AI Adoption by Country: Overview and Rankings

Facts as of 12 October 2026; the per-capita, investment, compute and rule items below were re-checked against primary sources that day, and items not re-read keep their 11 October reading. This page compares countries and links to four regional documents: Americas and Europe, Asia-Pacific, India and Middle East and Africa. Labels: "secondary" means trade-press, law-firm or aggregator coverage rather than the original publisher. "Single source" means only one source was found. "n/a" means no value was read in the research for this page; it does not mean zero.

Key takeaways

• Personal AI use is highest in small, rich, well-connected economies. In June 2026, 73.3% of working-age people in the UAE used generative AI, ahead of Singapore (64.3%), Ireland (49.9%), France (49.6%) and Norway (49.4%). The world average was 18.8%: 28.8% in the Global North and 16.2% in the Global South. The US (33.0%) ranked 21st of the 147 economies Microsoft measures.
• Firms lag people almost everywhere both are measured. About one firm in five uses AI in the EU (19.95% of firms with 10+ staff), the US (17-20%) and Canada (19.2%), while about half of US adults use chatbots. EU firm use ranges from 42.03% in Denmark to 5.21% in Romania. The exceptions are Denmark and Finland, where more firms use AI than individuals (42.03% versus 32.4%, and 37.82% versus 30.6%), and Japan, where a private survey finds 34.5% of firms using generative AI against 24.7% of people. Large firms adopt at two to three times the rate of small ones (EU 2025: 55.03% of firms with 250+ staff against 18.90% of SMEs with 10-249).
• Money and compute concentrate in the US, but absolute counts understate China. US private AI investment was 23 times China's, and in generative AI it exceeded China's and Europe's combined "by a wide margin". US cloud giants guided about $700-725 billion of 2026 capex (secondary), and the US holds 25 of the top 100 supercomputers. China, though, has the fastest listed supercomputer and about $184 billion of state guidance funding (2000-2023). Its model makers drew at least 37% of OpenRouter requests.
• Singapore is the most consistently prepared country. It tops the IMF index (0.801, 2023 data), is second on usage and on Anthropic's September 2025 per-capita index, and has among the highest values in Anthropic's June 2026 release (6.44 in April, 5.81 in May 2026). Scale-weighted rankings put the US first (Oxford Insights, Tortoise; secondary).
• Rules range from binding statutes to voluntary codes. The EU AI Act applies from 2 August 2026, with high-risk duties from December 2027. South Korea (22 January 2026) and Vietnam (1 March 2026) have binding AI laws that reach foreign firms. China regulates through filing and labelling rules. The US has binding state laws but no federal statute. Japan and Taiwan have framework laws without penalties. India, Singapore, Australia and Canada rely on guidelines plus existing law.
• India is large in total and small per person. In Anthropic's India brief (November 2025 data) it was second only to the US in share of Claude.ai use (5.8%) but 101st of 116 per capita; the June 2026 release puts it at 6.92% (April) and 7.12% (May 2026) of use, and 0.31 and 0.30 per capita. It has about 100 million weekly ChatGPT users. Microsoft puts it at 18.5% (64th of 147), just under the world average and above China. It scores 0.493 on the IMF index (28th of the 32 countries here with a score) and has one top-100 supercomputer.
• For a business adopting AI, the gap between personal and company use is the place to start, and compliance documentation now has fixed deadlines. EU transparency rules apply now. Colorado's automated-decision law starts on 1 January 2027, India's core data-protection duties on 13 May 2027, and EU high-risk duties in December 2027.

Thirteen yardsticks answer thirteen different questions

No single number captures "AI adoption". The indices quoted in news coverage count different things (people, firms, governments, money or machines), some per person and some in total, with data dates spanning three years. The table below sets out what each measure counts; read every later ranking against it. The glossary defines the terms, and the landscape overview places them in the wider market.

Each measure is cited where it is first used below. One figure needs a warning. Stanford's 2026 economy chapter gives country generative-AI adoption of 64% for the UAE, 61% for Singapore and 28.3% for the US (24th). They appear under a takeaway on how fast the population adopted generative AI ("Generative AI reached 53% adoption in three years"), and the page does not name the underlying data (Stanford Economy chapter, re-read 11 October 2026). They are sometimes quoted as organisational adoption, but the page does not describe them that way, and all three match the H2 2025 column of Microsoft's Q1 2026 report (UAE 64.0%, Singapore 60.9%, US 28.3%; the US then moved "from 24th to 21st") (Microsoft Q1 2026; Microsoft blog). They are therefore population figures, almost certainly Microsoft's (an inference from the exact match), and this page keeps them out of the enterprise ranking.

Master table: 33 countries on seven measures

The table puts the latest value of each measure side by side without combining them. Sources by column: Microsoft Q2 2026 with Q1 in brackets (Microsoft Q2 2026). Enterprise values are Eurostat 2025, firms with 10+ staff, unless another source is named (Eurostat API). The named sources are the US Census Bureau, Statistics Canada, the UK ONS and Teikoku Databank via Asia News Network (secondary). IMF values use 2023 data (IMF). AUI values are from the September 2025 report (Anthropic), with the June 2026 release's April / May 2026 values in brackets where they were read (Anthropic CSV; dataset); the June file was not extracted for the UK, Canada, Germany, France or most other European and American countries. The two releases measure differently (Israel 7.0 in 2025, about 3.0 in 2026), so do not rank across them. TOP500 counts systems in the top 100 of the June 2026 list (TOP500). The 27 countries listed there hold all 100 systems, so "0" means none. Readiness ranks are secondary: Oxford Insights and Tortoise via Comland. Regulation sources are in the regulation section.

The table supports three readings. Usage and readiness diverge: the UAE is first on usage but 18th of the 32 countries here with an IMF score, while Germany ranks 5th on the IMF index but only 23rd of 147 on usage. The largest economies dominate the absolute measures, such as supercomputers, investment and total users, but not the per-person ones. In most countries with an enterprise figure, it sits below the Microsoft figure for individuals. Japan is the exception in this table, and Sweden's two figures are close (35.04% of firms, 37.3% of people).

Usage rankings: the Gulf and small open economies lead

The top 15 by share of people using AI

Microsoft's June 2026 table places the UAE (73.3%) and Singapore (64.3%) far ahead of everyone else. Four European economies follow at 45-50%: Ireland, France, Norway and Spain. The world average was 18.8%, up from 17.8% in Q1, and the gap between the Global North (28.8%) and South (16.2%) reached 12.6 percentage points (Microsoft Q2 2026). Microsoft called that gap "continued widening" in May, and it widened again: it was 12.1 points in Q1 2026 (27.5 versus 15.4) (Microsoft blog). Among the 33 economies in the master table, Nigeria (10.4%) and Kenya (8.9%) are lowest, an eightfold gap from the UAE. Several economies outside the table sit lower still: Bangladesh 8.0% and Sri Lanka 7.6% in South Asia (Pakistan is at 11.9%), and Ethiopia 8.0%, Rwanda 7.6% and Syria 7.4% in Africa and the Middle East. Microsoft's full table of 147 economies bottoms out at 6.1%, in Cambodia (same source).

The ranks are Microsoft's own positions among the 147 economies in the Q2 2026 data table (PDF re-read 11 October 2026; Microsoft Q2 2026; method in arXiv:2511.02781). Qatar ties the Netherlands at 43.5% and is placed 10th, as in Q1 2026, so one outlet's "for the first time" line is wrong (Microsoft Q1 2026; Middle East AI News, 25 Sep 2026). South Korea climbed from 16th to 12th. The US was 21st in both Q1 and Q2 2026, with Sweden, Austria, Taiwan and Hungary also above it, after rising from 24th in the second half of 2025 (Microsoft Q1 2026). By this page's count of Microsoft's table, 31 economies were above 30% in June, against 26 in Q1.

The movers explain more than the levels do. South Korea had the largest absolute gain in Q2 (+3.5 points). Japan had the largest relative gain (+2.2 points, about 10%), moving to 45th after climbing from 56th to 48th in the previous edition. Saudi Arabia rose from 30th to 25th (Microsoft Q2 2026; Microsoft Q1 2026). Microsoft links the Asian gains partly to better AI capability in Asian languages, a sign that language support, not only income, gates usage. Over a longer span, the UAE rose from 59.4% in the first half of 2025 to 73.3%, the US from 26.3% to 33.0%, Japan from 16.7% to 24.7%, India from 14.2% to 18.5% and China from 15.4% to 17.5% (arXiv:2511.02781; Microsoft Q2 2026). Income still matters. In Anthropic's data, a 1% rise in GDP per capita goes with a 0.7% rise in Claude use per capita across countries (Anthropic, Sept 2025).

Anthropic's per-capita index has a different order at the top. In August 2025 the top five were Israel (7.0), Singapore (4.57), Australia (4.10), New Zealand (4.05) and South Korea (3.73), with the US (3.62) next among the countries the report names, while India (0.27) and Nigeria (0.2) were near the bottom (Anthropic, Sept 2025). Per-capita Claude use concentrates in high-income, English-speaking tech hubs. That lifts Israel (15th on Microsoft's measure) and Australia (11th) to the top three. Anthropic's June 2026 release (April / May 2026; Claude.ai chat and Cowork only) gives Singapore 6.44 / 5.81, Australia 5.83 / 6.40, the US 4.25 / 3.87, Israel 3.03 / 3.05, India 0.31 / 0.30 and Nigeria 0.21 / 0.24 (chart below; Anthropic Economic Index data, primary). Israel's 7.0 (September 2025) and the "4.9x" in March 2026 news come from different releases or series, so do not compare them with about 3.0; no ranks were computed from the June file. More of the comparison is in the Claude document.

Who uses the most in absolute terms

Absolute totals favour the largest countries. The US accounted for 21.6% of Claude.ai use and India for 7.2% in August 2025 (Anthropic, Sept 2025). Anthropic's January 2026 report names the US, India, Japan, the UK and South Korea as the leaders in overall use (Anthropic, Jan 2026). China reports the largest national user count found for this page. CNNIC counted more than 700 million generative-AI users by mid-2026, a penetration rate above 50%, up from 515 million in June 2025 (ECNS, 29 Sep 2026; CRN Asia). India has about 100 million weekly ChatGPT users, OpenAI's second-largest consumer market (TechCrunch, 15 Feb 2026; Business Today, 19 Jan 2026).

Worker surveys show higher numbers than population telemetry in emerging markets. Stanford reports that more than 80% of employees in India, China, Nigeria, the UAE, Egypt and Saudi Arabia used AI at work semi-regularly or regularly in 2025, against 58% globally (Stanford Public Opinion chapter). KPMG's 47-country study finds that 66% of people use AI regularly (KPMG). In the US, Pew found that 49% of adults use chatbots and 24% use them daily. Part of the rise from 33% in 2024 reflects a change in the sample (Pew; Pew, daily use).

Enterprise adoption: Nordic firms lead where measurement is cleanest

Europe has the only official, comparable cross-country measure of firm adoption. Eurostat's 2025 survey finds 19.95% of EU enterprises with 10+ staff using at least one AI technology, up from 13.48% in 2024 and 7.65% in 2021. The spread runs from Denmark (42.03%) to Romania (5.21%), about eightfold (Eurostat API; WAM citing Eurostat). The fastest gains were in Denmark (+14.5 points), Finland (+13.5) and Lithuania (+12.5), as the 2024 values confirm (Eurostat API, 2024; Implicator.ai). Denmark and Finland are the two countries on this page where more firms than people use AI: Microsoft puts individual use at 32.4% and 30.6% (Microsoft Q2 2026). The most common use was analysing written language, at 11.8% of enterprises (WAM).

Firm size matters as much as country. In the US, 37% of firms with 250+ employees use AI, against under 20% of firms with four or fewer, and 20-23% of firms expect to adopt within six months (US Census). In Japan, 63.6% of firms with more than 1,000 staff use generative AI, against 29.6% of firms with five or fewer (Asia News Network, secondary). Definitions also matter. A survey reported by Reuters found that only 16% of Japanese companies had integrated AI across the whole company, a much narrower test than "uses generative AI" (Mezha, secondary). No economy-wide enterprise survey was found for China, India, South Korea or the Gulf states. In the UK, the ONS business survey (BICS; a different survey from Eurostat's, so not like-for-like) found 25% of businesses using some AI technology in late December 2025 (44% of those with 250+ employees; 15% planning to adopt within three months) (ONS, 8 Jan 2026). In the EU, 55.03% of firms with 250+ staff used AI in 2025, against 18.90% of SMEs (10-249) (Eurostat API). For China, CNNIC reports only that more than 30% of manufacturing enterprises above designated size had adopted AI by June 2026 (ECNS, 29 Sep 2026). Stanford reports that China and Europe posted the largest year-on-year increases in organisational adoption, without country values (Stanford Economy chapter). The use-cases document covers which functions firms automate first.

Preparedness: Singapore first on data, the US first on scale

Preparedness indices measure capacity to absorb AI, not uptake, so they correlate only loosely with usage. On the IMF index, which covers 178 economies, Singapore (0.801), the US (0.771) and the Netherlands (0.766) lead the 33 countries in the master table. India (0.493) is 28th of the 32 of them with a value, between South Africa (0.497) and Vietnam (0.482) (IMF AI_PI, re-read 11 October 2026). Globally, Denmark (0.779) is second, behind Singapore and ahead of the US (same source). Secondary coverage of the 2024 release put South Sudan (0.11) and Afghanistan (0.13) at the bottom (Khan Global Studies, secondary). Neither country appears in the API data, where the lowest score is the Central African Republic's 0.184. The IMF API returns 2023 data only, so these scores predate the generative-AI boom they are used to explain.

Sources: IMF; Oxford Insights; Egypt Today; CADE project summary; Let's Data Science; Al Riyadh Daily; Comland.

The government-focused indices reward scale and state capacity. Oxford Insights' 2025 index of 195 governments puts the US 1st, China 8th (up from 23rd) and Canada 12th, according to a summary of its page. Egypt is 51st with a score of 57.5, first in Africa, and Saudi Arabia is first in the Middle East and North Africa and 7th globally on the governance pillar (Oxford Insights; Egypt Today; Al Riyadh Daily). Tortoise's 2025 index, which weights implementation, innovation and investment, ranks the US 1st, China 2nd, Singapore 3rd, the UK 4th and South Korea 5th (Comland, secondary). India is absent from every readiness rank recovered for this page except the IMF score.

Three mismatches show what the indices miss. The UAE leads usage but is 18th of 32 on the IMF index. Germany is 5th on preparedness but 23rd on usage (32.3%) and only modestly above the EU on enterprise use (25.97%). Japan scores 0.733 but had only 24.7% usage. Microsoft attributes Japan's recent climb to better Japanese-language model performance, which suggests that language support, not readiness, was the constraint (Microsoft Q1 2026). Singapore is the one country near the top of nearly every list: usage, per-capita Claude use, preparedness, and 81% public trust in its government to regulate AI, against a 54% global average and 31% in the US (Stanford Public Opinion chapter).

Investment and compute: the US builds, China is undercounted

Investment and compute are absolute measures, and the US dominates them. Stanford's 2026 AI Index states that the US committed 23 times more private AI investment than China, that in generative AI US investment "exceeded the combined total of China and Europe by a wide margin", and that private investment grew 127.5% (Stanford Economy chapter, re-read 11 October 2026). The report PDF gives the dollar values for 2025: $285.9 billion for the US and $12.4 billion for China (23.1x), $5.9 billion for the UK (US 48.5x), $20.92 billion for Europe and $344.7 billion globally (+127.5%). In generative AI, US private investment was $163.6 billion against $1.48 billion for China and $3.21 billion for Europe. The US had 1,953 newly funded AI companies (UK 172, China 161) (Quid data; funding rounds, not spending) (Stanford AI Index 2026 PDF, pp. 181-184). Stanford itself warns that private tallies understate China. Government guidance funds put an estimated $184 billion into Chinese AI firms between 2000 and 2023 (Stanford Economy chapter).

Capital spending tells the same story at a larger scale. Q1 2026 guidance from Amazon, Microsoft, Alphabet and Meta implies about $700-725 billion of 2026 capital expenditure, up from about $410 billion in 2025 (AI Weekly; Yahoo Finance, both secondary). OpenAI said in late April 2026 that it had signed contracts for 10 GW of compute. That is contracted capacity, not operating capacity (Bloomberg Law). The hardware and costs documents cover what this buildout means for chip and token prices.

The TOP500 list shows who builds the largest public supercomputers. The US has 25 of the top 100 systems in June 2026 and Japan 14, then South Korea 7, Germany 6, Italy and the UK 5 each, and Saudi Arabia and France 4 each. The European countries on the list hold 32 systems between them (a sum of the published counts), more than the US. The No. 1 system is Chinese: LineShine-LingKun at the National Supercomputing Centre in Shenzhen, which displaced the US El Capitan system that led in November 2025 (TOP500 June 2026; TOP500 Nov 2025). Yet China has only 2 systems in the top 100, which more likely reflects non-submission than a lack of capacity (an inference). TOP500 is not a measure of AI training capacity. CNNIC puts China's intelligent computing capacity at 2,185 exaflops in June 2026, up 177% in a year. No numeric precision is stated, so it cannot be compared with Western capacity numbers (ECNS on the CNNIC report, 29 Sep 2026). Even Epoch AI's dedicated GPU-cluster dataset covers only an estimated 10-20% of global GPU cluster performance (Epoch AI).

Outside the US, most compute figures are pledges or targets, not operating capacity. The table separates them.

Model building gives a different picture from hardware, and here China is close to the US. In the week beginning 28 September 2026, Chinese model makers took at least 37.0% of text requests on OpenRouter and US makers 42.2%. In the weekly token ranking read on 11 October, six of the ten models with the most tokens were Chinese (OpenRouter). These request and token rankings cover only traffic routed through OpenRouter, which says they measure neither users nor spend. Its volume leans toward low-priced "Flash" models, so these shares reflect price-sensitive developer traffic. The DeepSeek and other models documents cover the makers.

Why the indices disagree: four effects and four worked examples

Four effects explain nearly every contradiction between the rankings: telemetry versus surveys, per-capita versus absolute counts, consumer versus enterprise use, and edition dates. Each has a worked example.

Telemetry versus surveys: China at 17.5% and at 700 million

Microsoft's telemetry puts 17.5% of China's working-age population using generative AI in June 2026, below India and Turkey (Microsoft Q2 2026). CNNIC's national survey counts more than 700 million users and penetration above 50% at the same point in 2026 (ECNS). Stanford's survey finds more than 80% of Chinese employees using AI at work (Stanford Public Opinion chapter). All three figures can be correct. Microsoft extrapolates from its own operating system and software, adjusting for device share, and says "no single metric is perfect" (Microsoft blog). Microsoft also says its current measure misses use in China: broader tool coverage would raise measured use almost everywhere, "with the largest effect in China, where several widely used domestic tools are currently not included". It plans to widen coverage in its next report (Microsoft Q2 2026). Nigeria shows the same pattern: 10.4% on Microsoft's measure, while more than 80% of employees use AI at work. Workers are a subset of the population, so the two numbers are not directly comparable. Survey figures and telemetry should never share a ranking column.

Per-capita versus absolute: India is second and 101st at once

Anthropic's India brief gives the cleanest example. India accounts for 5.8% of global Claude.ai consumer use, second only to the US, yet ranks 101st of 116 countries per capita (Anthropic India brief). A country of India's size can be a top market for every AI vendor while sitting near the bottom of every per-person table. The same logic lifts Israel (AUI 7.0) and Singapore (4.57) above much larger economies in the September 2025 data (in the June 2026 release Israel is about 3.0 and Singapore 5.8-6.4). Composition differs too. Coding was over half of Claude use in India in August 2025, against about a third globally. The November 2025 brief, on a different basis, puts software-related tasks at 45.2%. Anthropic also finds that low-AUI countries lean toward automation and high-AUI countries toward augmentation, after controlling for task mix (Anthropic, Sept 2025). A low per-capita rank therefore does not mean shallow use. It means use is concentrated in a professional minority.

Consumer versus enterprise: the US is first, sixth, 21st and 24th

The US shows how far one country can move between indices. It is first on Oxford Insights, Tortoise, TOP500 top-100 system count, private investment and share of Claude use (21.6%). It is sixth among the countries Anthropic's September 2025 report names on its per-capita index (3.62). It is 21st of 147 on Microsoft's population measure (33.0%). Stanford's figure, which matches Microsoft's data for the second half of 2025, puts it 24th, at 28.3%. Firm-level adoption is lower still: 17-20% of US firms in the Census survey, against 49% of adults who use chatbots (US Census; Pew). The Census series has its own break: in November 2025 the question widened from AI use in producing goods or services to use in any business function (Federal Reserve FEDS Note). The UAE is the mirror case. It is first on usage at 73.3%, but its Claude per-capita value in the June 2026 data (2.73 in April and 2.84 in May) is modest. It is also 18th of the 32 countries here on the IMF's preparedness measure (Anthropic CSV; IMF).

Edition dates: a 2023 index next to June 2026 telemetry

The measures on this page span three years of data. IMF scores use 2023 data, Anthropic's September 2025 table one week of August 2025, Eurostat 2025, and Microsoft and TOP500 June 2026. Anthropic's newer release (26 June 2026) covers April and May 2026, and its country file has now been read for the countries in this page's tables (Anthropic data documentation; Anthropic Economic Index data). Edition drift also creates false conflicts: Israel is 7.0 in the September 2025 report, "4.9x" in a March 2026 news piece and 3.03 / 3.05 in the June 2026 release. These are different releases or series and should not be compared; this page dates each value (NRIGlobe; Anthropic, Sept 2025). The practical rule is to quote any country figure together with its measure and its data date.

Rules: binding statutes, targeted rules and voluntary codes

Regulatory strictness falls into six groups, plus a set of countries whose rules were not researched. Where a law exists, its dates matter more than its label. The governance document covers obligations in detail, and the timeline has the sequence of events.

Sources (US and EU items not listed here are cited in the text below): European Commission, AI Act; Gazzetta Ufficiale, Law 132/2025; Cooley on Korea; The Daily Star on Korean fines; IAPP on Korea's representative thresholds and grace period (opinion); Baker McKenzie on Vietnam; Digital Policy Alert on CAC filings; Digital Policy Alert filing tracker (secondary); Loeb & Loeb on labelling; Caixin; Clifford Chance on SB 53; FPF on Japan; Baker McKenzie on Taiwan; Indian Infrastructure on India's guidelines; CMS on Singapore; MinterEllison on Australia; LEGISinfo, Bill C-27; ISED, AI for All; Câmara dos Deputados, PL 2338/2023; Ecosistema Startup on Mexico; OECD.AI on Nigeria; Digital Policy Alert on Nigeria; The Standard on Kenya; OECD.AI on Egypt; gov.uk, DSIT; GOV.UK, Ministerial appointments July 2026; legislation.gov.uk, DUAA s.80; Focus Malaysia; Digital Policy Alert on Vietnam's first decree (secondary); Adnkronos on Japan's first Basic Plan (secondary); CourtListener, X.AI LLC v. Weiser; industry.gov.au, AI Safety Institute; HapaKenya on Kenya's AI Bill 2026 (secondary); OECD.AI national pages for Indonesia, Thailand, Israel, Türkiye and Egypt (OECD.AI listing, secondary); Tech For Good Institute on Indonesia (secondary); Daily Maverick on South Africa's withdrawal (secondary).

The EU delay changed the timing of its rules, not their scope. The AI Omnibus, in force since 27 July 2026, moved high-risk duties by about 16 months. It left transparency and general-purpose-model enforcement on schedule. It added two prohibitions, covering systems built to generate realistic intimate imagery or child sexual abuse material, from 2 December 2026. And it gave the AI Office power to impose periodic penalty payments of up to 5% of average daily turnover (NatLawReview). The AI-literacy duty in Article 4 survives, with no mandated level and no certificate required. National authorities have enforced it since August 2026 (European Commission AI literacy Q&A).

The US is moving in two directions at once. The December 2025 executive order directed an AI Litigation Task Force to challenge state laws (Federal Register). States kept legislating anyway. Colorado repealed its 2024 act and replaced it with a disclosure-based law on automated decisions. From 1 January 2027 it requires developers to give deployers technical documentation, and deployers to give point-of-interaction notice, a plain-language explanation within 30 days of an adverse outcome and human review on request; both must keep records for three years (Colorado General Assembly). Comments on the state attorney general's draft rules are open until 26 October 2026 (Colorado AG). A federal court order in X.AI LLC v. Weiser (27 Apr 2026) bars the AG from beginning enforcement of SB 24-205 or its replacement until 14 days after a preliminary-injunction ruling; whether that reaches SB26-189 is arguable (CourtListener). Texas's TRAIGA (effective 1 January 2026 per earlier sources; the date is not in the enrolled text) offers a safe harbour for substantial compliance with the NIST AI Risk Management Framework. Its enrolled text allows civil penalties of $80,000-200,000 for uncurable violations and up to $40,000 a day for continuing ones, after a notice-and-cure process (Texas HB 149, enrolled; Nelson Mullins). New York City has required independent bias audits of automated hiring tools since July 2023 (NYC DCWP). Washington also regulates through model access. A June 2026 US export-control directive (attributed to the Commerce Department in press reports; Al Jazeera, 1 Jul 2026) suspended foreign-national access to two Anthropic frontier models for 19 days (Anthropic, 12 Jun 2026).

Asian regulators act quickly against specific products even where no AI law exists. Indonesia, Malaysia and the Philippines blocked the Grok chatbot in January 2026 and lifted the blocks within weeks (Wikipedia, 2026 in AI, unverified lead). Public appetite for rules is broad. KPMG finds that 70% of people believe AI regulation is needed and only 46% are willing to trust AI systems (KPMG). Pew's 2026 survey (42,151 adults, 8 Feb-13 May 2026; 37 countries and territories listed) found more people expect AI to bring fewer jobs than more in 34 of 37; "more concerned than excited" rose most in Sweden (+9 points), Poland (+8), the Netherlands (+7) and Hungary (+6) (Pew, 17 Sep 2026).

Regional pictures in brief

Americas and Europe: high usage, slow firms, rules on two tracks

Western Europe has the highest population usage outside the Gulf and Singapore. Ireland (49.9%), France (49.6%), Norway (49.4%), Spain (45.1%), the UK (43.7%) and the Netherlands (43.5%) all exceed the US (33.0%) and Canada (38.5%) (Microsoft Q2 2026). Firms trail people almost everywhere: EU enterprise use is 19.95% (33-42% in Denmark, Finland, Sweden and the Benelux, 5-16% in Italy, Portugal, Greece, Poland, Bulgaria and Romania), US 17-20% and Canada 19.2%, with Denmark and Finland the exceptions. Latin America trails on usage: Colombia 25.9%, Chile 24.0%, Argentina 23.4%, Mexico 21.4% and Brazil 20.2%. Brazil's AI bill sits in a Chamber special committee "awaiting the rapporteur's opinion" (Chamber record, primary); MobileTime reported that the rapporteur expected a vote only after the October 2026 elections, a date the record does not set (MobileTime, 24 Aug 2026, not re-read on 12 Oct 2026). Chile's bill passed the lower house in October 2025 (IAPP). Colombia adopted a national AI policy, CONPES 4144, in February 2025 (DNP). The US dominates investment and compute, while the EU funds gigafactories and France and the UK depend largely on private pledges. Full detail is in Americas and Europe.

Asia-Pacific: Singapore leads, China is hard to measure, laws reach foreign firms

Singapore (64.3%) leads the region, followed by New Zealand (44.1%), Australia (40.7%), South Korea (40.6%), Taiwan (33.8%), Vietnam (27.9%), Japan (24.7%), Malaysia (22.6%) and the Philippines (20.7%). China (17.5%), Indonesia (14.8%) and Thailand (13.4%) trail on Microsoft's measure, and South Asia is lowest: Pakistan 11.9%, Bangladesh 8.0% and Sri Lanka 7.6% (Microsoft Q2 2026). China's survey count of more than 700 million users tells the opposite story. Its State Council "AI+" plan targets penetration of intelligent terminals and agents above 70% by 2027 and 90% by 2030 (gov.cn). The region holds all three regulatory models: binding laws in Korea and Vietnam, framework laws in Japan and Taiwan, and voluntary codes in Singapore and Australia. It also holds much of the world's compute supply chain: Japan's 14 top-100 supercomputers, Korea's 260,000-GPU Nvidia deal, TSMC's foundry share, Johor's data centres and China's reported 2,185 exaflops of intelligent computing (ECNS). Pakistan states a goal of $1 billion in AI investment by 2030, from a single government source (MoITT). See Asia-Pacific.

India: a top-two market by volume with a per-person gap

India is covered in its own section below and in the India document. In brief, it combines very large absolute use with low per-person use, a principles-based AI governance model with hard data-protection dates, and a fast-growing but still small compute base.

Middle East and Africa: the world's top user and some of its lowest

The UAE leads the world on usage (73.3%), and Qatar (43.5%) held 10th place, as it did in Q1 2026. Israel (38.7%) is third in the Middle East, and Saudi Arabia (31.4%) climbed five places (Middle East AI News; MediaPost). The Gulf builds through sovereign vehicles tied to US chips: Stargate UAE, Saudi Arabia's HUMAIN (with its Arabic-first ALLaM model) and Qatar's Qai, which will commercialise existing models rather than build its own (Wikipedia, Humain; Qatar Tribune, single outlet). Security is a risk too: in April 2026 Iran's Revolutionary Guard threatened the planned Abu Dhabi data centre (The Verge). Africa has the lowest usage read: South Africa 23.9%, Egypt 15.5%, Morocco 12.0%, Nigeria and Ghana 10.4%, Kenya 8.9%, Ethiopia 8.0% and Rwanda 7.6% (Microsoft Q2 2026). Its build-out rests on national strategies, a few compute projects and local-language models. Cassava's Nvidia-based "AI factories" are one example, and Vodafone Business and Cassava announced "Egypt's first sovereign AI factory" (Engineering News headline dated 25 Sep 2026; article paywalled, not read). Policy is thin: South Africa withdrew its draft national AI policy in April 2026, and Kenya's AI Bill 2026 is still a bill with an unclear status (see the regulation table). Nigeria's N-ATLaS (Yoruba, Hausa, Igbo and Nigerian English) is a Llama-3 8B fine-tune by Awarri with NITDA whose licence caps free use at 1,000 users and requires a separate commercial licence (Technology Times; Hugging Face, N-ATLaS model card). See Middle East and Africa.

India in every index: second in total, 101st per person

India's profile is the per-capita-versus-absolute problem in its sharpest form. It is a top-two market for both OpenAI and Anthropic. Anthropic calls India its second-largest market and opened a Bengaluru office in February 2026 (Fortune; Anthropic, Bengaluru office). Yet the typical Indian uses AI far less than the typical Singaporean or Emirati. Use is concentrated by region and by job. Maharashtra, Tamil Nadu, Karnataka and Delhi produce more than half of India's Claude use, and 51.3% of that use is work-related, against 46% globally. Indian users finish in 14.8 minutes tasks that would take 3.8 hours without AI, a 15x speedup against 12x globally (Anthropic India brief). On Microsoft's data India sits just below the world average, and it overtook China between the first half of 2025 (14.2% versus 15.4%) and June 2026 (18.5% versus 17.5%) (arXiv:2511.02781; Microsoft Q2 2026).

India's preparedness and infrastructure lag its usage. Its IMF score of 0.493 is below Brazil's and South Africa's. Data-centre capacity is about 1.6 GW and expected to reach 6 GW by 2029, needing about $110 billion of investment (IANS, 2 Sep 2026, reporting a JLL report; the JLL release was not read: secondary) (IANS citing JLL). Other 2030 forecasts range from about 3 GW (Avendus) to about 8 GW (Jefferies), as reported in the press (secondary; articles not read directly) (Outlook Business; IANS, May 2026). The IndiaAI Mission's outlay is INR 10,371.92 crore, about $1.1 billion (Wikipedia, AI in India, secondary). At the February 2026 AI Impact Summit, 88 countries and organisations endorsed a non-binding New Delhi Declaration, the count reported on the day; a later Observer Research Foundation analysis gives 91 (ORF). The electronics minister said more than $200 billion of AI and deep-tech investment is "expected" over two years, which is a forecast, not committed capital (Fortune).

India's rules are lighter in form than the EU's, but they carry hard dates. The governance guidelines of 5 November 2025 favour existing law and voluntary compliance (Indian Infrastructure). The binding instruments are the data-protection law, the IT Rules and CERT-In's 2022 Directions, which require service providers and body corporates to report listed cyber incidents, including attacks on AI/ML systems, within 6 hours (CERT-In Directions, 28 Apr 2022). Under the DPDP Act, consent-manager provisions start on 13 November 2026. Notice, consent, security, breach and data-subject-rights duties start on 13 May 2027 (Egazette, G.S.R. 843(E)). Penalties reach up to INR 250 crore for security-safeguard failures (Egazette, DPDP Act). The IT Rules amendment, in force since 20 February 2026, requires labels on synthetic audio and video and cuts the time allowed to act on government or court takedown orders from 36 hours to three, and to two hours for complaints about non-consensual intimate imagery (secondary) (iPleaders; Business Today). The DPDP Rules also require a detailed breach report to the Data Protection Board within 72 hours (MeitY, DPDP Rules 2025). The RBI's FREE-AI framework (13 August 2025) sets seven principles and 26 recommendations for banks and their vendors, and is not yet binding (Outlook Money; Bar and Bench). CERT-In's guidelines of 10 June 2026 (CISG-2026-03; title and date on CERT-In's listing, body not read, binding status unknown) reportedly ask technology providers for a secure development lifecycle, AI-assisted security testing, software bills of materials and faster patching (TeamLease RegTech, one secondary summary). The India document has the full picture, including Indian models and pricing.

What this means for your business

Your staff are probably ahead of your processes. In nearly every market where both are measured, personal use runs well ahead of company use; Denmark, Finland and Japan are the exceptions. Small firms also lag large ones by two to three times (in the EU, 18.90% of SMEs against 55.03% of large firms). Start by finding out which AI tools your staff already use informally, decide which are approved, and pick one or two processes to automate properly. That takes workflow design, integrations, governance and training, which a model subscription alone does not provide. The playbook and tool stack documents set out the steps.

India, the US and the EU now all have dated compliance obligations. Documentation is cheaper to build in at the start than to add later, so put these dates in your project plan and ask any vendor how its system meets them. The table's third column is inference, not measured fact.

In India, the 13 May 2027 data-protection deadline will reach every SME that handles customer data through chat, CRM or commerce automations; consent-manager provisions start on 13 November 2026. The DPDP Act has no "legitimate interests" ground, so reusing customers' personal data to train or evaluate models beyond the stated purpose likely needs fresh consent, an inference worth confirming with counsel (Hogan Lovells). For Indian-language voice and text, Sarvam's list prices give a starting budget: speech-to-text at Rs 30 per audio hour and text-to-speech at Rs 3 per 1,000 characters (Sarvam API pricing). Cheap Chinese open-weight models carry licence conditions for large hosted services: the Qwen3.8-Max licence requires a separate agreement above US$50M of annual revenue from model-as-a-service or AI-assistant businesses (Qwen3.8-Max licence).

In the US, check whether any automation supports consequential decisions. From 1 January 2027 Colorado requires developers of such tools to document intended uses, training-data categories, limitations and human-review instructions (Colorado General Assembly). If you buy such a tool, ask for that documentation; if you build one, write it once and keep it current.

Plan for a model being switched off. The June 2026 suspension cut off two frontier models for 19 days (Anthropic). Design automations that can switch models, and test the fallback before you need it. The OpenAI, Gemini and top models documents compare the alternatives.

Questions to ask any vendor or implementation partner: Which dated duties in the table apply, and who produces the records? Where is the data processed, and is it used to train models? What is the fallback model, and when was it last tested? Which licence, and which revenue thresholds, apply to any open-weight model used? Can an incident be reported within CERT-In's 6 hours and the DPDP Rules' 72 hours?

These rankings will keep moving: Microsoft's world figure rose a full point in one quarter while the IMF scores still use 2023 data, and Korea, Vietnam and the EU brought binding laws into application within about seven months of each other. The quarterly usage data and the legal dates are the series worth tracking. The future outlook considers where the curves go next.

Open questions and unverified items

• Microsoft's coverage change. Microsoft says its next report will add more AI tools, which it expects to raise measured use in almost every economy and most in China. Quarter-on-quarter comparisons across that change will need care. (The Q2 2026 ranks, including the US at 21st and New Zealand at 44.1%, are now read from Microsoft's data table.)
• Stanford's country adoption figures (UAE 64%, Singapore 61%, US 28.3%, 24th) match Microsoft's H2 2025 population values exactly, but Stanford's page does not name the source.
• Anthropic's June 2026 release: values for the UK, Canada, Germany, France and the other European and most American countries were not extracted, and no ranks were computed from the file.
• Oxford Insights and Tortoise ranks are secondary. The primary pages were blocked. Nigeria's Oxford rank is 70th or 72nd depending on source, and China's 8th needs checking against the report.
• IMF AIPI has no edition newer than 2023 data. Taiwan is not in the dataset.
• TOP500 counts for the top 100 and ranks 1-10 were re-checked on 11 October 2026. Ranks 101-500 were not read.
• KPMG and Ipsos country values were not available: the KPMG country PDFs were encrypted and the Ipsos AI Monitor was not found.
• Enterprise adoption has no official figure for China, India, South Korea or the Gulf states. The UK's ONS gives 25% (late December 2025; primary); a 29% figure for June 2026 ("nearly three in 10"; 49% for firms with 250+ staff) is single-source news, and Scotland's 36.9% (September 2026) was not read. Japan's figures range from 16% to 34.5% depending on the definition used.
• Regulation not researched or unresolved (Asia, the Gulf, Europe and the US): the UAE's national strategy and the rules in Saudi Arabia, Qatar, Switzerland and Malaysia. OECD.AI lists no strategy for Malaysia or Qatar, and its UAE, Saudi Arabia and South Africa pages returned errors; neither means "no strategy". For the UK, only the formal date and cause of DSIT's replacement stay open. Israel's, Turkey's, Indonesia's and Thailand's OECD.AI-listed strategies were not read in full, and Indonesia's two draft Presidential Regulations (AI roadmap; AI ethics with risk classes) are pending, expected by end-2026. Also open: New York's RAISE Act chapter amendments (reported signed 27 March 2026, secondary, while Wikipedia's February 2026 text still had them in committee), whether the X.AI v. Weiser stay reaches Colorado's SB26-189, Colorado's final rules after the 26 October 2026 hearing and the year of Malaysia's chip-permit rule.
• Regulation unresolved (Africa and Brazil): Kenya's AI Bill 2026 has no confirmed parliamentary status: secondary reports conflict on whether it has been tabled, and the Senate order paper was not checked. South Africa has no replacement timetable after withdrawing its draft policy on 26 April 2026 (Wikipedia gives 27 April). Brazil's post-election vote timing rests on one MobileTime report (24 August 2026, not re-read); the Chamber record sets no vote date.
• India specifics: the conflicting Microsoft India investment figures; the undated IndiaAI empanelled-GPU PDF (45,081); and the data-centre forecast of 1.6 GW (June 2026) to 6 GW by 2029, which rests on IANS reporting a JLL report (JLL release not read), while other 2030 forecasts differ (Jefferies about 8 GW, Avendus about 3 GW, an IANS range of 4-8 GW; all secondary).
• Gulf compute milestones: Stargate UAE's first 200 MW has two published targets ("by 2026" from Gulf News and G42; "third quarter" of 2026 from Mubadala) and no completion or operating notice by 12 October 2026. HUMAIN's capacity figures conflict (100 MW sites from Q2 2026 and about 1.9 GW by 2030; 250 MW from 2027 and up to 1 GW by 2030; 6 GW by 2034), and no go-live notice was found for the Riyadh or Dammam sites.
• China's enterprise adoption and total compute: CNNIC gives a manufacturing-only adoption figure (more than 30% of large manufacturers) and 2,185 exaflops of intelligent computing with no stated precision. No economy-wide enterprise survey or comparable compute total was found.

Sources

1. Microsoft AI Diffusion Report 2026 Q2 (PDF)
2. Microsoft AI Diffusion Report 2026 Q1 (PDF)
3. Microsoft On the Issues, the state of global AI diffusion in 2026, 7 May 2026
4. Microsoft technical paper, arXiv:2511.02781
5. Middle East AI News, UAE retains top world AI use ranking, 25 Sep 2026
6. MediaPost, Microsoft tracks AI adoption, 22 Sep 2026
7. Anthropic Economic Index, September 2025 report
8. Anthropic Economic Index, India Country Brief, 16 Feb 2026
9. Anthropic Economic Index, January 2026 report
10. Anthropic Economic Index, June 2026 country data (CSV) (country file read for the countries in this page's tables)
11. Anthropic Economic Index, June 2026 data documentation
12. NRIGlobe, Israel per-capita Claude usage, Mar 2026
13. Anthropic, Bengaluru office and partnerships across India
14. Anthropic, Fable 5 and Mythos 5 access update, 12 Jun 2026
15. IMF DataMapper API, AI Preparedness Index (AI_PI)
16. Khan Global Studies, AI Preparedness Index summary
17. Stanford HAI, AI Index 2026, Economy chapter
18. Stanford HAI, AI Index 2026, Public Opinion chapter
19. Stanford HAI, AI Index Report 2026 (PDF), pp. 181-184
20. Oxford Insights, Government AI Readiness Index
21. Egypt Today, Egypt tops Africa in Oxford Insights index
22. CADE project, Oxford index summary
23. Let's Data Science, Nigeria rises to 72nd in AI readiness
24. Al Riyadh Daily, Saudi Arabia in the Oxford index
25. Comland, Global AI Index 2025
26. TOP500, June 2026 list
27. TOP500, November 2025 list
28. Epoch AI, GPU clusters dataset
29. KPMG and University of Melbourne, Trust, attitudes and use of AI 2025
30. OpenRouter, rankings
31. Eurostat API, isoc_eb_ai 2025
32. WAM, 20% of EU enterprises use AI, citing Eurostat
33. Implicator.ai, Denmark leads Europe in business AI use
34. US Census Bureau, AI use by businesses, May 2026
35. Federal Reserve FEDS Note, monitoring AI adoption, 3 Apr 2026
36. Statistics Canada, Business conditions in Canada, Q2 2026
37. Pew Research, about half of US adults use AI chatbots
38. Pew Research, about a quarter of US adults use chatbots daily
39. Asia News Network, Japanese firms using generative AI (Teikoku Databank)
40. Mezha, more than 80% of Japanese companies use AI only in specific processes
41. ECNS, CNNIC generative AI report, 29 Sep 2026
42. CRN Asia, China's generative-AI user base hits 515 million
43. gov.cn / Xinhua, AI+ opinions, 29 Aug 2025
44. Digital Policy Alert, CAC generative-AI filings, May-June 2026
45. Loeb & Loeb, China's AI labelling measures, Mar 2025
46. Caixin Global, China amends Cybersecurity Law to add AI clause, 29 Oct 2025
47. TechCrunch, India has 100M weekly active ChatGPT users, 15 Feb 2026
48. Business Today, OpenAI bets big on India, 19 Jan 2026
49. AI Weekly, hyperscalers plan $725B AI capex in 2026
50. Yahoo Finance, hyperscalers hit $700 billion in 2026
51. Bloomberg Law, OpenAI meets AI computing capacity goal
52. EuroHPC JU, AI Gigafactories call, 30 Jul 2026
53. StorageNewsletter, EU AI gigafactories call, 18 Aug 2026
54. SiliconANGLE, France announces EUR 109B AI investment, 10 Feb 2025
55. UK Parliament, written statement on AI Growth Zones (May 2026; not re-read on 12 Oct 2026)
56. The Logic, Cohere data centre and Canada's compute strategy
57. TechRepublic, Nvidia and South Korea AI chips
58. Wikipedia, TSMC
59. Real Estate Asia, Johor on track to be 1 GW data-centre market
60. Gulf News, Stargate UAE first 200 MW phase
61. The National, US raises UAE export status, 19 Jul 2026
62. CryptoBriefing, HUMAIN data-centre fund
63. MEA Tech Watch, SDAIA Hexagon data centre, 2 Jan 2026
64. IANS, India's data-centre capacity expected to reach 6 GW by 2029, 2 Sep 2026 (secondary, reporting a JLL report)
65. TelecomTalk, IndiaAI Mission GPUs, citing MeitY (earlier announcement)
66. Communications Today, IndiaAI GPU hours sanctioned
67. Fortune, India's AI Impact Summit closes, 23 Feb 2026
68. IDC, Mexico's national AI plan, 9 Sep 2026
69. Times Brasil, Brazil AI plan R$23 billion
70. Africa.com, Cassava AI factory investment
71. Further Africa, Ghana launches $270m AI strategy, 29 Apr 2026
72. European Commission, AI Act regulatory framework
73. NatLawReview, EU AI Omnibus ("Digital Omnibus on AI") enters force
74. European Commission, AI literacy questions and answers
75. Gazzetta Ufficiale, Law 132 of 23 Sep 2025
76. Cooley, South Korea's AI Basic Act, 27 Jan 2026
77. The Daily Star, South Korea rolls out AI law
78. Baker McKenzie, Vietnam AI law, Feb 2026
79. Baker McKenzie, Taiwan AI Basic Act, Jan 2026
80. Future of Privacy Forum, Japan's AI Promotion Act
81. CMS, Singapore agentic AI governance framework
82. MinterEllison, Australia's National AI Plan
83. LEGISinfo, Bill C-27 (44-1)
84. ISED, Overview of Canada's National AI Strategy
85. Federal Register, EO 14365, Ensuring a National Policy Framework for AI
86. Clifford Chance, California SB 53, Oct 2025
87. Nelson Mullins, Texas HB 149 (TRAIGA)
88. Colorado General Assembly, SB26-189
89. Colorado Attorney General, ADMT and Chatbot Safety rulemaking
90. NYC DCWP, automated employment decision tools
91. Câmara dos Deputados, PL 2338/2023
92. MobileTime, Brazil AI bill vote at year end, 24 Aug 2026
93. IAPP, Chile AI bill approved in first stage
94. DNP, CONPES 4144 (PDF)
95. Ecosistema Startup, Mexico AI law 2026
96. OECD.AI, Nigeria National AI Strategy
97. Digital Policy Alert, Nigeria AI governance approach
98. The Standard, Kenya bets Sh152 billion on AI
99. OECD.AI, Egypt National AI Strategy
100. gov.uk, Department for Science, Innovation and Technology
101. Focus Malaysia, permit rule for US AI chips
102. Wikipedia, 2026 in artificial intelligence
103. Ministry of IT and Telecommunications, Pakistan
104. Wikipedia, Humain
105. Qatar Tribune, Qai launch
106. The Verge, Iran threatens Stargate UAE, 6 Apr 2026
107. Engineering News, Egypt's first national AI factory, 25 Sep 2026
108. Technology Times, Nigeria's N-ATLAS
109. Newswav, PayNearby MSME Digital Index 2026
110. Entrepreneur India, AI talent pool to touch 1.25 million by 2027
111. Careers360, IIM Ahmedabad AI/ML talent study
112. Wikipedia, Artificial intelligence in India
113. Indian Infrastructure, India AI Governance Guidelines, 7 Nov 2025
114. Egazette, DPDP commencement notification G.S.R. 843(E) (PDF)
115. Egazette, Digital Personal Data Protection Act 2023 (PDF)
116. Business Today, rules on AI-generated content, 10 Feb 2026
117. Outlook Money, RBI panel proposes ethical AI framework
118. Bar and Bench, RBI FREE-AI framework, 18 Sep 2025
119. Hogan Lovells, India's DPDP Act brought into force, 17 Nov 2025
120. Sarvam AI, API pricing
121. MeitY, Digital Personal Data Protection Rules 2025, G.S.R. 846(E) (PDF)
122. TeamLease RegTech, CERT-In guidelines on AI-accelerated vulnerability protection, Jun 2026
123. Qwen3.8-Max licence (Hugging Face)
124. Eurostat API, isoc_eb_ai 2024
125. IAPP, South Korea's AI Act designed to be all roar, no bite (opinion)
126. Digital Policy Alert, CAC generative-AI filing list tracker
127. The National, Stargate UAE's first phase to be completed in Q3 2026, 5 Dec 2025
128. AMD, AMD, Cisco and HUMAIN expand Saudi Arabia's AI infrastructure as AMD Instinct systems go live, 31 Aug 2026
129. Hugging Face, NCAIR1/N-ATLaS model card and licence
130. Observer Research Foundation, strategic outcomes of the India AI Impact Summit 2026
131. Al Jazeera: US lifts restrictions on powerful AI models Fable, Mythos, Anthropic says (1 Jul 2026)
132. Anthropic Economic Index dataset, Hugging Face (June 2026 release, 26 Jun 2026)
133. ONS, Business insights and impact on the UK economy, 8 Jan 2026 (BICS)
134. Pulse 2.0, EU launches AI gigafactories call (secondary)
135. GOV.UK, AI Growth Zones collection
136. GOV.UK, Ministerial appointments: July 2026
137. legislation.gov.uk, Data (Use and Access) Act 2025, section 80
138. silicon.co.uk, Saudi AI data centres (citing Bloomberg) (secondary)
139. IndiaAI compute portal, empanelled GPUs (PDF, undated)
140. IndiaAI compute portal, price calculator (updated 8 Oct 2026)
141. IndiaAI compute portal, Ready Reckoner (PDF)
142. Outlook Business, India's data-centre capacity likely to reach 3 GW by 2030 (secondary; not read directly)
143. IANS, India's data-centre capacity likely to grow up to 8 GW by 2030, May 2026 (secondary; not read directly)
144. Connecting Africa, Cassava deploys AI factory in South Africa
145. HapaKenya, summary of the Artificial Intelligence Bill 2026, 16 Mar 2026 (secondary)
146. OECD.AI, Indonesia national AI policies (OECD.AI listing, secondary)
147. OECD.AI, Thailand national AI policies (OECD.AI listing, secondary)
148. OECD.AI, Israel national AI policies (OECD.AI listing, secondary)
149. OECD.AI, Türkiye national AI policies (OECD.AI listing, secondary)
150. OECD.AI, Egypt national AI policies (OECD.AI listing, secondary)
151. Tech For Good Institute, Indonesia's AI developments in 2026, 10 Jul 2026 (secondary)
152. Daily Maverick, Malatsi withdraws draft AI policy, 27 Apr 2026 (secondary)
153. Digital Policy Alert, Vietnam decree implementing the Law on AI (AI authority governance) (secondary)
154. Adnkronos (Jiji), Japan adopts first Basic Plan on AI, 23 Dec 2025 (secondary)
155. industry.gov.au, Australia establishes AI Safety Institute, 24 Nov 2025 (headline read)
156. CourtListener, X.AI LLC v. Weiser (D. Colo. 1:26-cv-01515) docket
157. Texas Legislature, HB 149 enrolled text
158. Pew Research Center, globally more people expect AI to cause job loss than growth, 17 Sep 2026
159. CERT-In, Directions under section 70B(6), 28 Apr 2022 (PDF)
160. iPleaders, IT Rules 2026 amendment (secondary)