in/out

Explore the evidence / Evidence library

Follow the evidence.

Search the transcribed tables, read every original report page, and find all datasets, methods and reuse notes.

Find a chart, dataset, or source

Evidence library

Behind the headline numbers.

Exact tables and labelled values from five Microsoft reports. Each record retains its edition, units, source page and limitations.

Microsoft research in brief · ten findings with source pages

Microsoft research in brief.

Ten findings from Microsoft’s AI Diffusion research. Adoption means the estimated share of people aged 15–64 using generative AI during the reporting period; adoption changes are in percentage points (pp).

  1. Global use is rising broadly.

    Worldwide adoption reached 18.8% in Q2 2026, up from 17.8% in Q1 and 16.3% in H2 2025. In the supplied country dataset, 145 of 147 economies recorded an increase from Q1 to Q2.

    Q2 2026 · p. 5 ↗Country data ↓

  2. The UAE and Singapore lead adoption.

    The United Arab Emirates (73.3%) and Singapore (64.3%) have the highest Q2 2026 user shares. Ireland follows at 49.9%, then France at 49.6% and Norway at 49.4%.

    Q2 2026 · p. 13 ↗

  3. South Korea is gaining fastest.

    The largest Q1–Q2 2026 increases were South Korea (+3.5 pp), the UAE (+3.2 pp) and Japan (+2.2 pp), reaching 40.6%, 73.3% and 24.7% respectively. South Korea also leads growth since H1 2025, at +14.7 pp.

    Country data · calculated changes ↓

  4. Some economies remain near 6% adoption.

    Q2 2026 adoption is lowest in Cambodia (6.1%), followed by Afghanistan, Tajikistan and Turkmenistan (6.3% each). The weakest quarterly growth was in Papua New Guinea and Syria, the only declines in the CSV: −0.1 pp each, to 7.6% and 7.4%.

    Q2 2026 · p. 16 ↗

  5. The North–South divide is widening.

    In Q2 2026, adoption reached 28.8% in the Global North and 16.2% in the Global South. The gap grew from 10.6 pp in H2 2025 to 12.6 pp, as the North added more percentage points in both 2026 quarters.

    Q2 2026 · p. 5 ↗

  6. Platform choice has a geopolitical pattern.

    The H2 2025 report describes strong DeepSeek uptake in China, Russia, Iran, Cuba and Belarus, alongside parts of Africa. Microsoft connects its reach with restrictions on access to U.S. services, affordability and distribution partnerships. This is the report’s interpretation of observed patterns.

    H2 2025 · pp. 10–11 ↗

  7. U.S. metropolitan adoption is twice rural adoption.

    For Q1 2026, the U.S. report estimates average AI user shares of 32.9% in metropolitan counties, 21.8% in micropolitan counties and 16.2% in rural counties. These local estimates expose a substantial geographic divide within one country.

    U.S. Q1 2026 · p. 5 ↗

  8. Young-adult communities stand out.

    U.S. counties where more than 10% of residents are ages 18–24 average 28.6% AI user share, versus 20.3% elsewhere. The report associates younger populations and university communities with higher use; these county-level comparisons do not establish causation.

    U.S. Q1 2026 · p. 2 ↗

  9. Access barriers extend beyond AI tools.

    The Q1 2026 report compares internet access of 90.1% in the Global North with 65.7% in the Global South, using separate connectivity indicators. The H1 2025 research also identifies limited local-language support as a barrier, alongside electricity, connectivity and digital skills.

    Q1 2026 · p. 5 ↗H1 2025 · p. 13 ↗

  10. Learning has a larger role in Global South use.

    Among the consumer Copilot conversations studied, self-directed learning’s share was 29.2% higher in the Global South than globally; schoolwork’s share was 32.3% higher. These are relative differences, not pp, and describe the sampled early adopters rather than all AI users.

    Q2 2026 · pp. 5–6 ↗

Country comparisons follow the supplied CSV; report vintages and methods are documented below. Half-year and quarterly reporting windows differ. Adoption measures usage, not productivity or proficiency. Read the methods and limitations ↓

65 of 65 records

Download all records · JSON
01Historical economic comparison in report proseGlobal · H1 2025 · page 4 · 2 rows

Time interval for these quoted growth rates is not specified in paragraph. Report also says both economies had about $2,000 per capita income in 1960 (2025 dollars), and South Korea doubled living standards every eleven years. The GDP chart’s exact annual series is not printed.

Read source page · Open original PDF

Historical economic comparison in report prose
EconomyAnnual economic growth (%)
South Korea6.2
Philippines1.8
02Population and prerequisites for AI accessGlobal · H1 2025 · page 7 · 5 rows

Exact labels on source pyramid. “Datacenters” is an additional qualitative layer with no numeric count. Source references population 2024, electricity, internet 2024, and digital-literacy data.

Read source page · Open original PDF

Population and prerequisites for AI access
MeasureBillion people
World population8.1
With electricity7.4
Access to internet5.5
Basic digital skills4.2
AI users1.2
03Sub-Saharan Africa share of global population without electricityGlobal · H1 2025 · page 8 · 2 rows

Exact percentages in prose; report says over 750 million people globally lack electricity and 18 of the 20 countries with largest electricity-access deficits are in Sub-Saharan Africa.

Read source page · Open original PDF

Sub-Saharan Africa share of global population without electricity
PeriodShare (%)
201050
Report time85
04Zambia: connectivity and AI adoptionGlobal · H1 2025 · page 10 · 2 rows

Rounded narrative values. H1 appendix gives national diffusion 11.7%. Distinct denominators: not directly comparable counts.

Read source page · Open original PDF

Zambia: connectivity and AI adoption
Population denominatorAI adoption (%)
National population12
People with internet access34
05English representation in web content and populationGlobal · H1 2025 · page 12 · 2 rows

Rounded narrative shares (“half” and “just 5%”). Different denominators; these are not parts of a common whole.

Read source page · Open original PDF

English representation in web content and population
MeasureShare (%)
Open web content in English50
Native English speakers in world population5
06Further language-access countsGlobal · H1 2025 · page 12 · 2 rows

Counts and bounds printed in the report. The detailed language shares in the treemap are not numerically labelled.

Read source page · Open original PDF

Further language-access counts
MetricValue as stated
Malawi population speaking English fluentlyFewer than 4%
Other spoken languages in web-content illustration7,000+
07Language barrier: numerical claimsGlobal · H1 2025 · page 13 · 6 rows

Narrative bounds retained; no original detailed data table is supplied.

Read source page · Open original PDF

Language barrier: numerical claims
MetricValue as stated
Swahili speakersOver 200 million
German/Swahili digital content disparitySwahili has over 500 times less content
AI adoption in low-resource language countries20% lower, controlling for GDP/connectivity
LLM accuracy in EnglishAround 80%
LLM accuracy for some low-resource languages such as YorubaBelow 55%
Yoruba speakers across AfricaMore than 50 million
08Open model ecosystem claimGlobal · H1 2025 · page 15 · 1 row

Narrative claim attributed by report to VentureBeat reference 29.

Read source page · Open original PDF

Open model ecosystem claim
MetricValue as stated
LLaMA community variant modelsOver 7,000 within months of debut
09AI frontier index and months to frontierGlobal · H1 2025 · page 16 · 7 rows

Visually transcribed from original table. Frontier index is normalized; time lag is in months. Original prose rounds Israel lag to 11 months or less than a year.

Read source page · Open original PDF

AI frontier index and months to frontier
CountryBest modelFrontier indexMonths to frontier
United StatesGPT-5 (high)1.00.0
ChinaDeepSeek V3.1 Terminus (Reasoning)0.8415.3
South KoreaEXAONE 4.0 32B (Reasoning)0.8245.9
FranceMagistral Medium 1.20.7897.0
United KingdomGemma 3 27B Instruct0.7687.7
CanadaCommand A0.7677.8
IsraelJamba 1.7 Large0.65111.6
10Installed data center capacity, mid-2025Global · H1 2025 · page 17 · 14 rows

Visually transcribed original table. Report describes IEA estimates. Region labels preserved; Brazil and Central/South America both appear as printed. The prose claim “US and China 86%” is not supported by treating these rows as disjoint components.

Read source page · Open original PDF

Installed data center capacity, mid-2025
IEA regionInstalled capacity (GW)
United States53.7
China31.9
European Union11.9
Japan and South Korea6.9
India3.5
Other Asia Pacific3.1
United Kingdom2.6
Australia and New Zealand1.6
Other North America1.5
Africa1.5
Central and South America1.4
Eurasia1.2
Middle East1.1
Brazil0.6
11Additional quantitative claimsGlobal · H1 2025 · page 18 · 8 rows

Rounded report claims; broad-region exact figures appear in later H2 report. Capacity percentage conflicts with table if its regions are summed. Same claims recur on pages 2–3.

Read source page · Open original PDF

Additional quantitative claims
MetricValue as stated
AI usersMore than 1.2 billion in under 36 months
People lacking basic AI-access capabilitiesAbout four billion / nearly half of humanity
AI adoption in Global NorthApproximately 23%
AI adoption in Global SouthApproximately 13%
GDP per capita threshold with largest adoption disparitiesBelow $20,000
US and China share of global computeRoughly 86%
Countries hosting frontier-level modelsSeven
Windows devices underlying telemetryOver one billion
12AI diffusion data sourceGlobal · H1 2025 · page 19 · 74 rows

Values visually verified against original H1 appendix and match H1 column in H2 appendix. Rows sorted by printed diffusion; source does not explicitly assign ranks and equal values need not have unique ranks.

Read source page · Open original PDF

AI diffusion data source
EconomyH1 2025 AI diffusion (%)
United Arab Emirates59.4
Singapore58.6
Norway45.3
Ireland41.7
France40.9
Spain39.7
New Zealand37.6
United Kingdom36.4
Netherlands36.3
Qatar35.7
Australia34.5
Israel33.9
Belgium33.5
Canada33.5
Switzerland32.4
Sweden31.2
Austria29.1
Hungary27.9
Denmark26.6
Germany26.5
Poland26.4
Taiwan26.4
United States26.3
Czechia26.0
South Korea25.9
Italy25.8
Finland25.6
Bulgaria25.4
Jordan25.4
Costa Rica25.1
Lebanon24.8
Slovenia24.6
Saudi Arabia23.7
Oman22.6
Portugal22.4
Jamaica22.2
Slovakia22.1
Dominican Republic22.0
Croatia21.8
Vietnam21.2
Lithuania21.0
Uruguay20.9
Colombia20.4
Panama20.3
Serbia19.7
Chile19.6
South Africa19.3
Malaysia18.3
Bosnia and Herzegovina18.2
Argentina17.8
Kuwait17.7
Greece17.7
Georgia17.3
Philippines17.1
Ecuador17.0
Japan16.7
Mexico16.7
Moldova16.6
Albania15.8
Brazil15.6
China15.4
Romania15.3
El Salvador14.6
India14.2
Azerbaijan14.2
Guatemala13.7
Peru13.4
Türkiye13.4
Namibia13.0
Botswana12.8
Libya12.7
Kazakhstan12.7
Mongolia12.6
Egypt12.5
13AI diffusion data sourceGlobal · H1 2025 · page 20 · 73 rows

Values visually verified against original H1 appendix and match H1 column in H2 appendix. Rows sorted by printed diffusion; source does not explicitly assign ranks and equal values need not have unique ranks.

Read source page · Open original PDF

AI diffusion data source
EconomyH1 2025 AI diffusion (%)
Honduras12.4
Senegal12.4
Gabon12.3
Nepal12.3
Tunisia12.3
Indonesia11.7
Zambia11.7
Algeria11.3
Bolivia10.9
Côte d’Ivoire10.8
Gambia10.6
Morocco10.5
Iraq10.3
Paraguay10.1
Nicaragua10.0
Pakistan9.7
Iran9.6
Thailand9.1
Ukraine9.1
Angola8.9
Madagascar8.9
Malawi8.9
Mozambique8.9
Lesotho8.8
Benin8.7
Burkina Faso8.7
Ghana8.7
Guinea8.7
Guinea-Bissau8.7
Liberia8.7
Mali8.7
Mauritania8.7
Niger8.7
Nigeria8.7
Sierra Leone8.7
Togo8.7
Myanmar8.4
French Guiana8.3
Guyana8.3
Suriname8.3
Venezuela8.3
Kenya7.8
Belarus7.6
Kyrgyzstan7.6
Russia7.6
Papua New Guinea7.2
Haiti7.1
Cameroon7.0
Central African Republic7.0
Chad7.0
Congo7.0
Congo (DRC)7.0
Zimbabwe6.9
Syria6.7
Bangladesh6.5
Burundi6.4
Eritrea6.4
Ethiopia6.4
Somalia6.4
South Sudan6.4
Sudan6.4
Tanzania6.4
Uganda6.4
Armenia6.2
Sri Lanka6.2
Laos6.0
Rwanda6.0
Uzbekistan5.7
Cuba5.7
Afghanistan5.1
Tajikistan5.1
Turkmenistan5.1
Cambodia4.6
14AI diffusion by broad regionGlobal · H2 2025 · page 2 · 3 rows

Also plotted on page 5 for North and South.

Read source page · Open original PDF

AI diffusion by broad region
RegionH1 2025 (%)H2 2025 (%)Change (pp)
Global North22.924.71.8
Global South13.114.11.0
World15.116.31.2
15Top 30 diffusion over timeGlobal · H2 2025 · page 4 · 30 rows

Distinct rank and change labels from page 4. Source calls Czechia “Czech Republic” here. Chart changes differ from appendix for UAE, Switzerland, Sweden, Austria and United States. Do not silently merge change labels.

Read source page · Open original PDF

Top 30 diffusion over time
EconomyH1 rankH2 rankH1 diffusion (%)H2 diffusion (%)Chart change (pp)
United Arab Emirates1159.464.04.6
Singapore2258.660.92.3
Norway3345.346.41.1
Ireland4441.744.62.9
France5540.944.03.1
Spain6639.741.82.1
New Zealand7737.640.52.9
Netherlands9836.338.92.6
United Kingdom8936.438.92.5
Qatar101035.738.32.6
Australia111134.536.92.4
Israel121233.936.12.2
Belgium141333.536.02.5
Canada131433.535.01.5
Switzerland151532.434.82.4
Sweden161631.233.32.1
Austria171729.131.42.3
South Korea251825.930.74.8
Hungary181927.929.81.9
Denmark192026.628.72.1
Germany202126.528.62.1
Poland212226.428.52.1
Taiwan222326.428.42.0
United States232426.328.32.0
Czechia242526.027.81.8
Italy262625.827.82.0
Bulgaria282725.427.31.9
Finland272825.627.31.7
Jordan292925.427.01.6
Costa Rica303025.126.51.4
16Global North–South adoption gapGlobal · H2 2025 · page 5 · 2 rows

As printed. Calculating from rounded H1 endpoints 22.9−13.1 gives 9.8.

Read source page · Open original PDF

Global North–South adoption gap
PeriodGap (percentage points)
H1 20259.8
H2 202510.6
17Ten economies with largest published gainsGlobal · H2 2025 · page 6 · 10 rows

Source table prints UAE change as 4.50%, whereas 64.0−59.4 = 4.6 pp and page 4 prints 4.6%.

Read source page · Open original PDF

Ten economies with largest published gains
EconomyH1 2025 (%)H2 2025 (%)Published change (pp)
South Korea25.930.74.8
United Arab Emirates59.464.04.5
France40.944.03.1
New Zealand37.640.52.9
Ireland41.744.62.9
Qatar35.738.32.6
Netherlands36.338.92.6
United Kingdom36.438.92.5
Saudi Arabia23.726.22.5
Belgium33.536.02.5
18AI user base growth: labelled South Korea barGlobal · H2 2025 · page 7 · 1 row

Chart title says “H1 2025 to H2 2025”, but adjacent prose discusses growth since October 2024. Other grey bars have no labels. Do not interpret this as the change between the report’s H1/H2 diffusion rates.

Read source page · Open original PDF

AI user base growth: labelled South Korea bar
EconomyGrowth (%)
South Korea81.4
19Growth since October 2024: narrative comparisonsGlobal · H2 2025 · page 7 · 3 rows

Period from prose; growth metric is user base, distinct from diffusion percentage-point change. Chart calls South Korea 81.4% but has a conflicting period label.

Read source page · Open original PDF

Growth since October 2024: narrative comparisons
EconomyGrowth as stated
South KoreaMore than 80%
World35%
United States25%
20Korean SAT / CSAT benchmark scoreGlobal · H2 2025 · page 8 · 3 rows

Exact scores from prose, not percentiles. Source cites Korean SAT LLM leaderboard.

Read source page · Open original PDF

Korean SAT / CSAT benchmark score
ModelScore
GPT-3.516
GPT-4o75
GPT-5100
21English and Korean benchmark percentilesGlobal · H2 2025 · page 8 · 6 rows

Ranges/bounds preserved. Original chart incorrectly repeats title “AI User Share in the Global South and Global North”; benchmark content and adjacent text identify the intended subject.

Read source page · Open original PDF

English and Korean benchmark percentiles
ModelLanguageReported percentile
GPT-3.5English~70–80th
GPT-3.5Korean<5th
GPT-4oEnglish~90–95th
GPT-4oKorean~70th
GPT-5English>95th
GPT-5Korean>95th
22AI trust in 2025Global · H2 2025 · page 9 · 3 rows

Approximate figures in report prose attributed to 2025 Edelman Trust Barometer. Report calls UAE–US difference 35 pp.

Read source page · Open original PDF

AI trust in 2025
PopulationTrust in AI (%)
United Arab Emirates67
United States32
Western European nations average32
23DeepSeek market share: explicitly labelled economiesGlobal · H2 2025 · page 10 · 4 rows

Only four printed numeric labels. The map additionally encodes bins (≤5%, 6–10%, 11–15%, 16–20%, 21–25%, >25%); do not infer exact shares for other countries from colour.

Read source page · Open original PDF

DeepSeek market share: explicitly labelled economies
EconomyDeepSeek market share (%)
China89
Belarus56
Cuba49
Russia43
24DeepSeek usage in Africa relative to other regionsGlobal · H2 2025 · page 10 · 1 row

Estimated range stated in prose. No constituent country values are printed.

Read source page · Open original PDF

DeepSeek usage in Africa relative to other regions
ComparisonReported multiplier
Africa relative to other regions2–4×
25AI diffusion appendixGlobal · H2 2025 · page 14 · 40 rows

Original vintage. Published change labels transcribed as printed; they may differ from differences calculated from rounded endpoint values. Source prints percentage signs for change.

Read source page · Open original PDF

AI diffusion appendix
EconomyH1 2025 (%)H2 2025 (%)Published change (pp)
United Arab Emirates59.464.04.5
Singapore58.660.92.3
Norway45.346.41.1
Ireland41.744.62.9
France40.944.03.1
Spain39.741.82.1
New Zealand37.640.52.9
Netherlands36.338.92.6
United Kingdom36.438.92.5
Qatar35.738.32.6
Australia34.536.92.4
Israel33.936.12.2
Belgium33.536.02.5
Canada33.535.01.5
Switzerland32.434.82.5
Sweden31.233.32.2
Austria29.131.42.2
South Korea25.930.74.8
Hungary27.929.81.9
Denmark26.628.72.1
Germany26.528.62.1
Poland26.428.52.1
Taiwan26.428.42.0
United States26.328.32.1
Czechia26.027.81.8
Italy25.827.82.0
Bulgaria25.427.31.9
Finland25.627.31.7
Jordan25.427.01.6
Costa Rica25.126.51.4
Slovenia24.626.52.0
Saudi Arabia23.726.22.5
Lebanon24.825.70.9
Oman22.624.21.6
Portugal22.424.21.8
Slovakia22.123.81.7
Croatia21.823.71.9
Vietnam21.223.52.3
Dominican Republic22.022.70.8
Uruguay20.922.51.6
26AI diffusion appendixGlobal · H2 2025 · page 15 · 40 rows

Original vintage. Published change labels transcribed as printed; they may differ from differences calculated from rounded endpoint values. Source prints percentage signs for change.

Read source page · Open original PDF

AI diffusion appendix
EconomyH1 2025 (%)H2 2025 (%)Published change (pp)
Lithuania21.022.41.3
Jamaica22.222.1-0.1
Colombia20.422.01.6
Panama20.321.51.2
Serbia19.721.51.8
South Africa19.321.11.8
Chile19.620.81.2
Malaysia18.319.71.4
Argentina17.819.61.8
Bosnia And Herzegovina18.219.51.3
Kuwait17.719.11.4
Greece17.719.11.4
Japan16.719.12.4
Philippines17.118.31.2
Georgia17.318.20.9
Mexico16.717.81.1
Ecuador17.017.70.8
Brazil15.617.11.5
Moldova16.617.00.4
Albania15.816.50.7
China15.416.30.9
Romania15.316.20.9
El Salvador14.616.21.6
India14.215.71.4
Azerbaijan14.215.51.3
Guatemala13.714.81.1
Peru13.414.71.2
Türkiye13.414.61.2
Mongolia12.614.31.7
Namibia13.013.80.9
Libya12.713.71.1
Kazakhstan12.713.71.1
Botswana12.813.70.9
Gabon12.313.41.1
Egypt12.513.40.9
Honduras12.413.10.7
Nepal12.313.00.8
Senegal12.412.90.5
Indonesia11.712.71.1
Tunisia12.312.70.4
27AI diffusion appendixGlobal · H2 2025 · page 16 · 40 rows

Original vintage. Published change labels transcribed as printed; they may differ from differences calculated from rounded endpoint values. Source prints percentage signs for change.

Read source page · Open original PDF

AI diffusion appendix
EconomyH1 2025 (%)H2 2025 (%)Published change (pp)
Zambia11.712.30.5
Algeria11.312.00.8
Cote D'Ivoire10.811.70.8
Bolivia10.911.60.7
Iraq10.311.20.9
Paraguay10.111.00.9
Morocco10.510.90.3
Gambia10.610.90.2
Thailand9.110.71.6
Nicaragua10.010.70.7
Iran9.610.71.1
Pakistan9.710.30.7
Angola8.99.70.8
Madagascar8.99.70.8
Malawi8.99.70.8
Mozambique8.99.70.8
Benin8.79.30.6
Burkina Faso8.79.30.6
Ghana8.79.30.6
Guinea8.79.30.6
Guinea-Bissau8.79.30.6
Liberia8.79.30.6
Mali8.79.30.6
Mauritania8.79.30.6
Niger8.79.30.6
Nigeria8.79.30.6
Sierra Leone8.79.30.6
Togo8.79.30.6
Lesotho8.89.10.4
Myanmar8.49.10.7
Ukraine9.19.0-0.1
French Guiana8.39.00.7
Guyana8.39.00.7
Suriname8.39.00.7
Venezuela8.39.00.7
Belarus7.68.40.8
Kyrgyzstan7.68.20.7
Kenya7.88.10.3
Russia7.68.00.4
Cameroon7.07.80.7
28AI diffusion appendixGlobal · H2 2025 · page 17 · 27 rows

Original vintage. Published change labels transcribed as printed; they may differ from differences calculated from rounded endpoint values. Source prints percentage signs for change.

Read source page · Open original PDF

AI diffusion appendix
EconomyH1 2025 (%)H2 2025 (%)Published change (pp)
Central African Republic7.07.80.7
Chad7.07.80.7
Congo7.07.80.7
Congo (DRC)7.07.80.7
Haiti7.17.60.5
Zimbabwe6.97.60.6
Papua New Guinea7.27.30.2
Syria6.77.10.4
Bangladesh6.57.10.6
Burundi6.46.80.4
Eritrea6.46.80.4
Ethiopia6.46.80.4
Somalia6.46.80.4
South Sudan6.46.80.4
Sudan6.46.80.4
Tanzania6.46.80.4
Uganda6.46.80.4
Laos6.06.70.8
Armenia6.26.60.4
Sri Lanka6.26.60.4
Uzbekistan5.76.30.6
Rwanda6.06.30.2
Cuba5.76.10.4
Afghanistan5.15.60.4
Tajikistan5.15.60.4
Turkmenistan5.15.60.4
Cambodia4.65.10.5
29Economies above 30% AI diffusionGlobal · Q1 2026 · page 2 · 1 row

Q1 2026.

Read source page · Open original PDF

Economies above 30% AI diffusion
MeasureValue
Economies exceeding 30% of working-age population using AI26
30Global AI diffusion through Q1 2026Global · Q1 2026 · page 5 · 4 rows

Gap is percentage points, although the Q1 report prints percent signs. Q1 change compares Q1 2026 with H2 2025.

Read source page · Open original PDF

Global AI diffusion through Q1 2026
RegionH1 2025 (%)H2 2025 (%)Q1 2026 (%)Q1 change (pp)
Global North22.924.727.52.8
Global South13.114.115.41.3
World15.116.317.81.5
North–South gap9.810.612.1—
31Foundations for AI accessGlobal · Q1 2026 · page 5 · 2 rows

The report cites World Bank Population 2024, Our World in Data electricity access, ITU Internet use 2024 and ITU information/data-literacy skills. Different indicators and populations; do not interpret as a single conversion funnel.

Read source page · Open original PDF

Foundations for AI access
RegionAI user share (%)Digital skills (%)Internet access (%)Electricity access (%)Population baseline (%)
Global North27.570.190.198.1100
Global South15.448.265.788.9100
32Fastest growing economies since June 2025Global · Q1 2026 · page 6 · 15 rows

Relative growth Q1 2026 versus H1 2025, as labeled in the chart; values may differ from ratios computed on rounded appendix values.

Read source page · Open original PDF

Fastest growing economies since June 2025
EconomyIncrease in AI user share (%)
South Korea43.2
Thailand36.4
Japan34.1
Mongolia32.3
Iran31.4
Laos31.2
Turkey30.3
Belarus27.6
El Salvador25.7
Kazakhstan25.6
Kyrgyzstan25.5
Uzbekistan25.5
Vietnam24.9
Cambodia24.8
Russia24.6
33Japan adoption and Japanese exam performanceGlobal · Q1 2026 · page 8 · 5 rows

Report narrative says 3.4pp growth is more than three times global average; its table reports global change 1.5 pp, so that narrative comparison is internally inconsistent. Exam endpoints summarize different cited studies rather than a harmonized series.

Read source page · Open original PDF

Japan adoption and Japanese exam performance
MeasureValueUnit
Japan rank H1 202556rank
Japan rank Q1 202648rank
Japan Q1 adoption increase3.4pp
Earlier Japanese professional-exam accuracy50.8% (approximately)
Recent Japanese professional-exam accuracy lower bound90% (over)
34English and Japanese benchmark performanceGlobal · Q1 2026 · page 9 · 4 rows

Visually verified color mapping: English dark blue, Japanese light blue. MMLU and MMLU-Pro are different benchmarks and should be charted separately.

Read source page · Open original PDF

English and Japanese benchmark performance
BenchmarkModelReleaseEnglish accuracy (%)Japanese accuracy (%)
MMLUGPT-3.5 TurboMar 20237050
MMLUGPT-4oAug 20248980
MMLU-ProGPT-4oAug 20246975
MMLU-ProGPT-5Aug 20258587
35Year-over-year growth in Git pushesGlobal · Q1 2026 · page 9 · 2 rows

Source named by report: GitHub. Year-over-year growth during Q1 2026.

Read source page · Open original PDF

Year-over-year growth in Git pushes
GeographyYear-over-year growth (%)
Japan129
Global78
36MMLU: English and Japanese accuracy (%)Global · Q1 2026 · page 9 · 2 rows

A separate benchmark; scores should not be compared across MMLU and MMLU-Pro. Duplicate numeric view of the combined benchmark table.

Read source page · Open original PDF

MMLU: English and Japanese accuracy (%)
ModelReleaseEnglish (%)Japanese (%)
GPT-3.5 TurboMar 20237050
GPT-4oAug 20248980
37MMLU-Pro: English and Japanese accuracy (%)Global · Q1 2026 · page 9 · 2 rows

A separate benchmark; scores should not be compared across MMLU and MMLU-Pro. Duplicate numeric view of the combined benchmark table.

Read source page · Open original PDF

MMLU-Pro: English and Japanese accuracy (%)
ModelReleaseEnglish (%)Japanese (%)
GPT-4oAug 20246975
GPT-5Aug 20258587
38Global Git pushes: labeled endpointsGlobal · Q1 2026 · page 10 · 2 rows

Chart annotation reports +78% YoY. Unlabeled historical line is preserved in the source page; these are the only explicitly labeled values.

Read source page · Open original PDF

Global Git pushes: labeled endpoints
PeriodGit pushes (millions)
Q1 2025213
Q1 2026380
39GPT-5.3-Codex benchmark claimsGlobal · Q1 2026 · page 10 · 2 rows

Approximately/around values in report prose; specific benchmark versions not provided in this paragraph.

Read source page · Open original PDF

GPT-5.3-Codex benchmark claims
BenchmarkAccuracy (%)
Terminal-Bench77
OSWorld65
40New GitHub repositories: labeled endpointsGlobal · Q1 2026 · page 11 · 2 rows

Chart annotation reports +45% YoY. Historical line is not numerically labeled.

Read source page · Open original PDF

New GitHub repositories: labeled endpoints
PeriodNew repositories (millions)
Q1 202514.7
Q1 202621.3
41GitHub pull requests associated with AI agents: labeled endpointsGlobal · Q1 2026 · page 12 · 2 rows

Chart annotation reports 28× in 10 months. Narrative page 11 says since June 2025, conflicting with chart May 2025. The full historical line has no exact labels.

Read source page · Open original PDF

GitHub pull requests associated with AI agents: labeled endpoints
PeriodAgentic pull requests (millions)
May 20250.083
Mar 20262.3
42US software developer employmentGlobal · Q1 2026 · page 12 · 3 rows

Reported from BLS CPS. These measures do not establish causal effects of AI on employment.

Read source page · Open original PDF

US software developer employment
MeasureValueUnitPeriod
Employment2.2million (approximately)2025
Employment year-over-year growth8.5%2025
Employment year-over-year growth4% (about)March 2026
43Economy appendix, page 17Global · Q1 2026 · page 17 · 40 rows

Original report snapshot; compare with CSV vintages rather than silently overwriting prior values.

Read source page · Open original PDF

Economy appendix, page 17
EconomyH1 2025 (%)H2 2025 (%)Q1 2026 (%)Q1 change (pp)
United Arab Emirates59.464.070.16.1
Singapore58.660.963.42.5
Norway45.346.448.62.2
Ireland41.744.648.43.8
France40.944.047.83.8
Spain39.741.844.22.4
New Zealand37.640.543.02.5
United Kingdom36.438.942.23.3
Netherlands36.338.942.13.2
Qatar35.738.341.83.5
Australia34.536.939.52.6
Belgium33.536.039.03.0
Israel33.936.138.12.0
Switzerland32.434.837.83.0
Canada33.535.037.32.3
South Korea25.930.737.16.4
Sweden31.233.336.12.8
Austria29.131.434.12.7
Hungary27.929.832.22.4
Taiwan26.428.431.83.4
United States26.328.331.33.0
Denmark26.628.731.22.5
Germany26.528.631.12.5
Poland26.428.531.02.5
Italy25.827.830.22.4
Czech Republic26.027.830.12.3
Jordan25.427.029.72.7
Bulgaria25.427.329.72.4
Finland25.627.329.52.2
Saudi Arabia23.726.229.43.2
Slovenia24.626.529.02.5
Costa Rica25.126.528.52.0
Lebanon24.825.727.31.6
Vietnam21.223.526.53.0
Oman22.624.226.52.3
Portugal22.424.226.42.2
Croatia21.823.726.12.4
Slovak Republic22.123.826.12.3
Dominican Republic22.022.724.82.1
Uruguay20.922.524.62.1
44Economy appendix, page 18Global · Q1 2026 · page 18 · 40 rows

Original report snapshot; compare with CSV vintages rather than silently overwriting prior values.

Read source page · Open original PDF

Economy appendix, page 18
EconomyH1 2025 (%)H2 2025 (%)Q1 2026 (%)Q1 change (pp)
Colombia20.422.024.52.5
Lithuania21.022.424.31.9
Serbia19.721.524.12.6
Jamaica22.222.124.01.9
Panama20.321.523.31.8
South Africa19.321.123.12.0
Chile19.620.822.71.9
Japan16.719.122.53.4
Bosnia And Herzegovina18.219.522.12.6
Argentina17.819.621.92.3
Malaysia18.319.721.82.1
Kuwait17.719.121.12.0
Greece17.719.120.81.7
Georgia17.318.220.52.3
Mexico16.717.820.12.3
Philippines17.118.320.11.8
Ecuador17.017.719.51.8
Brazil15.617.119.12.0
Albania15.816.518.52.0
Moldova16.617.018.51.5
El Salvador14.616.218.32.1
Azerbaijan14.215.517.72.2
India14.215.717.61.9
Romania15.316.217.51.3
Turkey13.414.617.42.8
Mongolia12.614.316.72.4
Guatemala13.714.816.41.6
Peru13.414.716.41.7
China15.416.316.40.1
Kazakhstan12.713.715.92.2
Namibia13.013.815.11.3
Gabon12.313.415.01.6
Libya12.713.715.01.3
Egypt12.513.414.81.4
Botswana12.813.714.81.1
Nepal12.313.014.21.2
Indonesia11.712.714.11.4
Honduras12.413.114.00.9
Senegal12.412.913.91.0
Tunisia12.312.713.50.8
45Economy appendix, page 19Global · Q1 2026 · page 19 · 40 rows

Original report snapshot; compare with CSV vintages rather than silently overwriting prior values.

Read source page · Open original PDF

Economy appendix, page 19
EconomyH1 2025 (%)H2 2025 (%)Q1 2026 (%)Q1 change (pp)
Algeria11.312.013.21.2
Zambia11.712.313.10.8
Cote D'Ivoire10.811.713.11.4
Bolivia10.911.612.71.1
Iran9.610.712.61.9
Iraq10.311.212.51.3
Thailand9.110.712.41.7
Paraguay10.111.012.21.2
Nicaragua10.010.711.81.1
Morocco10.510.911.70.8
Gambia10.610.911.40.5
Pakistan9.710.311.41.1
Angola8.99.710.91.2
Madagascar8.99.710.91.2
Malawi8.99.710.91.2
Mozambique8.99.710.91.2
French Guiana8.39.010.31.3
Guyana8.39.010.31.3
Suriname8.39.010.31.3
Venezuela8.39.010.31.3
Benin8.79.310.10.8
Burkina Faso8.79.310.10.8
Ghana8.79.310.10.8
Guinea8.79.310.10.8
Guinea-Bissau8.79.310.10.8
Liberia8.79.310.10.8
Mali8.79.310.10.8
Mauritania8.79.310.10.8
Niger8.79.310.10.8
Nigeria8.79.310.10.8
Sierra Leone8.79.310.10.8
Togo8.79.310.10.8
Myanmar8.49.110.00.9
Lesotho8.89.19.80.7
Belarus7.68.49.61.2
Kyrgyzstan7.68.29.51.3
Russia7.68.09.51.5
Ukraine9.19.09.40.4
Kenya7.88.18.70.6
Cameroon7.07.88.70.9
46Economy appendix, page 20Global · Q1 2026 · page 20 · 27 rows

Original report snapshot; compare with CSV vintages rather than silently overwriting prior values.

Read source page · Open original PDF

Economy appendix, page 20
EconomyH1 2025 (%)H2 2025 (%)Q1 2026 (%)Q1 change (pp)
Central African Republic7.07.88.70.9
Chad7.07.88.70.9
Congo7.07.88.70.9
Democratic Republic Of The Congo7.07.88.70.9
Zimbabwe6.97.68.50.9
Haiti7.17.68.50.9
Laos6.06.77.81.1
Bangladesh6.57.17.80.7
Papua New Guinea7.27.37.70.4
Burundi6.46.87.60.8
Eritrea6.46.87.60.8
Ethiopia6.46.87.60.8
Somalia6.46.87.60.8
South Sudan6.46.87.60.8
Sudan6.46.87.60.8
Tanzania6.46.87.60.8
Uganda6.46.87.60.8
Syria6.77.17.50.4
Armenia6.26.67.40.8
Sri Lanka6.26.67.30.7
Rwanda6.06.37.20.9
Uzbekistan5.76.37.20.9
Cuba5.76.16.70.6
Afghanistan5.15.66.10.5
Tajikistan5.15.66.10.5
Turkmenistan5.15.66.10.5
Cambodia4.65.15.70.6
47Global AI diffusion through Q2 2026Global · Q2 2026 · page 5 · 4 rows

Gap columns use percentage points. Measure covers ages 15–64; change compares Q2 with Q1.

Read source page · Open original PDF

Global AI diffusion through Q2 2026
RegionH2 2025 (%)Q1 2026 (%)Q2 2026 (%)Q2 change (pp)
Global North24.727.528.81.3
Global South14.115.416.20.8
World16.317.818.81.0
North–South gap10.612.112.6—
48Copilot intent prevalence relative to the global conversation mixGlobal · Q2 2026 · page 5 · 8 rows

Relative differences from each intent’s global share, not percentages of conversations or percentage-point gaps. Visually verified chart color mapping. Subset of consumer Copilot conversations; early adopters, not all AI users; intents ordered by global prevalence.

Read source page · Open original PDF

Copilot intent prevalence relative to the global conversation mix
IntentGlobal North difference (%)Global South difference (%)
Searching for information6.5-14.1
Create/edit text or code-10.322.4
Feedback or advice12.6-27.5
Shopping/product research16.9-36.9
Self-directed learning-13.429.2
Create/edit images-14.932.3
Technical support3.8-8.3
Schoolwork/academic support-14.932.3
49Copilot traffic shares in the intent analysisGlobal · Q2 2026 · page 5 · 2 rows

Shares of the analyzed Copilot conversations; not population shares.

Read source page · Open original PDF

Copilot traffic shares in the intent analysis
RegionTraffic share (%)
Global North68.5
Global South31.5
50OpenRouter open-weight token usageGlobal · Q2 2026 · page 7 · 2 rows

Late 2025 stated as roughly one-third, not a precise 33.3%. Analysis refers to >100 trillion tokens. OpenRouter is developer-oriented; excludes direct-provider and local usage and is not global market share.

Read source page · Open original PDF

OpenRouter open-weight token usage
PeriodOpen-weight token share (%)
Late 2025 (roughly)33.333333333333336
2026 recent analysis75
51Artificial Analysis frontier labelsGlobal · Q2 2026 · page 7 · 2 rows

Explicit final labels in chart; full model release-date series is unlabeled. Inspect original page for exact visual context. Data through September 2026, beyond the Q2 adoption reporting period.

Read source page · Open original PDF

Artificial Analysis frontier labels
Model classIntelligence Index
Closed weights53.4
Open weights44.9
52GLM 5.3 Flash cost compared with GPT 5.6 Sol highGlobal · Q2 2026 · page 8 · 1 row

Prose says similar intelligence performance at roughly one third the cost per task. Approximate ratio, not a direct measured price; full scatter values are unlabeled.

Read source page · Open original PDF

GLM 5.3 Flash cost compared with GPT 5.6 Sol high
ComparisonRelative cost
GLM 5.3 Flash / GPT 5.6 Sol high0.3333333333333333
53Hugging Face Spaces: labeled countsGlobal · Q2 2026 · page 9 · 3 rows

Chart and prose disagree on dates. Chart says 1.8× since first report; prose says approximately 83% increase. Projects are surviving Spaces, not distinct commercial products. All historical intermediate values are unlabeled.

Read source page · Open original PDF

Hugging Face Spaces: labeled counts
PeriodCumulative surviving Spaces
End 2023 (approximately)124000
First report: chart Oct 2025 / prose Nov 2025797000
Chart Aug 2026 / prose Sep 20261460000
54Bring Your Own Language performance improvementGlobal · Q2 2026 · page 9 · 1 row

Report summary across a set of benchmarks, not exact per-language scores.

Read source page · Open original PDF

Bring Your Own Language performance improvement
MeasureImprovement (%)
Chichewa and Māori benchmark mean improvement (approximately)12
55Economy appendix, page 13Global · Q2 2026 · page 13 · 40 rows

Original report snapshot; compare with CSV vintages rather than silently overwriting prior values.

Read source page · Open original PDF

Economy appendix, page 13
EconomyQ1 2026 (%)Q2 2026 (%)Q2 change (pp)
United Arab Emirates70.173.33.2
Singapore63.464.30.9
Ireland48.449.91.5
France47.849.61.8
Norway48.649.40.8
Spain44.245.10.9
New Zealand43.044.11.1
United Kingdom42.243.71.5
Netherlands42.143.51.4
Qatar41.843.51.7
Australia39.540.71.2
South Korea37.140.63.5
Belgium39.040.31.3
Switzerland37.839.11.3
Israel38.138.70.6
Canada37.338.51.2
Sweden36.137.31.2
Austria34.135.41.3
Taiwan31.833.82.0
Hungary32.233.31.1
United States31.333.01.7
Denmark31.232.41.2
Germany31.132.31.2
Poland31.032.31.3
Saudi Arabia29.431.42.0
Italy30.231.41.2
Czech Republic30.131.21.1
Jordan29.731.21.5
Bulgaria29.730.71.0
Finland29.530.61.1
Slovenia29.030.11.1
Costa Rica28.529.71.2
Vietnam26.527.91.4
Lebanon27.327.70.4
Oman26.527.61.1
Portugal26.427.41.0
Croatia26.127.21.1
Slovak Republic26.127.11.0
Colombia24.525.91.4
Dominican Republic24.825.81.0
56Economy appendix, page 14Global · Q2 2026 · page 14 · 40 rows

Original report snapshot; compare with CSV vintages rather than silently overwriting prior values.

Read source page · Open original PDF

Economy appendix, page 14
EconomyQ1 2026 (%)Q2 2026 (%)Q2 change (pp)
Uruguay24.625.71.1
Serbia24.125.31.2
Lithuania24.325.10.8
Jamaica24.025.01.0
Japan22.524.72.2
Panama23.324.20.9
Chile22.724.01.3
South Africa23.123.90.8
Argentina21.923.41.5
Bosnia And Herzegovina22.123.21.1
Malaysia21.822.60.8
Kuwait21.122.61.5
Georgia20.521.61.1
Greece20.821.50.7
Mexico20.121.41.3
Philippines20.120.70.6
Ecuador19.520.30.8
Brazil19.120.21.1
El Salvador18.320.11.8
Albania18.519.20.7
Moldova18.519.10.6
Turkey17.418.91.5
Azerbaijan17.718.71.0
India17.618.50.9
Romania17.518.00.5
Mongolia16.717.91.2
China16.417.51.1
Peru16.417.30.9
Guatemala16.417.30.9
Kazakhstan15.916.91.0
Gabon15.015.70.7
Libya15.015.50.5
Egypt14.815.50.7
Namibia15.115.40.3
Botswana14.815.00.2
Indonesia14.114.80.7
Nepal14.214.60.4
Honduras14.014.50.5
Senegal13.914.40.5
Tunisia13.513.90.4
57Economy appendix, page 15Global · Q2 2026 · page 15 · 40 rows

Original report snapshot; compare with CSV vintages rather than silently overwriting prior values.

Read source page · Open original PDF

Economy appendix, page 15
EconomyQ1 2026 (%)Q2 2026 (%)Q2 change (pp)
Algeria13.313.80.5
Cote D'Ivoire13.113.70.6
Thailand12.413.41.0
Bolivia12.713.30.6
Zambia13.113.30.2
Iraq12.513.00.5
Iran12.612.80.2
Paraguay12.212.60.4
Nicaragua11.812.50.7
Morocco11.712.00.3
Pakistan11.411.90.5
Gambia11.511.50.0
Mozambique10.911.40.5
Angola10.911.40.5
Madagascar10.911.40.5
Malawi10.911.40.5
Guyana10.311.31.0
Suriname10.311.31.0
Venezuela10.311.31.0
French Guiana10.311.31.0
Myanmar10.010.40.4
Togo10.110.40.3
Guinea10.110.40.3
Niger10.110.40.3
Liberia10.110.40.3
Nigeria10.110.40.3
Benin10.110.40.3
Mali10.110.40.3
Ghana10.110.40.3
Guinea-Bissau10.110.40.3
Sierra Leone10.110.40.3
Mauritania10.110.40.3
Burkina Faso10.110.40.3
Belarus9.610.30.7
Kyrgyzstan9.510.00.5
Russia9.59.90.4
Lesotho9.89.90.1
Ukraine9.59.50.0
Chad8.79.10.4
Central African Republic8.79.10.4
58Economy appendix, page 16Global · Q2 2026 · page 16 · 27 rows

Original report snapshot; compare with CSV vintages rather than silently overwriting prior values.

Read source page · Open original PDF

Economy appendix, page 16
EconomyQ1 2026 (%)Q2 2026 (%)Q2 change (pp)
Cameroon8.79.10.4
Democratic Republic Of The Congo8.79.10.4
Congo8.79.10.4
Haiti8.58.90.4
Kenya8.78.90.2
Zimbabwe8.58.90.4
Laos7.88.40.6
Bangladesh7.88.00.2
Uganda7.68.00.4
Ethiopia7.68.00.4
Somalia7.68.00.4
Eritrea7.68.00.4
South Sudan7.68.00.4
Sudan7.68.00.4
Burundi7.68.00.4
Tanzania7.68.00.4
Armenia7.47.70.3
Papua New Guinea7.77.6-0.1
Sri Lanka7.37.60.3
Rwanda7.27.60.4
Uzbekistan7.27.60.4
Syria7.57.4-0.1
Cuba6.76.80.1
Tajikistan6.16.30.2
Afghanistan6.16.30.2
Turkmenistan6.16.30.2
Cambodia5.86.10.3
59College-age population and AI adoptionUnited States · Q1 2026 · page 2 · 2 rows

Narrative comparison; report does not specify weighting in this paragraph.

Read source page · Open original PDF

College-age population and AI adoption
County population share ages 18–24Average AI user share (%)
More than 10%28.6
Smaller shares20.3
60AI adoption by county urbanization classUnited States · Q1 2026 · page 5 · 3 rows

Report classifications: metropolitan generally urban core ≥50,000; micropolitan 5,000–50,000; rural outside both. Population proportions rounded.

Read source page · Open original PDF

AI adoption by county urbanization class
County classAI user share (%)Share of population (%)
Metropolitan32.987
Micropolitan21.810
Rural16.23
61Urbanization and trustUnited States · Q1 2026 · page 5 · 3 rows

The 0.17 is an association, not causation. Trust percentages come from National AI Opinion Monitor survey cited in report, not Microsoft telemetry.

Read source page · Open original PDF

Urbanization and trust
MeasureValueUnit
AI user share per 1% more urban population0.17pp
Urban respondents: AI likely acts in public interest53%
Rural respondents: AI likely acts in public interest38%
62Top 15 counties by AI user share, minimum 10,000 populationUnited States · Q1 2026 · page 7 · 15 rows

13 of 15 exceed 95th percentile of young-adult share nationally, per prose. Population cutoff matters: this is not the unrestricted ranking of all CSV counties.

Read source page · Open original PDF

Top 15 counties by AI user share, minimum 10,000 population
RankCounty-equivalent areaStateAge 18–24 share (%)AI user share (%)Anchor institution
1WilliamsburgVirginia36.273.2College of William & Mary
2HarrisonburgVirginia32.767.5James Madison University
3MadisonIdaho46.467.2Brigham Young University-Idaho
4BrazosTexas26.464Texas A&M University
5StoryIowa30.663.8Iowa State University
6DentonTexas8.960.1University of North Texas
7ClarkeGeorgia2659University of Georgia
8DouglasKansas2357.2University of Kansas
9CharlottesvilleVirginia20.151.7University of Virginia
10ClevelandOklahoma1551.6University of Oklahoma
11HaysTexas14.251.4Texas State University
12LeeAlabama16.550.8Auburn University
13MontgomeryVirginia27.850.8Virginia Tech
14UtahUtah16.850.2Brigham Young University
15FredericksburgVirginia1849.3University of Mary Washington
63Top-county contextUnited States · Q1 2026 · page 7 · 3 rows

Top 15 ranking applies to counties with at least 10,000 residents.

Read source page · Open original PDF

Top-county context
MeasureValueUnit
Top 15 counties with a college or university15count
Top 15 above 95th percentile in age 18–24 share13count
Denton technology-access national rank12rank
6435 largest metropolitan areasUnited States · Q1 2026 · page 8 · 35 rows

Prose selection criterion is working-age population above 1.5 million. Figure caption merely says population > 1.5 million; preserve the more explicit prose definition.

Read source page · Open original PDF

35 largest metropolitan areas
RankMetroStatesAI user share (%)
1WashingtonDC, VA, MD, WV39.7
2DallasTX39.4
3MiamiFL38.6
4HoustonTX38.4
5AtlantaGA38.0
6San FranciscoCA38.0
7AustinTX37.8
8SeattleWA36.8
9ChicagoIL, IN36.6
10Las VegasNV36.5
11San DiegoCA36.3
12OrlandoFL36.2
13San AntonioTX35.4
14New YorkNY, NJ35.2
15BostonMA, NH35.1
16RiversideCA35.1
17NashvilleTN34.7
18Los AngelesCA34.7
19TampaFL33.8
20BaltimoreMD33.7
21CharlotteNC, SC33.6
22PhoenixAZ33.4
23CincinnatiOH, KY, IN32.9
24DenverCO32.8
25PhiladelphiaPA, NJ, DE, MD32.6
26Kansas CityMO, KS32.4
27PortlandOR, WA31.1
28SacramentoCA29.6
29MinneapolisMN, WI29.2
30ClevelandOH29.1
31DetroitMI29.1
32IndianapolisIN28.8
33ColumbusOH28.7
34St. LouisMO, IL26.8
35PittsburghPA25.8
65State AI diffusion rankingUnited States · Q1 2026 · page 13 · 50 rows

50 states; Washington DC is not a state and excluded.

Read source page · Open original PDF

State AI diffusion ranking
RankStateAI user share (%)
1Maryland36.3
2Utah35.7
3Texas35.3
4Virginia34.7
5New Jersey34.5
6Nevada34.2
7California34.0
8Connecticut34.0
9Georgia33.7
10Florida33.6
11Massachusetts33.4
12Illinois33.3
13New York32.7
14Rhode Island32.5
15Colorado32.3
16Washington32.2
17Arizona31.4
18Hawaii30.6
19Delaware30.6
20New Hampshire30.2
21North Carolina30.1
22South Carolina29.1
23Oklahoma28.9
24Idaho28.8
25Kansas28.6
26Tennessee28.5
27Oregon28.4
28Ohio28.3
29Wisconsin28.2
30North Dakota28.2
31Michigan27.4
32South Dakota27.4
33Alabama27.3
34Pennsylvania27.2
35Indiana26.8
36Missouri26.8
37Nebraska26.4
38Minnesota26.3
39Louisiana26.1
40Arkansas26.0
41Wyoming25.5
42Kentucky25.1
43Iowa24.4
44New Mexico23.9
45Alaska23.6
46Vermont23.3
47Mississippi22.9
48Montana22.7
49Maine21.4
50West Virginia20.8
23 original charts without exact point-level data

These graphics remain available in the reader. No values have been estimated from curves, map colours or pixel heights.

  1. Real GDP per capita: Philippines vs. South Korea

    Annual curves, 1960s–2020s; no exact datapoint labels or underlying series in folder.

    Global · H1 2025 · page 4 · Read source page

  2. Household diffusion of AI and earlier technologies

    Six curves (AI, smartphone, cellphone, radio, internet, electricity), percent of U.S. households against years since introduction; no exact datapoint labels.

    Global · H1 2025 · page 5 · Read source page

  3. AI diffusion and GDP

    Country scatter against GDP per capita (USD) 2023. Diffusion values are in appendix; GDP coordinates are not printed.

    Global · H1 2025 · page 6 · Read source page

  4. Access to electricity by economy

    Choropleth bins: >80%, 60–80%, 40–60%, 20–40%, 0–20%, no data. Exact economy shares absent.

    Global · H1 2025 · page 8 · Read source page

  5. Where AI data center regions are located

    Bubble map, legend counts 1/10/20/30; individual region counts not printed.

    Global · H1 2025 · page 9 · Read source page

  6. Internet connectivity by economy

    Choropleth bins 0–20%, 20–40%, 40–60%, 60–80%, 80–100%, no data; exact country shares absent.

    Global · H1 2025 · page 10 · Read source page

  7. Global digital adoption by economy

    World Bank Digital Adoption Index choropleth with 20-point bins; exact country values absent.

    Global · H1 2025 · page 11 · Read source page

  8. Open web content by language

    Language treemap / Voronoi-style cells with labels but no numerical shares. Only narrative English 50% is recoverable.

    Global · H1 2025 · page 12 · Read source page

  9. Language coverage in AI resources

    Categorical country map: high, intermediate, low, no data. No numerical scale.

    Global · H1 2025 · page 13 · Read source page

  10. AI models by performance, cost, and release date

    Bubble plot; exact intelligence scores, dates and price values not printed. Model names printed only for selected points.

    Global · H1 2025 · page 15 · Read source page

  11. AI user base growth H1 2025 to H2 2025

    Only South Korea bar has country/value labels (81.4%). Grey bars have neither country labels nor exact values. Period title conflicts with adjoining since-October-2024 prose.

    Global · H2 2025 · page 7 · Read source page

  12. DeepSeek market share by economy

    Four exact country labels extracted. Remaining economies appear only as colour bins; not numerically recoverable from folder.

    Global · H2 2025 · page 10 · Read source page

  13. English MMLU versus14-language MMMLU over time

    Model/date points and values are not explicitly labeled. All chart marks preserved in source page; no fabricated numeric series.

    Global · Q1 2026 · page 7 · Read source page

  14. Japan and world monthly adoption, three-month average

    Monthly values and annotated model-release dates are not tabulated. Period-aggregate country CSV cannot reproduce monthly curve.

    Global · Q1 2026 · page 8 · Read source page

  15. Historical global Git pushes

    OnlyQ1 2025 andQ1 2026 endpoints labeled; intermediate monthly counts not provided.

    Global · Q1 2026 · page 10 · Read source page

  16. Historical new GitHub repositories

    OnlyQ1 2025 andQ1 2026 endpoints labeled; intermediate counts not provided.

    Global · Q1 2026 · page 11 · Read source page

  17. Historical GitHub pull requests associated with AI agents

    OnlyMay 2025 andMarch 2026 endpoints labeled; intermediate counts not provided.

    Global · Q1 2026 · page 12 · Read source page

  18. Artificial Analysis Intelligence Index by model release date

    Individual model points and frontier progression not numerically labeled; only 53.4 / 44.9 endpoint labels.

    Global · Q2 2026 · page 7 · Read source page

  19. Artificial Analysis intelligence vs cost per task

    Named points shown, but exact per-model costs and intelligence scores not labeled. Axes alone cannot establish exact values.

    Global · Q2 2026 · page 8 · Read source page

  20. Hugging Face surviving Spaces historical trajectory

    Three narrative/annotation counts extracted; all other trajectory values unlabeled.

    Global · Q2 2026 · page 9 · Read source page

  21. County AI share vs urban population share

    County AI share exists in CSV; urban shares and populations do not. Cannot faithfully reconstruct scatter from supplied tables.

    United States · Q1 2026 · page 5 · Read source page

  22. County AI share vs young-adult population share

    County AI share exists in CSV; age 18–24shares exist only for top 15table. Cannot reconstruct full scatter.

    United States · Q1 2026 · page 6 · Read source page

  23. Industry composition and AI usage estimates and95% confidence intervals

    14 industry point estimates and confidence-interval endpoints shown graphically without numeric labels. Positive categories: professional/technical services, services overall, headquarters/management, health care, information/media, finance, wholesale. Near-zero: education, utilities. Negative: manufacturing, agriculture, mining/oil/gas, construction, goods-producing. Do not infer precise coefficients from pixels.

    United States · Q1 2026 · page 9 · Read source page

US college guidance and preparation · September 2026

Dartmouth US college sample · agreement or strong agreement · September 2026
StatementShareItem sample
Mostly developed AI skills independently, outside formal coursework79.3%1,003
Professors and courses provided meaningful guidance on effective and ethical AI use68.1%1,003
Institution adequately prepares students for AI use in their future career49.5%1,003

Separate, overlapping agreement measures, not complements or an adoption rate. Weighted online-panel self-reports; working paper, peer review not verified. Design effects were not verified. No causal learning outcomes or direct comparison with HEPI. Dartmouth working paper, Table 12 (PDF p. 22) · Source and methods. The 65 Microsoft transcription records and five-report reader below retain their original scope.

UK business AI · 2026

UK current AI use, including pilots · 7 January–21 April 2026 · weighted percentages
Business sizeAI useUnweighted base
All eligible businesses32%3,503
Micro · 5–9 employees27%1,347
Small · 10–49 employees34%1,579
Medium · 50–249 employees47%447
Large · 250+ employees52%130

Skills barrier among nonusers planning adoption: 78% (unweighted n=236). Broad AI, including pilots, not only generative AI. Selected sectors and a five-employee threshold differ from Eurostat; no direct UK–EU ranking. Self-reports, not a census or causal productivity evidence. Low response creates nonresponse risk; weighting does not establish its absence. Methods and source qualifications · Six observations · CSV · Original government tables.

Reuters Institute · use and search exposure

Reuters Institute / Oxford · separate measures and country sets · weighted percentages
MeasureWaveShareRespondents
Weekly use of standalone generative AI tools — six countries202534%12,565
Weekly use of standalone generative AI tools — six countries202651%12,774
AI-generated answers or assistants encountered in search engines in the previous week — nine countries202657%18,921

Self-reported use and exposure, subject to recall and interpretation; online samples underrepresent people offline. No conventional sampling margin of error; not a worldwide estimate or directly comparable with Microsoft usage estimates. Different country sets and questions mean the weekly-use and search-exposure figures must not be subtracted as an exposure gap. Weekly-use comparison covers the same six countries, not a longitudinal panel. The prompted standalone-tool list changed from 13 tools in 2025 to 14 in 2026; this qualification accompanies the trend. Methods and source qualifications · Three observations · CSV.

OECD · regional enterprise adoption

In the selected 2025 data, the simple average regional share of enterprises using AI was 11.43% across 43 Central and Eastern European regions, compared with 26.10% across 48 other covered regions. These are unweighted means of regional rates across 12 countries, not enterprise-weighted adoption rates for Europe.

OECD analysis of existing Eurostat isoc_r_eb_ain2 data, not an independent survey. Selected 2025 regional coverage in 12 countries; not Europe-wide. Regional achieved sample counts and confidence intervals are not supplied in the workbook. Employment exposure in column D is excluded: it has a different denominator.

Enterprises using AI, 2025 · selected OECD regional coverage · percent
CountryRegionCodeAI useOECD grouping
AustriaBurgenlandAT1118.33%North, South & West Europe
AustriaCarinthiaAT2123.63%North, South & West Europe
AustriaLower AustriaAT1221.81%North, South & West Europe
AustriaSalzburgAT3231.45%North, South & West Europe
AustriaStyriaAT2225.55%North, South & West Europe
AustriaTyrolAT3330.28%North, South & West Europe
AustriaUpper AustriaAT3132.93%North, South & West Europe
AustriaViennaAT1337.36%North, South & West Europe
AustriaVorarlbergAT3436.42%North, South & West Europe
BelgiumBrussels Capital RegionBE142.32%North, South & West Europe
BelgiumFlemish RegionBE234.84%North, South & West Europe
BelgiumWalloon RegionBE330.20%North, South & West Europe
CroatiaAdriatic CroatiaHR0310.98%East Europe
CroatiaCity of ZagrebHR0525.44%East Europe
CroatiaNorthern CroatiaHR0611.98%East Europe
CroatiaPannonian CroatiaHR0210.51%East Europe
CzechiaCentral Bohemian RegionCZ0213.47%East Europe
CzechiaCentral MoraviaCZ0715.57%East Europe
CzechiaMoravia-SilesiaCZ0815.72%East Europe
CzechiaNortheastCZ0517.81%East Europe
CzechiaNorthwestCZ049.47%East Europe
CzechiaPragueCZ0125.50%East Europe
CzechiaSoutheastCZ0615.49%East Europe
CzechiaSouthwestCZ0314.06%East Europe
DenmarkCentral JutlandDK0450.82%North, South & West Europe
DenmarkCopenhagen regionDK0133.05%North, South & West Europe
DenmarkNorthern JutlandDK0532.13%North, South & West Europe
DenmarkSouthern DenmarkDK0334.68%North, South & West Europe
DenmarkZealandDK0242.73%North, South & West Europe
HungaryBudapestHU1117.66%East Europe
HungaryCentral TransdanubiaHU217.72%East Europe
HungaryNorthern Great PlainHU325.48%East Europe
HungaryNorthern HungaryHU316.26%East Europe
HungaryPestHU127.40%East Europe
HungarySouthern Great PlainHU337.07%East Europe
HungarySouthern TransdanubiaHU237.98%East Europe
HungaryWestern TransdanubiaHU226.11%East Europe
LithuaniaCentral and Western LithuaniaLT0217.98%East Europe
LithuaniaVilnius RegionLT0126.84%East Europe
NorwayAgder and Sør-ØstlandetNO0924.22%North, South & West Europe
NorwayInnlandetNO0223.64%North, South & West Europe
NorwayNorthern NorwayNO0720.38%North, South & West Europe
NorwayOslo and VikenNO0835.49%North, South & West Europe
NorwayTrøndelagNO0627.18%North, South & West Europe
NorwayWestern NorwayNO0A25.56%North, South & West Europe
PolandGreater PolandPL417.50%East Europe
PolandKuyavian-PomeraniaPL616.36%East Europe
PolandLesser PolandPL218.86%East Europe
PolandLodzkiePL717.57%East Europe
PolandLower SilesiaPL519.68%East Europe
PolandLublin ProvincePL815.86%East Europe
PolandLubuszPL435.58%East Europe
PolandMazowiecki regionPL922.94%East Europe
PolandOpole regionPL527.40%East Europe
PolandPodkarpaciaPL827.02%East Europe
PolandPodlaskiePL845.56%East Europe
PolandPomeraniaPL639.24%East Europe
PolandSilesiaPL228.05%East Europe
PolandSwietokrzyskiePL725.31%East Europe
PolandWarmian-MasuriaPL624.53%East Europe
PolandWarsawPL9115.99%East Europe
PolandWest PomeraniaPL427.92%East Europe
Slovak RepublicBratislava RegionSK0128.36%East Europe
Slovak RepublicCentral SlovakiaSK0314.00%East Europe
Slovak RepublicEast SlovakiaSK0413.36%East Europe
Slovak RepublicWest SlovakiaSK0214.00%East Europe
SpainAndalusiaES6115.83%North, South & West Europe
SpainAragonES2412.68%North, South & West Europe
SpainAsturiasES1215.08%North, South & West Europe
SpainBalearic IslandsES5311.87%North, South & West Europe
SpainBasque CountryES2122.42%North, South & West Europe
SpainCanary IslandsES7015.85%North, South & West Europe
SpainCantabriaES1310.99%North, South & West Europe
SpainCastile and LeónES4117.13%North, South & West Europe
SpainCastile-La ManchaES4218.93%North, South & West Europe
SpainCataloniaES5124.04%North, South & West Europe
SpainExtremaduraES4314.88%North, South & West Europe
SpainGaliciaES1117.06%North, South & West Europe
SpainLa RiojaES2314.61%North, South & West Europe
SpainMadridES3027.42%North, South & West Europe
SpainMurciaES6215.09%North, South & West Europe
SpainNavarraES2220.35%North, South & West Europe
SpainValenciaES5219.82%North, South & West Europe
SwedenCentral NorrlandSE3218.43%North, South & West Europe
SwedenEast Middle SwedenSE1229.23%North, South & West Europe
SwedenNorth Middle SwedenSE3122.09%North, South & West Europe
SwedenSmåland with IslandsSE2129.53%North, South & West Europe
SwedenSouth SwedenSE2231.32%North, South & West Europe
SwedenStockholmSE1144.77%North, South & West Europe
SwedenUpper NorrlandSE3331.51%North, South & West Europe
SwedenWest SwedenSE2337.08%North, South & West Europe

Regional values · CSV · Data and provenance · JSON · OECD workbook ↗ · Methods and coverage

OECD / Technology combinations

Using AI alongside cloud and analytics.

Around 12% is the unweighted mean across 28 covered OECD countries for enterprises with at least ten employees using all three technologies. It measures co-use; it does not establish successful workflow integration or returns.

Published mean and 28 country rates (displayed to four decimals)
OECD Figure 3.4 — published mean and 28 covered country rates
CountryReference periodAI + cloud + analytics (%)
OECD2025 or most recent (2021–2025)12.2162
Australia (2022)2021–20221.4000
Austria202511.1124
Belgium202522.7793
Canada (2023)20231.8000
Czech Republic202511.0392
Denmark202529.9112
Estonia202516.1476
Finland202525.4826
France20258.3080
Germany202512.9291
Greece20254.6703
Hungary20255.7149
Ireland202513.4635
Italy202510.1717
Japan (2023)2021–20232.9417
Korea (2024)202420.8704
Latvia20258.1669
Lithuania202514.6097
Luxembourg202514.2910
Netherlands202523.7163
Norway202516.1432
Poland20255.2906
Portugal20257.2088
Slovak Republic20258.6340
Slovenia202510.9384
Spain202510.6999
Sweden202520.8595
Türkiye20252.7536

Mostly 2025, with older observations and differing national definitions. Definitions, coverage and calculation.

OECD chapter / Figure 3.4 ↗ · Original workbook ↗ · Workbook snapshot · 29 observations · CSV · Data & provenance · JSON

United States / Census BTOS

Two business measures, separate questions.

US BTOS — 5–18 October 2026
MeasureShare of businessesStandard error (pp)
AI use in any business function24.2%0.61
Business reports employees using generative AI for work26.8%0.56

Both figures are percentages of covered nonfarm employer businesses, not employees. Reference period 5–18 October; collection 19 October–1 November; released 5 November 2026. The questions overlap. Wording changed to SI in this collection, so no trend is inferred or direct UK/EU ranking made. Sampling and comparability.

5 November release ↗ · Survey notes ↗ · Methodology ↗ · National workbook snapshot · Two observations · CSV · Data & provenance · JSON

Separate workflow evidence

Science · research sessions by task

Eight categories, two interfaces, 16 observations. The existing 335-observation ledger is unchanged.

Study / Figure 3 ↗ · Technical appendix ↗ · 16 workflow observations · CSV · Data & provenance · JSON

Chart and exact values → · Methods and limitations →

Australian work evidence

Recruitment and productivity remain separate measures.

ABS adds four published shares of businesses reporting perceived AI effects on vacancies. Their unexplained 99.2% total is retained and flagged. RBA adds approximate liaison context on productivity gains. Neither source is pooled with adoption rates.

All 335 core ledger observations and 16 separate science workflow observations remain unchanged.

ABS values and chart → · RBA context → · ABS · CSV · ABS provenance · JSON · RBA provenance · JSON · Australian sources and methods →

Original presentation

Five reports. Every page.

All 94 pages preserve maps, unlabeled series, appendices, methods and references. Physical PDF page numbers are used throughout.

Browse all reports and 94 page images directly

Global · H1 2025

Original PDF · Extracted text

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24

Global · H2 2025

Original PDF · Extracted text

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18

Global · Q1 2026

Original PDF · Extracted text

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21

Global · Q2 2026

Original PDF · Extracted text

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17

United States · Q1 2026

Original PDF · Extracted text

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14

Reusable evidence

Data, reports and source files.

Study / Figure 3 ↗ · Technical appendix ↗ · 16 workflow observations · CSV · Data & provenance · JSON

OECD chapter / Figure 3.4 ↗ · Original workbook ↗ · Workbook snapshot · 29 observations · CSV · Data & provenance · JSON

5 November release ↗ · Survey notes ↗ · Methodology ↗ · National workbook snapshot · Two observations · CSV · Data & provenance · JSON

Original source bytes are retained. JSON representations and text extracts are supplied separately; they do not replace the original reports.

Microsoft geographic data

Survey, sector and geographic evidence

  • Survey and sector observationsCSV
  • Evidence and provenanceJSON
  • EU country observationsCSV
  • Claude country observationsCSV
  • All geographic profilesJSON

Integrated evidence · HTML · All geographic profiles · HTML

Five original reports

Transcriptions and provenance

  • 65 records, 23 chart references and source notesJSON
  • 12 original-file sizes and SHA-256 fingerprintsJSON
  • Research overview and methodological linksMarkdown
  • Microsoft MIT licence noticeText
  • Original repository security guidanceMarkdown
Original source-file catalogue · 12 files

Microsoft research and original data: Microsoft AI Economy Institute. HTML views, calculated differences and visual presentation: in/out. Other contributing sources retain their own attribution and reuse terms.

Read the denominators

Methods, caveats and attribution.

Population use, firm adoption, employee experience and student use describe different things. Keep the source, question, population and observation window attached to every number.

Thirteen source groups: coverage, limits and reuse

Know what each number can carry.

Publication date is not observation date. Open the extracts for exact values, question bases, page references and source limitations.

Research coverage and observation windows
SourcePublishedObservedPopulation / coverageInterpretation limit
Microsoft AI Economy Institute2025–2026 report seriesH1 2025–Q2 2026; US Q1 2026147 economies; 50 states + DC, 35 metros, 3,143 countiesAdjusted population telemetry and modeled local estimates. Usage is not proficiency or productivity.
Anthropic26 June 2026Geography: May 2026; output analysis: 10 Apr–10 Jun 2026Claude chat and Cowork; country publication thresholds applyProvider-specific conversations; classifier-inferred purpose. The per-capita index is not a user percentage.
European Commission JRC / EurostatJRC: 14 October 20262021–2025; Eurostat snapshot updated 15 June 2026EU27; eligible enterprises with 10+ persons employedBroad AI, not only GenAI. Strategic-sector coverage varies; the country map uses the broader enterprise series.
McKinsey25 August 20264 May–8 June 20261,719 respondents across 97 countriesSelf-reports, GDP weighting, non-census sample. Reported financial effects are not causal estimates.
Gallup / JFF / Families & Workers Fund6 October 202626 January–24 March 202615,482 US employees; AI-effect questions: 7,845 AI usersEmployee assessments; frequency and benefit questions have different bases. Job quality association is not causation.
HEPI / Kortext12 March 2026December 2025 for the 2026 edition1,054 full-time UK undergraduatesWeighted self-report survey. Earlier waves differ; 2026 adds wellbeing uses. Tool provision is student-reported.
National Literacy Trust18 August 2026Early 2026; pupils and teachers January–March40,543 pupils aged 13–18; 2,567 teachers; 3,073 parentsSchool recruitment; pupil sample largely England. Questions use different routed bases; findings are not national prevalence estimates.
Dartmouth College30 October 2026 · working paper10–23 September 20261,003 US four-year college students; College Pulse weighted online panelSeparate, overlapping agreement measures, not complements or an adoption rate. Weighted online-panel self-reports; working paper, peer review not verified. Design effects were not verified. No causal learning outcomes or direct comparison with HEPI. Table 12, PDF p. 22.
UK Government · AI Adoption Survey29 October 20267 January–21 April 2026UK private-sector businesses with at least five employees in SIC A–N and R–S. 3,503 telephone interviews with technology-responsible respondents; Market Location commercial sampling frame; weighted by business size and sector; 5% response rate.Broad AI, including pilots, not only generative AI. Selected sectors and a five-employee threshold differ from Eurostat; no direct UK–EU ranking. Self-reports, not a census or causal productivity evidence. Low response creates nonresponse risk; weighting does not establish its absence. The report compares 32% with 16% in the prior wave. These are repeated cross-sections, not a panel; questionnaire and achieved samples changed. Tables 27 and 61.
Reuters Institute / Oxford29 October 20262025 and 2026 waves; 2026: 14 August–28 SeptemberSix-country weekly use; nine-country search exposure. YouGov online questionnaire; nonprobability samples with quotas and weighting for age, gender, region, education and political leaning. Combined-respondent estimates use YouGov weights, not an unweighted average of country percentages.Self-reported use and exposure, subject to recall and interpretation; online samples underrepresent people offline. No conventional sampling margin of error; not a worldwide estimate or directly comparable with Microsoft usage estimates. Different country sets and questions mean the weekly-use and search-exposure figures must not be subtracted as an exposure gap. Weekly-use comparison covers the same six countries, not a longitudinal panel. The prompted standalone-tool list changed from 13 tools in 2025 to 14 in 2026; this qualification accompanies the trend.

Comparisons are source-aware

No pooled adoption score is calculated. Percentages retain their population, question and time window; percentage-point changes use comparable published endpoints where available. Missing evidence is never treated as zero.

Reported effects are not causal effects

Speed, quality, profit and support figures are respondent assessments unless explicitly described otherwise. These studies cannot jointly establish that AI caused a country’s growth, changed learning outcomes or improved job quality.

Small multiples, quiet scales

Charts favour direct labels, shared scales within a comparison and restrained reference lines. Static SVG maps and charts remain in the initial HTML. Tables and downloads provide every plotted value without requiring interaction.

Geography has limits

World reach does not make every study global. US workforce and UK education evidence remain local lenses. Natural Earth borders are cartographic references; US state geometry derives from Census boundaries.

Attribution and reuse by source
  • Microsoft AI Economy Institute. Microsoft source collection uses MIT; attribution retained.
  • Anthropic. June dataset documentation specifies CC-BY without a version; attribution retained. Report linked at source.
  • European Commission JRC / Eurostat. JRC report CC BY 4.0; Eurostat source and reuse terms linked.
  • McKinsey. Facts transcribed and charts redrawn; original report remains with publisher.
  • Gallup / JFF / Families & Workers Fund. Facts transcribed and charts redrawn; original remains with publisher.
  • HEPI / Kortext. Report supplied for this analysis; publisher PDF and data remain at source.
  • National Literacy Trust. Facts transcribed and charts redrawn; original reports remain with publisher.

Natural Earth: public domain; US Atlas: Census-derived geometry, ISC package licence. No single licence is asserted over all contributing sources.

Excluded or uncertain comparisons

HEPI’s internally inconsistent encouragement percentage is omitted. National Literacy Trust frequency results with an unclear routed base are not presented as a share of all pupils; conflicting prior-year teacher-training figures are omitted. Eurostat time-series breaks and the expanded 2025 AI category remain in the downloadable data. No values are inferred from chart heights or unreported geographies.

Reuse documentation: Anthropic Economic Index data documentation · US Atlas original licence.

Microsoft methods, 17 transcription caveats and original-file fingerprints

The Microsoft collection: definitions and provenance.

What diffusion measures

The global reports define AI diffusion as the share of people aged 15–64 who used a generative AI product in the reported period. Aggregated Microsoft telemetry is adjusted for operating system and device share, internet penetration, and population. Usage does not measure capability, productivity, or economic impact.

How comparisons are made

Country changes subtract published rounded percentages. A rise from 10% to 12% is +2 percentage points. Periods are labeled categories; half-year and quarterly windows have different durations. No global or national user rate is calculated by averaging local percentages.

U.S. estimates have a distinct method

County figures are spatial small-area estimates, designed to address small samples, IP-geolocation noise, and infrastructure differences. State and metro aggregates use working-age population weights. The supplied CSVs contain no uncertainty intervals or population weights; none are fabricated here.

Design and completeness

Inspired by Edward Tufte’s principles: direct labels, common scales, quiet reference lines, small multiples, and data beside its explanation. All 3,376 CSV records and 3,817 rate observations are retained. Additional exact report data is transcribed; all 94 original pages cover the remaining graphical and narrative evidence.

Source inconsistencies and editorial decisions

The main country table follows AI_Diffusion_Q22026_Update.csv. The Q2 report appendix revises four Q1 2026 values upward by 0.1 percentage points: Algeria 13.2 → 13.3; Gambia 11.4 → 11.5; Ukraine 9.4 → 9.5; Cambodia 5.7 → 5.8. Q2 values agree. Each report snapshot is retained separately.

One economy name has a malformed byte in the CSV (T\x9frkiye). The main display uses “Turkey,” matching the PDF spelling. Original file bytes and source cell values remain intact in the original source downloads.

The H2 2025 report shows different UAE change labels: 4.6 pp in one chart and 4.50% in its appendix. Calculated changes in the main view always subtract the supplied rounded endpoints. Other printed changes may differ because of rounding.

The H1 2025 data-center capacity table totals 122.5 GW. Its U.S. and China entries total 85.6 GW (69.9%), while the accompanying prose states 86%. The original table and prose are both preserved; the capacity comparison uses the table values.

The U.S. PDF ranks 50 states, while the CSV contains 50 states plus the District of Columbia. The main table follows all 51 CSV entries. The 35 metros in the report have working-age populations above 1.5 million. Published top-county tables apply a minimum population of 10,000; the complete county CSV does not apply that display filter.

Report transcriptions preserve stated approximations, differing definitions, and source dates in their notes. Point-level values are not inferred from unlabeled curves or map colors. Full original pages provide coverage for those figures and all other narrative data.

Edition-specific transcription notes

The H1 PDF is image-only. Exact tables and key labels were visually read from rendered pages; OCR text is provided for navigation and may contain errors.

The 2025 H2 original appendix is an earlier vintage. Later CSV/report revisions should remain separate instead of being overwritten with these numbers.

H2 appendix changes are printed with % signs although differences are percentage points; several differ by 0.1 pp from subtraction of rounded endpoints.

H2 UAE change is 4.6 pp on page 4 but 4.50% on pages 6/14. Page 4 additionally differs from appendix for Switzerland (2.4 vs 2.5), Sweden (2.1 vs 2.2), Austria (2.3 vs 2.2), United States (2.0 vs 2.1).

H1 page 17 tabulated capacities sum to 122.5 GW; US+China total 85.6 GW = 69.9% of that sum. Pages 3/18 claim 86%. Original values preserved without resolving the inconsistency.

H2 page 7 user-base-growth chart says H1 to H2 2025 but prose says since October 2024. Treat 81.4% as printed with period ambiguous.

H2 page 8 benchmark chart carries an unrelated Global North/South chart title; English/Korean labels and adjacent text establish the intended measure.

Original PDFs mix population and working-age-population descriptions of diffusion. Retain the report-vintage methodology caveat; do not interpret user-share numbers as general-population counts.

Country PDFs use Turkey, while CSV apparently has corrupted encoding for Türkiye. Normalize entity identity without claiming the CSV spelling was valid.

Q2 report appendix revises prior Q1 values: Algeria 13.3 (previous 13.2); Gambia 11.5 (previous 11.4); Ukraine 9.5 (previous 9.4); Cambodia 5.8 (previous 5.7). Keep vintages visible.

Q1 Japan narrative claims3.4 pp increase is more than3 times global increase, but global table says1.5 pp. Preserve values, do not repeat claim.

Q1 agentic pull-request chart startsMay 2025; prose saysJune 2025.

Q2 Hugging Face figure dates first report Oct 2025 and endpoint Aug 2026; prose uses Nov 2025 and Sep 2026.

Q2 includes model/activity evidence through September 2026 although adoption snapshot is June 2026.

Global denominator explicitly ages 15–64. US report refers to working-age population; README describes distinct Bayesian spatial small-area estimates and working-age aggregation. The two methods should not be treated as identical.

US population and job-mix associations are not causal estimates.

Exact labeled numerical data were transcribed/extracted. Unlabeled graphics are retained through source-page images rather than estimated.

Reproducibility and original-file fingerprints

This page preserves the 2025–2026 source collection accompanying Microsoft’s AI Diffusion research. Filters and derived changes run locally in your browser. Source downloads reproduce the original bytes. CSV exports from filtered views are derived convenience files. SHA-256 fingerprints identify each original file.

Original fileSHA-256
LICENSE.md2dc1bdc83678f2c0fa39d1b8a6c4e143877cffe07789ce9135189b0a282e224f
README.md586827c455a8d5813c89ab8581c200d392b2b444c3f272dd86f31c270a3dc128
SECURITY.mdb02bf4e236e9335b27298a77688c751d18ca7e6052873d0d3d29eda37d0a72fa
data/AI_Diffusion_Q22026_Update.csv4a076d6aded6d8c9e60ddfff8a28476a72bb94e5293ae8feaa40c6bda06bd92a
data/US/County_AI_User_Share_2026Q1.csv5dc2355ef4c7b17af940710c959c6a78ff8f59ac55ff8541161c5ca53a5bf351
data/US/MSA_Ranking_2026Q1.csv4ea3d0afecbde178bfe5bce23995e8bc85a1479a0afc4c9e2fe9704b1e7c5d50
data/US/State_Rankings_2026Q1.csv6b79be96c85862e616bcce24b9262c8835046b7e1ef8e1986ddd3b26e2dd1ebe
reports/Microsoft-AI-Diffusion-Report-2025-H1.pdfa1220d061a69362b21f0767b7ef853594d64a6e7933fbd0a0ea2eee042246070
reports/Microsoft-AI-Diffusion-Report-2025-H2.pdf4f841408823012e8052143b83968b155d4f5acdb61017cf57412a7b13e50b267
reports/Microsoft-AI-Diffusion-Report-2026-Q1.pdf3bf9f87b633724128c8389dc0701c0f4e03c943730e4bec5b04878368b655782
reports/Microsoft-AI-Diffusion-Report-2026-Q2.pdfd9397e9ae9cced6d589ea9616ea34a7ad5d80744bd6d00f728914ca3577a9209
reports/Microsoft-US-AI-Diffusion-Report-2026-Q1.pdf084326ae828956d03ff0a5d3d7082582d1563bc38bb36d499eabb4e2d1b4fd89
Microsoft collection: attribution and MIT Licence

Research and source data © 2026 Microsoft, produced by the Microsoft AI Economy Institute (AIEI). Source: microsoft/ai-diffusion-report. Repository materials are provided under the MIT Licence. The original notice is reproduced below. This visual presentation was prepared for in/out.

MIT License

Copyright (c) [2026] [Microsoft]

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

Integrated analysis and visualisation by Alessandro L. Piana Bianco for in/out. Research remains attributed to its respective creators. No single licence is asserted over all contributing sources.

Menlo: spending intensity, separately measured

Two supplementary survey observations: average 4.7× and median 2.2× reported spending growth. Fieldwork 8–26 September; published 3 November 2026. Not pooled with adoption measures.

Primary source ↗ · Report PDF ↗ · CSV · Data and provenance · JSON · Methods · Chart and exact values

Reuters: acceptance of AI-produced news

Very or somewhat comfortable; four separate scenarios, not parts of a total. Nine-country weighted online sample, n=18,921; YouGov fieldwork 14 August–28 September 2026, published 29 October. This base differs from the six-country use trend; the figures are not a worldwide estimate. This extends the existing Reuters source, not an independent study.

Primary report · section 4 ↗ · Four observations · CSV · Data and provenance · JSON · Methods · Chart and exact values

Stanford AI100 · Infrastructure and agency

Published November 2026; expert synthesis of developments from 2021–2026. The infrastructure discussion (pp. 98–105) covers interdependent models, data, networks, chips and data centres, and the consequences of concentrated capacity for sovereignty and agency.

Contextual synthesis, not a population survey, adoption rate or causal estimate. Access to an AI model does not establish the capacity to operate, adapt or govern it. No new quantitative ledger observations are drawn from this source.

Primary source ↗ · Methods and qualifications · Context and provenance · JSON

McKinsey Global Institute · AI fluency in job requirements

Agents, robots, and us: Skill partnerships in the age of AI, November 2025. Exhibit 7 combines Lightcast job-posting analysis and US Bureau of Labor Statistics employment: occupations where AI fluency appeared in at least 5% of postings employed about 1.0 million workers in 2023 and 7.0 million in 2025. The report identifies May 2025 postings in its methods.

Employees in occupations crossing a posting threshold, not AI users, people with AI skills, vacancies or new jobs. The rounded values are retained; no causal productivity or employment outcome is inferred. This is older supporting evidence, not a November 2026 release. Automation-potential and economic-value scenarios are not measured outcomes.

Primary source ↗ · Methods and qualifications · Context and provenance · JSON

Dubai Future Foundation · AI in creative industries

Guidelines for the development, regulation, and use of artificial intelligence in creative industries (2024). Research ran from October 2023 to September 2024: two virtual and four in-person participatory-design workshops with 67 participants, expert interviews, and small-scale surveys and opinion polls. The report addresses authorship, compensation and cultural representation.

Qualitative expert consultation, explicitly not for statistical analysis. Senior experts were selected and participation expanded through snowball sampling; this is not a representative survey of creative workers, industries or Dubai. Used for governance context, with no quantitative adoption observations.

Primary source ↗ · Methods and qualifications · Context and provenance · JSON

Kyndryl · deployment and organisational redesign

Three separate leadership-survey observations, fielded 10 July–2 September 2026. Agreement thresholds differ; not a national adoption estimate or an integration-gap calculation.

Data
Kyndryl 2026–27 Readiness Report. Senior-leader responses, n=3,700; fieldwork 10 July–2 September 2026. Different response thresholds; not a subtraction-based integration gap.
StatementShareResponse thresholdSource
AI deployed beyond pilot projects84%AgreeQ8; p.9
Roles formally redesigned or reorganised around AI27%Completely agreeQ6; p.25
Formal AI change-management programmes fully implemented29%Completely agreeQ6; p.25

Methods and qualifications · CSV · JSON