Skip to content

AI data market

Data market geographic expansion: count usable supply, not firms

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

Data market geographic expansion adds usable supply only where a country's official firm count matches the eligibility threshold and its records pass four gates: English on the records, rights and transfer review, workflow fit and an approved processing route. Illustrative gates keep 41 to 98 percent of an English-majority frame and 1 to 20 percent of a non-English one; Australia's size bands alone leave a 13.7-fold range.

Countries and languages

Data market geographic

This page covers data market geographic expansion: count usable supply, not firms.Illustrative

Key takeaways

  • A new country adds firms in proportion to its count but usable supply in proportion to four gate pass rates, none of which is published for any country.
  • Official frames are not comparable: the UK counts 46,770 businesses with 50 or more employees (start of 2025), while Australia's bands bound the same group only between 5,366 and 73,691 (June 2026).
  • Household language is a poor proxy for record language: English is the main home language for 48.3 percent of Singapore residents (2020) and 8.7 percent of South Africans (2022).
  • With illustrative gates, an English-majority case keeps 41 to 98 percent of its frame and a non-English-majority case keeps 1 to 20 percent, driven by the language gate.
  • Credit new-country supply only with a sourced frame, English measured on offered records and pass rates by gate, and ask SourceX the same questions.

What does geographic expansion add to a data licensing market model?

Geographic expansion adds firms to a data licensing market model at once, but it adds usable supply only after each new country's records pass the gates that decide whether a dataset can be licensed. A model that multiplies a foreign firm count by a home-market yield skips those gates and overstates what expansion delivers.

Usable supply here means companies holding datasets that would clear a licensing company's eligibility rules and its rights, privacy and processing review. The investor's question is whether adding a country grows that number roughly in line with the country's firm count, or by much less. Four kinds of evidence answer it: an official firm count defined at the same threshold as the eligibility rule, the measured English share of records actually offered, a rights and transfer pass rate, and the share of companies whose systems and workflows fit buyer demand.

SourceX makes a clear test case because its rules are published. It works with U.S., Canadian and EU companies that had 50 or more full-time employees at peak, and it can consider companies in the UK, Australia, New Zealand, India, Singapore and South Africa, and in Latin America, the UAE and Saudi Arabia, when the offered data is at least 90 percent English as confirmed on the dataset itself. Eligibility in those countries turns on the dataset rather than the company's location, and every dataset still faces rights, privacy, quality and buyer-fit review. Those rules are the gates this article prices.

Why a country's firm count is not a supply frame

A firm count becomes a supply frame only when its unit, size band and date match the eligibility rule, and national statistics rarely line up. The same 50-employee threshold reads very differently across the three official tables this article could open.

Eurostat's release of 9 December 2025, on preliminary 2024 data, counts 55,000 large enterprises (more than 249 people employed) and 251,000 medium-sized enterprises (50 to 249 people employed) in the EU business economy, 306,000 in all. The UK's Business population estimates 2025, published by the Department for Business and Trade on 2 October 2025 as official statistics in development, count 38,435 medium-sized private sector businesses with 50 to 249 employees and 8,335 large businesses at the start of 2025, 46,770 in all.

Australia's business counts release has no 50-employee cut. The Australian Bureau of Statistics release covering July 2022 to June 2026 counts 5,366 businesses with 200 or more employees and 68,325 with 20 to 199. Businesses with 50 or more employees therefore number somewhere between 5,366 and 73,691, a 13.7-fold range before any gate is applied.

Even the matching counts differ in what they count. Eurostat counts people employed and the UK counts employees; neither measures full-time equivalents or size at a company's peak, and both count legal units that may sit inside one corporate group. For seven of the nine extension markets named in SourceX's rules, this article read no official frame at all, so the table below lists them without a count:

Why a country's firm count is not a supply frame
FrameSource and dateDefinitionWhat it does not measure
306,000 EU enterprisesEurostat, 2024 preliminary data, released 9 December 202555,000 with more than 249 people employed plus 251,000 with 50 to 249Full-time equivalents, size at peak, enterprise groups, record language
46,770 UK businessesDepartment for Business and Trade, start of 2025, published 2 October 202538,435 private sector businesses with 50 to 249 employees plus 8,335 with 250 or moreSame gaps; official statistics in development
5,366 to 73,691 Australian businessesAustralian Bureau of Statistics, release to June 20265,366 with 200 or more employees; 68,325 with 20 to 199; no 50-employee cutWhere the 50-employee line falls inside the 20 to 199 band
No frame readIndia, Singapore, South Africa, New Zealand, Latin America, UAE, Saudi ArabiaLeft as reader inputsEverything; supply claims for these markets need a sourced count first

Four gates every expansion case has to pass

Each expansion case has to pass four gates, language on the records, rights and transfer, workflow fit and an approved processing route, and none of their pass rates is published for any country. Each gate removes companies the firm count still includes.

Language comes first for the extension markets. SourceX measures the English share on the records actually offered, so a company whose help desk runs in two languages can fail even in a country where English is an official language. The study design for measuring that share is set out in the business dataset English coverage protocol; no country-level result exists yet.

Rights and transfer come second. SourceX's legal framework page says it reviews customer contracts, privacy notices and employee policies and leaves out data a company has no right to share, and that data covered by GDPR is only included when there is a valid legal basis and the right transfer safeguards. Each origin country has its own transfer rules, mapped country by country in the guide to cross-border data transfer mechanisms; the pass rate depends on counsel's reading in each case.

Workflow fit and processing come last. SourceX's how-it-works page says qualifying companies list at least five systems, and its public guidance says buyers usually look for years of consistent history and tens of thousands of records or more in one system, with smaller sets considered when the work is rare. A country whose mid-sized firms run fewer systems, or keep shorter histories, passes this gate less often. The processing gate asks whether an approved de-identification route exists for that jurisdiction's records.

Is household language a usable proxy for English records?

No. Census language figures measure homes, the gate measures business records, and the two can diverge in either direction, so a household figure cannot stand in for the English share of a dataset.

Two official censuses show how far household language sits from any business assumption. Singapore's Census of Population 2020 found English the language most frequently spoken at home for 48.3 percent of residents aged 5 and over. South Africa's Census 2022 found English the language most often spoken in the household for 8.7 percent of the population aged one and over. A model that used those shares as the English pass rate would rate South Africa's records far below Singapore's without looking at a single help desk or ledger.

Demand does not settle the question either. OpenAI's data partnerships page, as summarized in a search excerpt, says it is interested in data across any language and describes a partnership aimed at improving a model's Icelandic. A 90 percent English rule is one company's supply rule, not a law of the market, which matters again in the risks below.

The expansion workbook: frames multiplied by gates

The workbook multiplies each frame by four gate pass rates and reports the result as home-firm equivalents: the number of home-market companies that would contribute the same usable supply. Every gate value is an illustrative assumption, not a measurement, and the gates for rights, workflow and processing are set relative to the home market and capped at 1, meaning an expansion case is assumed never to be easier than home.

Arithmetic for the UK, central case: 46,770 times 0.93 times 0.90 times 0.95 times 0.95 equals about 35,330 home-firm equivalents. The low case, 46,770 times 0.85 times 0.75 times 0.80 times 0.80, gives about 19,082; the high case, 46,770 times 0.98, gives about 45,835. For Australia the frame itself is a range, so the honest result runs from 5,366 times 0.408, about 2,189, to 73,691 times 0.98, about 72,217, and no central figure can be stated without a 50-employee count.

For a non-English-majority market the language gate dominates. Per 10,000 firms in its frame, the same rights, workflow and processing values with an English pass rate of 0.02, 0.08 or 0.20 leave about 96, 650 or 2,000 home-firm equivalents. To turn home-firm equivalents into datasets, multiply by whatever home-market yield a model already uses; this article does not estimate one. The inputs and results:

The expansion workbook: frames multiplied by gates
Input (illustrative unless sourced)English-majority case: low / central / highNon-English-majority case: low / central / highEvidence that would replace it
Frame (firms at the threshold)UK 46,770; Australia 5,366 to 73,691 (sourced)Reader input; shown per 10,000 firmsAn official count at 50 employees, same unit and date
English on offered records, at least 90 percent0.85 / 0.93 / 0.980.02 / 0.08 / 0.20Language measured on sampled records from offered datasets
Rights and transfer, relative to home0.75 / 0.90 / 1.000.75 / 0.90 / 1.00Counsel's pass rate on reviewed companies, by country
Workflow fit, relative to home0.80 / 0.95 / 1.000.80 / 0.95 / 1.00Systems listed and history depth from assessments
Processing route, relative to home0.80 / 0.95 / 1.000.80 / 0.95 / 1.00Approved de-identification routes by jurisdiction
Combined multiplier0.408 / 0.755 / 0.9800.0096 / 0.065 / 0.200Product of the four rows above
Home-firm equivalentsUK 19,082 / 35,330 / 45,835; Australia 2,189 to 72,21796 / 650 / 2,000 per 10,000 firmsFrame times multiplier, rounded half up at the end

Which input moves usable supply most in an expansion case?

For English-majority markets the frame definition moves usable supply far more than any gate; for non-English-majority markets the language gate moves it most. That ranking, not the totals, is what an investor can carry into a diligence call.

Varying one gate at a time on the UK frame, with the others at central values, the language gate moves the result from 32,291 to 37,229, 1.15 times; rights and transfer from 29,441 to 39,255, 1.33 times; workflow and processing each from 29,751 to 37,189, 1.25 times. All four together, low against high, move it 2.40 times. Australia's published size bands alone move the frame 13.7 times.

In the non-English-majority case, moving the English pass rate from 0.02 to 0.20 is a tenfold change on its own and drives most of the 20.8-fold range from low to high. The figure shows both results: the share of each frame left after the gates, and the Australian frame bracket.

Left, two range bars of the share of an official frame left after four illustrative gates, 41 to 98 percent for an English-majority case and 1 to 20 percent for a non-English-majority case, each with a central mark; right, two bars from zero for Australian businesses with 200 or more employees, 5,366, and 20 or more, 73,691, between which the 50-employee count must lie.
Gate values are illustrative assumptions; the Australian counts are from the ABS release covering July 2022 to June 2026.

Could AI budgets, data volume or venture rounds make expansion bigger?

No, because none of those measures counts licensable records in a country. A larger AI budget, more stored data or a new funding round can raise buyer demand or a company's capacity, but it does not add one eligible dataset to a market whose records fail the language or rights gate.

The tempting shortcut is to scale a country's supply by its share of AI investment or data center capacity. Those figures describe where models are trained and bought, not where business records sit, in what language, under which contracts. A country can lead on AI spending and still contribute little usable supply to a company whose rules require English records, and the reverse can also hold.

The one demand signal that does bear on supply is language. If buyers pay for non-English records, as OpenAI's page suggests some do, a company that limits itself to English leaves that supply to others; that is a strategic choice to price, not evidence that the supply is absent.

What could make this expansion analysis wrong

Five developments could make this analysis too pessimistic or too generous. Each points to evidence an investor can check rather than a forecast.

  • English shares could be higher than assumed. Subsidiaries of multinationals in non-English markets may keep records in English; a measured sample of offered datasets would replace the 0.02 to 0.20 range.
  • Eligibility rules can change. SourceX's list of extension markets and its 90 percent threshold are company rules; a lower threshold, or added translation and multilingual processing, would raise the language gate overnight.
  • Transfer regimes can tighten or loosen. New adequacy decisions, localization rules or court rulings move the rights gate country by country, and counsel's pass rates, not this workbook, should carry that change.
  • Official frames get revised. Eurostat's count is preliminary and the UK release is labeled official statistics in development, so the frames above can move between releases.
  • Group structure can shrink the frame. Several legal units in a count may belong to one rights holder, so the number of companies able to sign can be smaller than any official count.

Questions for any company claiming new-country supply, SourceX included

Ask for the frame, the gates and the pass rates, each defined in writing, before accepting any claim that a new country expands supply. The same questions apply to SourceX, whose eligibility rules are public but whose pass rates are not.

SourceX shares company and commercial metrics with investors directly, as its investor page says, and none of its assessment, pass-rate or supply figures appear in this article. SourceX is raising from accredited investors under Rule 506(c), and the investor page is where to request its materials and put these questions to the company. Ask:

  • Which official count defines the frame in each new country, at what size threshold, unit and date?
  • How is the English share measured on offered records, on what sample, and what share of assessed companies passed the threshold?
  • What share of companies in each country cleared rights and transfer review, and which clauses or rules caused most failures?
  • How many systems and years of history do assessed companies in each country list, against the home market?
  • Does an approved processing route exist for each jurisdiction's records, including regulated ones?
  • Which new-country datasets reached signed terms, counted against which denominator of assessments?

Methods note: frames, workbook inputs and what was not measured

Frames were read on 2026-10-28 from each publisher's own page through domain-restricted search excerpts: Eurostat's news release of 9 December 2025, the GOV.UK Business population estimates 2025 statistical release and the Australian Bureau of Statistics Counts of Australian Businesses latest release. The census language shares come from Singapore's Census of Population 2020 Statistical Release 1 and Statistics South Africa's Census 2022 statistical release. SourceX's rules come from its public facts and its how-it-works and legal framework pages.

Workbook: home-firm equivalents equal the frame times L times R times W times P, where L is the share of frame firms whose offered dataset is at least 90 percent English and R, W and P are pass rates relative to the home market, capped at 1. All four are illustrative. The UK frame adds 38,435 and 8,335; the Australian bounds are 5,366 and 5,366 plus 68,325. Calculations use exact fractions in bin/expansion_workbook.py in this batch's scratch folder and are rounded half up to whole firms only at the end; the one-at-a-time sensitivity holds the other three gates at central values.

Not measured: any country's English share on business records, any gate pass rate, frames for India, Singapore, South Africa, New Zealand, Latin America, the UAE and Saudi Arabia, the home-market yield, and buyer demand by language. Nothing here reports SourceX assessments, pass rates or supply.

Frequently asked questions

Does adding countries increase the AI data market's addressable supply?

Only after the new country's records pass the same gates as home: language on the records, rights and transfer, workflow fit and a processing route. With illustrative gates, an English-majority case keeps 41 to 98 percent of its official frame and a non-English-majority case keeps 1 to 20 percent.

Why can't I add official firm counts across countries?

Because they count different units at different size bands and dates. Eurostat counts people employed and the UK counts employees, and Australia publishes no 50-employee cut, so its 50-plus count lies anywhere between 5,366 and 73,691. Align the definitions before adding anything.

Is a country's English-speaking population a good proxy for English business records?

No. Census figures measure household language: 48.3 percent of Singapore residents speak English most at home (2020) against 8.7 percent of South Africans (2022). The relevant figure is the English share measured on the records a company actually offers.

Which countries does SourceX consider outside the U.S., Canada and the EU?

Nine markets beyond those three: Latin America, the UAE and Saudi Arabia, plus the UK, Australia, New Zealand, India, Singapore and South Africa, provided the offered data is at least 90 percent English when measured on the dataset. Each dataset still has to clear rights, privacy, quality and buyer-fit review.

What should an investor ask a data company that claims international supply?

Ask for the official frame behind each country, the measured English share on offered records, pass rates for rights and transfer review, and how many new-country datasets reached signed terms against a stated denominator. SourceX shares its company metrics with investors directly.

Sources

  • Eurostat (9 December 2025, preliminary 2024 data): 55,000 large enterprises with more than 249 people employed and 251,000 medium-sized enterprises with 50 to 249 in the EU business economy. Source
  • UK Business population estimates 2025 (DBT, published 2 October 2025, official statistics in development): 38,435 medium-sized businesses with 50 to 249 employees and 8,335 large businesses at the start of 2025. Source
  • ABS Counts of Australian Businesses, release covering July 2022 to June 2026: 5,366 businesses with 200 or more employees and 68,325 with 20 to 199. Source
  • Singapore Census of Population 2020: English was the language most frequently spoken at home for 48.3 percent of residents aged 5 and over. Source
  • Statistics South Africa, Census 2022: English was the language most often spoken in the household for 8.7 percent of the population aged one and over. Source
  • OpenAI's data partnerships page, summarized in a search excerpt: interest in data across any language, and a partnership to improve a model's Icelandic. Source
  • SourceX's how-it-works page: qualifying companies list at least five systems; qualifying does not by itself produce an offer. Source
  • SourceX's legal framework page: contracts, notices and employee policies are reviewed; GDPR data is included only with a valid legal basis and the right transfer safeguards. Source
  • SourceX's investor page: company and commercial metrics are shared with investors directly. Source

Request SourceX investor materials

SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.

See if your company qualifies