peru, Saint Lucia, Trinidad and Tobago and Uruguay are not in Taxing Wages: their 2025 income tax and employee social security rules for a single employee are coded from the tax offices’ own tables (Receita Federal and INSS, ARCA and ANSES, SUNAT, DGII and TSS, the Panamanian Fiscal Code and CSS, Guatemala’s Decreto 10-2012, Uruguay’s DGI and BPS, El Salvador’s Decreto Ejecutivo 10 of 2025, Honduras’s SAR and IHSS, Ecuador’s SRI and IESS, the Guyana Revenue Authority, Nicaragua’s Ley de Concertación Tributaria and INSS, Saint Lucia’s Inland Revenue Department; for Trinidad and Tobago and Barbados, PwC and KPMG summaries and the NIBTT), with the ILO’s mean monthly earnings of employees times 12 as the average wage (raised to 2025 by IMF consumer prices where the latest figure is older; Nicaragua: the central bank’s average salary of INSS-insured workers; extra months such as the 13th-month salary left out), the average decile ratios of the four Latin American OECD members as the spread, and OECD Revenue Statistics in Latin America and the Caribbean 2025 for revenue (CC BY 4.0; ILOSTAT CC BY 4.0). Botswana, Côte d’Ivoire, Egypt, Eswatini, Ghana, Kenya, Lesotho, Mauritius, Namibia, Nigeria, Rwanda, Senegal, South Africa, Tanzania, Uganda and Zambia are built the same way from their 2025 rules (SARS; KRA; Deloitte’s 2025 tax highlights for Botswana, Egypt, Ghana, Kenya, Nigeria, Senegal, Uganda and Zambia; PwC for Côte d’Ivoire, Eswatini, Mauritius, Namibia, Rwanda, Senegal and Tanzania; RSSB; Lesotho’s 2025/26 amendments; Mauritius’s Finance Act 2025), with revenue from OECD Revenue Statistics in Africa 2025 (none for Botswana, Côte d’Ivoire or Tanzania). The French-speaking systems divide income by family shares; the lab’s single worker has one share. Armenia, Cambodia, China, Georgia, India, Indonesia, Kazakhstan, Malaysia, Mongolia, Pakistan, the Philippines, Singapore, Sri Lanka, Thailand, Uzbekistan and Vietnam are built the same way (Deloitte’s 2025 tax highlights, with PwC for allowances, the Union Budget 2025 for India’s new regime, Sri Lanka’s Inland Revenue (Amendment) Act 2025, Pakistan’s Finance Act 2025, the CPF Board’s 2025 contribution table for Singapore, and payroll guides where noted), with revenue from OECD Revenue Statistics in Asia and the Pacific (none for India, Pakistan or Uzbekistan); Pakistan’s average wage is the Pakistan Bureau of Statistics’ Labour Force Survey 2024-25 figure and Singapore’s the Department of Statistics’ average monthly earnings per employee for 2025. Bulgaria, Croatia, Cyprus, Malta and Romania, EU members outside the OECD, are built the same way from their 2025 rules (Deloitte’s 2025 tax highlights; the City of Zagreb’s rates; Malta’s Class 1 table; Romania’s Fiscal Code), with Eurostat’s 2025 average worker (a full-time worker in industry, construction and services) as the average wage, the average decile ratios of the eight central and eastern European OECD members as the spread, Eurostat national accounts for wages and salaries, and OECD Global Revenue Statistics for revenue (Cyprus, not in it: Eurostat government finance statistics); at the average wage each is checked against Eurostat’s own income tax and social security figures (Eurostat, CC BY 4.0). Moldova and Serbia are built the same way (PwC; Moldova’s State Tax Service; for Serbia, the 2025 non-taxable amount and contribution bases), with the same spread; Moldova’s average wage is the ILO’s and its revenue OECD Global Revenue Statistics, Serbia’s average wage is its Statistical Office’s 2025 average gross salary and Eurostat national accounts give its wages and salaries (no revenue figure: Serbia is not in the OECD’s revenue statistics). Hong Kong, Bangladesh, Kyrgyzstan, Jordan, Fiji, Azerbaijan, Bhutan, Ukraine and the six Gulf states (Saudi Arabia, the United Arab Emirates, Qatar, Kuwait, Bahrain, Oman, which tax no wages; their nationals’ social insurance is shown) are built the same way from the best sources found (PwC; tax offices; payroll guides; Hong Kong’s 2025 Annual Earnings and Hours Survey, whose percentiles give its spread of pay and, through the lab’s earnings model, its average wage; for the United Arab Emirates, Kuwait and Oman, which publish no average wage, the labour income share of GDP over the number of employees, ILO and IMF), and each says in the lab what is less certain. Bolivia, Paraguay, Burkina Faso, Mali, Togo, Benin, Chad, Seychelles, Lao PDR, Albania and Montenegro rest on the thinnest sources of all (PwC where it covers them; otherwise tax offices’ brochures, tax commentaries and payroll firms), and say so. Tunisia (Finance Law 2025), Morocco (Finance Law 2025; the average wage declared to the CNSS), Mozambique (RSM’s 2025 tax guide), Ethiopia (Proclamation 1395/2025) and North Macedonia (the State Statistical Office’s 2025 average wage) are built the same way. So are Russia (Rosstat’s 2025 average wage), Belarus (Belstat’s), Kosovo (the Agency of Statistics’), Bosnia and Herzegovina (the Federation’s rules; the Agency for Statistics’ wage for the whole country), Taiwan (PwC; the DGBAS average total pay per employee), Nepal (the FY 2082/83 slabs), Niger (the General Tax Code), Angola (the IRT table in force through 2025) and Cameroon (the DGI’s IRPP guide); of these only Niger and Cameroon are in the OECD’s revenue statistics. Where the ILO’s latest wage figure was years old, the national statistics office’s 2025 average is used instead (South Africa, Kenya, Malaysia, Kazakhstan, Uzbekistan, Kyrgyzstan, Azerbaijan, China, Georgia, Armenia, Indonesia, Ukraine, Albania and Montenegro; mostly enterprise surveys, so closer to formal jobs than the ILO’s labour force figures), and for India, Taiwan, Russia and Serbia, which are in none of the OECD’s revenue statistics, income tax revenue comes from the tax office or finance ministry over GDP (the World Bank’s unless stated). For Albania, Belarus, Bosnia and Herzegovina, Ecuador, Jordan, Kosovo, North Macedonia and Tanzania it comes from UNU-WIDER’s Government Revenue Dataset (2025 edition, taxes on individuals, CC BY 4.0, read through Our World in Data), which gives the same figure as the OECD at the median across the 50 countries both cover. Each country’s panel names its figures.
Who gains from taxes and spending. What if > Tax policy. 99 fiscal incidence studies in 77 countries from the CEQ Standard Indicators (CEQ Data Center on Fiscal Redistribution, CEQ Institute, Tulane University, April 2024 version; used under citation): the Gini of market income (with contributory pensions), disposable, consumable and final income, the budget’s taxes and spending as shares of GDP, and for 56 studies the incidence of direct taxes, cash transfers, indirect taxes, subsidies and in-kind education and health by tenth of the population. The lab rebuilds the ten groups’ average incomes from the incidence table (their Gini is 96% of the reported one across studies) and moves the reported Gini by the same proportion; static, no behaviour.
Consumption tax lab. What if > Tax policy opens a full-screen lab for VAT and fuel duty in 38 countries (the EU, the UK, Norway, Japan, Canada’s federal GST, Mexico, Turkey, the western Balkans, and a new VAT for the US). VAT rates by spending category in force on 1 October 2026 come from the European Commission’s Taxes in Europe Database and PwC Worldwide Tax Summaries; where food is taxed at several rates the lab uses a rough average and says so. Budget shares by fifth of households come from the national household surveys behind Who pays. Each country is calibrated to the VAT it collects (OECD Revenue Statistics): the share of ‘everything else’ that carries VAT is set so the rates give the revenue. Price changes reach shops at 90% by default; estimates are static and for one year.
US tax lab. What if > Tax policy opens a full-screen lab for 2026 federal income and payroll taxes. The tax rules are the open-source Tax-Calculator model (version 6.8.5, PSLmodels, CC0 public domain), translated line by line into JavaScript and checked against it: the same answer for every one of its 280,005 sample tax units, and for ten test reforms, under 2026 law after the 2025 tax act. The households are Tax-Calculator’s Census CPS file aged to 2026; the page runs a stratified sample of 60,643 (all with income over $500,000), within 0.1% of the full file’s revenue. Estimates are static (no change in behaviour), federal only and for one year. The CPS is a survey without capital gains, partnership and S-corporation income, tips, overtime or car-loan interest, and it misses much of the top of the income distribution, so revenue levels are lower than official ones, changes aimed at the top are understated and the 2025 tips, overtime and car-loan deductions are not offered.
Power grid. The Power grid switch draws every transmission line and substation of 110 kV and up that OpenStreetMap mappers have recorded (745,900 line sections, about 4.5 million km, and 119,500 substations, pulled through Geofabrik’s Postpass service on 21 October 2026, ODbL), colored by voltage: lines of 220 kV and up show from the world view and 110 to 199 kV lines appear as you zoom in. Lines tagged as being built or planned are dashed, and HVDC links are drawn in their own colour. Where a line carries several voltages the highest is shown. The lines are vector tiles read straight from PMTiles files (one for the world view, regional files as you zoom in), so only the part of the world on screen is downloaded. Three analyses sit on top: each data center’s nearest substation and its voltage (open a data center), data-center clusters (5 or more within 25 km of each other) against the operating power plants within 50 km from Global Energy Monitor, and a “what’s within 25 km” panel for any point you click (plants, substations, data centers, hospitals, ports). In the earthquake and solar-storm what-ifs the lines and substations in the zone turn red or amber, and the panel counts the kilometres of line and the substations inside it; for a solar storm only lines of 300 kV and up are flagged, since long high-voltage lines carry the strongest induced currents. Limits: OpenStreetMap is close to complete for transmission in Europe and North America but thinner in China, Russia, parts of Africa and the Gulf, and voltage tags are sometimes missing (those lines are left out). Grid stress is a rough first pass, not a forecast: it shows what lies in the zone, not what would trip.
What is still thin, and why. Some gaps have no open worldwide data to fill them: factory headcounts (only US OSHA and EU registries publish them), rail network owners outside North America, US-style local government offices (DMVs and courthouses have no common tag elsewhere), and supplier lists, which only US companies publish in their SEC filings. The trade layer stands in for supply links between countries.
Companies so far: every company in the S&P Composite 1500, about 800 large companies in Russia and Africa, about 1,100 large companies elsewhere from Wikidata, about 940 energy and industry owners from GEM, and several hundred suppliers and partners. The deepest coverage is for Meta, Coca-Cola, Nike, Apple, Amazon, Microsoft, Alphabet, Walmart, Kroger, Costco, Sysco, PepsiCo, Procter & Gamble, Tesla, the big oil, chip, pharma, defense, rail, utility and food companies, with their supply chains. More are coming, sector by sector. Spot something wrong? Email me.