AI in the Middle East and Africa, Country by Country

Facts as of 12 October 2026; the Stargate, Saudi compute, per-capita, strategy and policy items were re-checked against primary sources where available that day, and items not re-read keep their 11 October reading. This document covers the Gulf states, Israel, Turkey, North Africa, Sub-Saharan Africa and the African Union. Global comparisons are in the country overview. India, the Americas and Europe and Asia-Pacific have their own documents. Labels used below: "secondary" means a news outlet or aggregator reporting someone else's figures; "single source" means only one source was found; "Wikipedia-only" means the fact comes from a Wikipedia article whose underlying reference was not opened; "headline only" means only a headline or listing was read; "search summary" means the figure came from a search-result summary and the page itself was not opened; "OECD.AI listing" means an entry in the OECD's national AI policy navigator (secondary). Unlabelled facts come from a primary source or an outlet read directly. Coverage is uneven. Some countries have only one or two data points, and the text says so.

Key takeaways

• The Gulf leads the world in measured AI use, and most of Africa trails the global average. Microsoft's June 2026 data put the UAE first worldwide at 73.3% of working-age people, Qatar 10th at 43.5%, Israel 15th at 38.7% and Saudi Arabia 25th at 31.4%, up from 30th. South Africa (23.9%) is the only Sub-Saharan country above the 18.8% world figure. Turkey (18.9%) sits on that figure. Egypt (15.5%), Morocco (12.0%), Nigeria and Ghana (both 10.4%), Kenya (8.9%), Ethiopia (8.0%) and Rwanda (7.6%) fall below it.
• Gulf compute runs on US chips under US licences. The US raised the UAE to "A5" export status on 10 July 2026, two months after the first Nvidia shipments arrived. Stargate UAE's first 200 MW has two published targets: "to complete by 2026" (Gulf News, 16 Oct 2025) and "the third quarter" of 2026 (Mubadala's chief executive, The National, 5 Dec 2025). No completion or operating notice had appeared by 12 October 2026. Saudi Arabia's HUMAIN is the only large Gulf project in the research to report running hardware: AMD says its MI355X systems went live for HUMAIN customers on 31 August 2026, but gives no capacity figure. Qatar's new national AI company needs US licences for its chips.
• Gulf capital flows out as fast as chips flow in. Abu Dhabi's MGX is a partner in the US Stargate venture, co-led Anthropic's $30bn Series G in February 2026 (Qatar's QIA also invested) and backs a 1 GW campus in France. Two UAE entities supplied about 86% of chipmaker Cerebras's 2025 revenue (per its IPO filing, as relayed by Wikipedia).
• National models are few, and their licences need checking. Saudi Arabia's ALLaM powers a free consumer app. Nigeria's N-ATLaS is a Llama-3 8B fine-tune for Yoruba, Hausa, Igbo and Nigerian English, released with four speech-recognition models. Its licence caps free use at 1,000 users and requires a separate licence for commercial use. The new Falcon models that the UAE's TII published in 2026 are either small (under 1B parameters) or built for OCR and vision. Its other 2026 uploads are quantised versions of earlier models, its newest 7B language model dates from October 2025, and Jais could not be verified. Qatar's Qai says it will not build its own large language models at all.
• Governments are the anchor buyers. The UAE cabinet adopted a framework in April 2026 to put agentic AI into half of government sectors, services and operations within two years (single outlet). Saudi Arabia's data authority broke ground on a 480 MW government data centre in January 2026.
• Africa has strategies, but little public compute and almost no AI law. Nigeria, Kenya, Egypt, Morocco, Ghana, Rwanda and Ethiopia have national AI strategies or policies (Ethiopia's 2024 policy per OECD.AI's listing), and a 2024 African Union strategy frames them. South Africa withdrew its draft national AI policy in April 2026 after its reference list was found to contain fictitious citations. Compute is mostly private, led by Cassava with Nvidia. No binding AI-specific law was verified anywhere in the region, and one policy tracker says Nigeria has none. Kenya's risk-based AI Bill 2026 is still a bill.
• For a business automating in or selling to the region, the research points to three conclusions (inferences). Gulf government agent programmes set the pace, and Arabic-language capability matters there. Egypt joins India and Vietnam as a source of AI-assisted software work. Gulf-hosted deployments need model and hosting portability, because US licensing decisions now reach Gulf capacity directly.

Regional summary: the region on one page

Each cell is sourced in its country section below. "n/a" means the research did not find or extract the value. It does not mean the value does not exist. The usage values for Jordan, Oman, Kuwait, Iran, Morocco, Ghana, Rwanda and Ethiopia, and the IMF values for Qatar and those countries, were read from the primary Microsoft report and IMF API on 11 October 2026.

A nearly tenfold usage gap from Abu Dhabi to Kigali

Why the measures disagree

The numbers in the summary table measure different things, so they should not be merged into one ranking. Microsoft's AI User Share estimates the share of people aged 15 to 64 who used a generative-AI product in the period. It is built from aggregated, anonymised Microsoft telemetry and adjusted for operating-system and device share, internet penetration and population. It therefore measures reach, not depth or business value (Microsoft AI Diffusion Report Q2 2026). Worldwide use was 18.8% in June 2026, split into 28.8% in the Global North and 16.2% in the Global South (same source). Microsoft itself says "no single metric is perfect, and this one is no exception" (same source). Its table also gives identical values to whole groups of African economies. Nigeria, Ghana and ten other West African economies all show 10.1% in Q1 and 10.4% in Q2, and Ethiopia, Tanzania, Uganda and five other East African economies all show 7.6% and 8.0%. The report does not explain this, but it suggests regional estimates where telemetry is thin, so small gaps between these countries mean little (inference). The IMF AI Preparedness Index is a 0-to-1 macro-structural score. Its primary API covers 178 economies and returns only 2023 data, the 2024 release (IMF DataMapper, AI_PI, read 11 Oct 2026). Anthropic's per-capita usage index (usage_per_capita_index) divides a country's share of Claude.ai usage by its share of the working-age (15-64) population, with 1.0 meaning proportional. Its June 2026 release (26 Jun 2026; April and May 2026 data; Claude.ai chat and Cowork on Free, Pro and Max plans only) gives, April then May: Israel 3.03 / 3.05; UAE 2.73 / 2.84; Qatar 1.66 / 1.66; Morocco 0.86 / 0.90; Saudi Arabia 0.79 / 0.93; Turkey 0.68 / 0.64; South Africa 0.43 / 0.46; Ghana 0.40 / 0.38; Kenya 0.34 / 0.34; Egypt 0.31 / 0.36; Nigeria 0.21 / 0.24; Rwanda 0.18 / 0.20 (Anthropic Economic Index data, primary). The September 2025 table, from a one-week August 2025 sample, is a different release and should not be compared with these values (Anthropic Economic Index, Sept 2025). Stanford's AI Index 2026 quotes country adoption rates of 64% for the UAE, 61% for Singapore and 28.3% (24th) for the US. These match Microsoft's H2 2025 user-share values exactly (64.0%, 60.9% and 28.3%, 24th), so they restate Microsoft's population measure and are not independent evidence (inference from the match) (Stanford AI Index 2026, Economy; Microsoft Q1 2026). Stanford's other country measure, the share of employees using AI at work, does not name its survey on the page read (Stanford AI Index 2026, Public Opinion). TOP500 counts only the supercomputers that operators choose to submit, and its country counts here were parsed by row matching (TOP500, June 2026). Earlier Microsoft values (H1 2025, H2 2025) come from the comparison columns of its Q1 2026 report. Its H1 2025 column matches the technical paper (arXiv:2511.02781).

The Gulf: the world's highest usage, US-licensed compute and outbound capital

The Gulf states run AI policy through state-owned companies and sovereign funds, not through legislation. Their two binding constraints in 2025 were US chip-export approval and construction. By October 2026 the first had eased for the UAE. The second was still largely unproven: only HUMAIN reported live capacity, and without a size.

United Arab Emirates: first in usage, still waiting on its first 200 MW

First in the world on reach, though not on every measure

The UAE tops Microsoft's usage leaderboard. Microsoft's June 2026 figure of 73.3% is up from 70.1% in Q1 2026, 64.0% in H2 2025 and 59.4% in H1 2025, a gain of 13.9 points in a year (Microsoft Q2 2026; Microsoft Q1 2026). The Q1 to Q2 rise of 3.2 points is the largest absolute gain among the region's measured countries. Singapore is second worldwide at 64.3%, so the UAE leads it by nine points (same source). WAM and Middle East AI News both reported the ranking (WAM; Middle East AI News, 25 Sep 2026). Stanford's 2026 AI Index repeats the lead with 64% for the UAE, but that figure is Microsoft's H2 2025 value, not a separate survey of firms (see above). The workplace survey is independent: more than 80% of UAE employees say they use AI at work semi-regularly or regularly, against a 58% global average (Stanford AI Index 2026, Public Opinion). Stanford also names the UAE among the three places where AI engineering skills are growing fastest (see South Africa below).

Other measures show less of a lead. The UAE's Claude usage index was 2.73 in April and 2.84 in May 2026, meaning its share of Claude use was nearly three times its share of the world's working-age population. These values come from Anthropic's June 2026 country file (Claude.ai chat and Cowork; Free, Pro and Max plans; April and May 2026) and cannot be compared with the September 2025 country table, which used a different sample (Anthropic data, June 2026 release). The IMF preparedness score of 0.628 is well below Israel's 0.725 (IMF). The UAE's lead is therefore a lead in reach rather than in macro-structural readiness. The research looked for an explanation, such as a nationwide chatbot access deal or AI in the school curriculum, and found none. The national AI strategy site blocked automated access.

Stargate UAE: 1 GW announced, 200 MW still unconfirmed

Stargate UAE is the region's flagship compute project. It was announced on 22 May 2025 by G42, OpenAI, Oracle, Nvidia, SoftBank and Cisco (Wikipedia: Stargate LLC, Wikipedia-only). It is a 1 GW AI cluster in Abu Dhabi. G42 builds it through its data-centre company Khazna, and OpenAI and Oracle operate it (OpenAI, "Introducing Stargate UAE"; Enterprise AM, 23 May 2025; Gulf News). Its first 200 MW is also described as the start of a planned 5 GW UAE-US AI Campus. At GITEX in October 2025, G42's Talal Al Kaissi said that 200 MW should be operational by the end of 2026 and that G42 was in deep talks with other US hyperscalers for the remaining 4 GW (MIT Sloan Management Review Middle East, undated). The same month G42 said construction was "well underway", all long-lead equipment had been procured and the first 200 MW was "on track for handover in 2026" (MEA Tech Watch, 17 Oct 2025).

A G42 spokesperson told Gulf News the same week that the rest of the 1 GW cluster and the 5 GW campus had no confirmed completion timelines (Gulf News, 16 Oct 2025). The most specific date came in December 2025. Mubadala's chief executive, Khaldoon Al Mubarak, said the first 200 MW "will be in place and ready at some point in the third quarter of next year". He added that more than 5,000 workers were on site and more than 100,000 cubic metres of concrete had been poured (The National, 5 Dec 2025). Nothing later confirms that it is running. The Verge wrote on 6 April 2026 that "it's not clear how much of the Abu Dhabi data center is actually finished", citing nothing newer than the October 2025 update. It also gave figures of $30bn and "16 gigawatts" that conflict with the 1 GW used everywhere else, so this document does not use them (The Verge, 6 Apr 2026). As of 12 October 2026, Mubadala's Q3 2026 date has passed and no completion or operating notice has been found; the "by 2026" target (Gulf News and G42, October 2025) is the remaining milestone. Stargate sites elsewhere have slipped: Stargate UK was paused in about April 2026 over energy costs and regulation (secondary, several outlets; TechRadar), and Microsoft took over the Norway capacity (headlines only, Hacker News listing). For scale, Epoch AI's 93 tracked AI data centres total 13.8 GW of IT capacity, so a finished 1 GW cluster would equal about 7% of everything Epoch tracks (Epoch AI; the percentage is a calculation).

Chips arrived after an export upgrade and a set of assurance deals

The UAE secured US chips by accepting US conditions on how they are used. According to Wikipedia, the US Commerce Department authorised exports equivalent to 35,000 Nvidia Blackwell chips to G42 in November 2025 (Wikipedia: G42). Mubadala's chief used the same number in December 2025, saying the UAE would go from "a couple of thousand" chips to 35,000 "through this agreement" (The National, 5 Dec 2025). Microsoft and G42 announced a 200 MW data-centre expansion in the UAE on 5 November 2025 (Reuters headline in the same article's references, not opened). G42 launched a "Digital Embassies" framework in January 2026. In February 2026 it signed a framework on governance and assurance for AI infrastructure built on US-origin chips (same source). The UAE's ambassador to the US, Yousef Al Otaiba, announced in May 2026 that the first Nvidia chip shipments had arrived. On 10 July 2026 the US upgraded the UAE to "A5" export status, easing transfers of advanced AI and military technology. Core42's chief executive, Talal Al Kaissi, said the UAE had signed an intergovernmental assurance agreement that led to a "regulated technology environment", designed with input from the US government and Microsoft (The National, 19 Jul 2026, re-read 11 Oct 2026). A Bloomberg headline of the same day read "US Eases Export Curbs on UAE, Opening Door for AI Chip Sales" (headline only, via Wikipedia). China, by contrast, moved only to case-by-case licence review for Nvidia H200 and AMD MI325X chips from 15 January 2026, with performance caps and conditions (Gigazine, 15 Jan 2026). With the licences in place, the UAE's compute timeline now depends mainly on construction, power and regional security.

MGX and G42 export capital as well as importing chips

Abu Dhabi is a large investor in the US AI supply chain as well as a customer of it. MGX was set up in 2024 by Mubadala and G42 with a target of $100bn in assets under management. Wikipedia reports that it raised close to $50bn for an AI infrastructure fund in June 2026 (Wikipedia-only, single source). It also took part in a $40bn Aligned Data Centers deal in October 2025 and a 1 GW campus in France with Mistral and Nvidia in 2025 (Wikipedia: MGX). At the February 2025 Paris summit, the UAE's pledge to France was reported at EUR 30-50bn, with MGX expected to provide the initial financing. Figures vary by outlet (The Media Leader, French). MGX is one of the four partners in Stargate LLC, the US joint venture with OpenAI, SoftBank and Oracle that plans up to $500bn of US AI infrastructure by 2029 (Wikipedia: Stargate LLC). It also co-led Anthropic's $30bn Series G at a $380bn post-money valuation on 12 February 2026. Qatar Investment Authority is listed among the round's other significant investors (Anthropic, re-read 11 Oct 2026). On the buying side, Cerebras's IPO filing, as relayed by Wikipedia, attributed 62% of its 2025 revenue of $510m to MBZUAI and 24% to G42 (Wikipedia: Cerebras, Wikipedia-only). Together that is about 86%, or roughly $440m (calculation). The UAE is therefore more than an export market for US AI. It funds the US labs and data centres it then buys from, and that two-way stake plausibly strengthens its claim on scarce chips (inference).

The government wants agents in half its services within two years

The UAE government also plans to be a major AI customer itself. Middle East AI News reports that the cabinet adopted a "world-first" framework in April 2026 to deploy agentic AI across 50% of government sectors, services and operations within two years. It also reports that an AI-powered Proactive Government Performance System already tracks 150 million data points a month (Middle East AI News, 25 Sep 2026, single outlet; the government statement was not read). The minister in charge, Omar Sultan Al Olama, has held the AI portfolio since October 2017. Since 2023 he has also been Director General of the Prime Minister's Office (Wikipedia: Omar Sultan Al Olama). At least one vendor is hiring delivery staff locally. OpenAI's job board lists Abu Dhabi among the non-US locations for forward-deployed engineers, the staff who embed with customers to build production deployments (OpenAI job board API). Chinese technology is present as well. Baidu's Apollo Go runs a fully driverless commercial service in Dubai in partnership with Dubai Taxi and plans a fleet in Abu Dhabi with Autogo (Wikipedia: Apollo Go, Wikipedia-only).

Falcon and Jais: small new models and thin evidence

The evidence on the UAE's own models is uneven. TII's eight most-downloaded models on Hugging Face date from 2023 to 2025: the 2023 falcon-7b models under Apache-2.0, and falcon-mamba-7b (July 2024), Falcon3-7B-Instruct (November 2024) and Falcon-H1 (0.5B to 7B, May 2025) under a custom licence (Hugging Face API, tiiuae). Sorted by date, the same account shows that TII kept publishing in 2026, but at small scale. It released the Falcon-H1-Tiny family in January 2026 (90M to 0.6B parameters by their names, custom licence), then Falcon-OCR and Falcon-Perception in February under Apache-2.0. In April it posted quantised versions of the earlier Falcon-E-3B and Falcon3-10B. The newest 7B language model is Falcon-H1R-7B, from October 2025 (Hugging Face API, tiiuae, newest first, read 11 Oct 2026). The research could not verify the current status of Jais. The practical point is that Falcon's licence changes between models, even within one year, and must be checked model by model (see Other models).

Saudi Arabia: HUMAIN's gigawatt targets and the first systems going live

Saudi Arabia runs AI through a single sovereign champion. Crown Prince Mohammed bin Salman launched HUMAIN under the Public Investment Fund on 12 May 2025 with Tareq Amin as chief executive. It signed partnerships with Nvidia, AMD and Qualcomm that year and plans "several gigawatts" of capacity (Wikipedia: Humain). The launch deals were large. Nvidia will supply several hundred thousand GPUs over five years, starting with an 18,000-unit GB300 Grace Blackwell supercomputer on InfiniBand. The AMD partnership is worth up to $10bn over five years to deploy 500 MW of compute on AMD's ROCm software, with multi-exaflop capacity targeted for early 2026 (The National, 14 May 2025; TechFinitive). AWS later expanded its deal to deploy Nvidia GPUs and its own Trainium chips, up to 150,000 accelerators in a Riyadh "AI Zone" (Seeking Alpha; November 2025 date from a search summary). In 2026 HUMAIN added a European partner: Mistral announced a partnership on 24 August without giving details (Mistral news). A week later, on 31 August 2026, AMD said that MI355X GPUs on Cisco networking were "now live in Saudi Arabia and serving HUMAIN customers" as GPU-as-a-service. The release gives no megawatt or GPU count. It adds plans for up to 250 MW of MI400-series capacity from 2027, and says the AMD-Cisco-HUMAIN venture "remains on track to deploy up to 1 GW of AI infrastructure by 2030" (AMD press release, 31 Aug 2026, vendor claim, read 11 Oct 2026).

The targets run well ahead of verified delivery, and they are stated on different bases. HUMAIN aims for about 1.9 GW of AI data-centre capacity by 2030 (silicon.co.uk, citing Bloomberg; cairoscene; secondary), starting with Riyadh and Dammam data centres of up to 100 MW each planned from Q2 2026, and a search summary of a CryptoBriefing report gives 6 GW by 2034 (one report 6.6 GW) and a fund of up to $3bn for data-centre expansion (CryptoBriefing, undated, search summary). Separately, the national data-centre strategy issued in June 2025 by MCIT with SDAIA targets up to 1.5 GW by 2030 (trade.gov, search summary). The AMD venture's 1 GW by 2030 is a third figure. The targets are on different bases and may overlap, so they cannot be added. The state is also building for its own use. SDAIA broke ground in early January 2026 on "Hexagon", a 480 MW government data centre in Riyadh billed as the largest government data centre by megawatts (MEA Tech Watch, 2 Jan 2026; MIT Sloan ME). One listing gives a $2.7bn contract with Albawani as main contractor, and the sources disagree on whether it is rated Tier IV or Tier V (OGN News, single source). For scale, HUMAIN's 2034 goal of 6 GW equals about 43% of the 13.8 GW that Epoch AI tracks across 93 AI data centres worldwide (Epoch AI; calculation).

HUMAIN also has its own model. ALLaM is an Arabic-first multimodal LLM, and ALLaM 34B was reportedly trained on more than 500bn Arabic tokens. It powers Humain Chat, a free app on iOS, Android and the web, and Humain OS, launched in October 2025 (Wikipedia: Humain, Wikipedia-only). SDAIA's role in ALLaM, beyond HUMAIN's own description, was not verified. Measured adoption is rising but still trails the UAE, Qatar and Israel. Saudi Arabia went from 29.4% to 31.4% between Q1 and Q2 2026 and climbed from 30th to 25th. Microsoft calls that the largest rise in rank of any economy, though it leaves the country below the US at 33.0% (Microsoft Q2 2026; MediaPost, 22 Sep 2026). Its government readiness scores run ahead of its population's usage. Oxford Insights ranks it first in MENA, 7th in the world on governance and 9th on public-sector adoption (Al Riyadh Daily, secondary). It also holds four of the TOP500 top 100 systems, more than any other country in the region (TOP500). The open question is chip licensing. One report says HUMAIN's imports depend on US approvals (TechFinitive). Saudi Arabia approved locally the import of up to 18,000 Nvidia Blackwell chips for the Riyadh and Dammam sites, but US export clearance was described as still pending in early 2026, and no go-live notice for those sites was found (silicon.co.uk; cairoscene; secondary). This is a separate figure from Nvidia's 18,000-GB300 first phase and is not merged with it. The July 2026 export upgrade covers the UAE, and no comparable Saudi decision appears in the sources read. HUMAIN's live AMD systems suggest that some licences were granted, but no record of them was found (inference).

Qatar: top-ten usage and a national AI company that will not build LLMs

Qatar has the region's second-highest usage, at 43.5%, which places it 10th in the world, level with the Netherlands and just behind the UK at 43.7% (Microsoft Q2 2026). It was already 10th in Q1 2026, so one outlet's line that Qatar "entered the global top ten for the first time" is wrong; the same article also says it "retained 10th place" (Microsoft Q1 2026; Middle East AI News, 25 Sep 2026). Its IMF score of 0.535 is close to Turkey's and below Saudi Arabia's (IMF). Its strategy is unlike its neighbours'. QIA set up Qai to develop, operate and invest in AI infrastructure in Qatar and abroad (Gulf Times). According to one outlet citing Bloomberg, Qai "will not build its own large language models". It will instead evaluate and commercialise existing models, work on AI agents and seek US export licences for Nvidia and AMD chips (Qatar Tribune, single outlet). No investment amount or compute capacity for Qai was found. The size of its parent is disputed, at $524bn or $557bn (MEA Tech Watch). QIA, like Abu Dhabi's MGX, invested in Anthropic's Series G. The rest of the data on Qatar is thin: the government's own page refused automated access, and regulation and public-sector use were not researched. On the evidence available, Qatar is positioning itself as a buyer and integrator of other people's models, which makes it the Gulf state whose stated plan relies most on outside models and integration work (inference).

Oman, Kuwait, Bahrain and Iran: index values only, and one security warning

The research did not cover the policies of Oman, Kuwait, Bahrain or Iran. The index tables cover most of them. Microsoft's June 2026 figures are 27.6% for Oman, 22.6% for Kuwait and 12.8% for Iran, and Bahrain is not in Microsoft's table. Their IMF scores are 0.533 for Oman, 0.515 for Bahrain, 0.461 for Kuwait and 0.378 for Iran (Microsoft Q2 2026; IMF; both read 11 Oct 2026). Otherwise Iran appears only as a security factor. On 3 April 2026 the IRGC published a video threatening OpenAI's planned Abu Dhabi data centre if the US attacked Iran's power plants. The video came amid US threats against Iran over the Strait of Hormuz (The Verge, 6 Apr 2026). It is the first explicit state threat against a named AI data centre in the research, and no source estimates its effect on timelines or insurance. Iran's AI policy, compute and the effect of sanctions on its AI access were not researched.

Israel, Turkey and North Africa: high readiness, product blocks and new strategies

This group has the region's widest spread. Israel has the strongest structural base, Turkey enforces against specific products in court, and Egypt and Morocco have new strategies and their first AI-factory projects.

Israel: high Claude intensity, the region's top readiness score and a defence-AI surge

Israel is an intensive AI user per person, though not the region's widest. In Anthropic's June 2026 release its per-capita index was 3.03 in April and 3.05 in May 2026, above the UAE (2.73 and 2.84) but below Singapore (6.44 and 5.81) and the US (4.25 and 3.87) (Anthropic Economic Index data, primary). The September 2025 report's 7.0, then the world's highest, and a March 2026 article's 4.9x come from different releases or series and should not be compared with the 2026 values (Anthropic Economic Index, Sept 2025; NRIGlobe). On Microsoft's broader reach measure, Israel is 15th worldwide and third in the Middle East at 38.7%, behind the UAE and Qatar (Microsoft Q2 2026; rank by position in the table). The two results fit together: Claude intensity tracks income, and Anthropic's September 2025 report finds that a 1% rise in GDP per capita goes with a 0.7% rise in per-capita Claude use (Anthropic, Sept 2025). Israel's IMF score of 0.725 is the highest in the region, close to Australia's and South Korea's 0.727 (IMF).

The only investment signal in the research is in defence. Defence-tech startups working with the Defence Ministry raised nearly $3bn in the first half of 2026, about three times their total for 2025. Defence and dual-use firms took nearly 30% of the $8.4bn invested in Israeli high-tech over the period (Israeldesks, secondary, citing forum data). An earlier Viola Ventures analysis counted more than 300 defence-tech startups taking about 4% of Israeli tech funding. It used a different definition, so it should not be combined with the 30% figure (VC Cafe, secondary). On policy, OECD.AI's navigator lists a National Program for AI (2021), an AI for Science Moonshot (2025) and a multi-stakeholder AI regulation forum (2025) (secondary; gov.il unreachable). The research found no AI statute, total AI-investment figure or civilian adoption survey.

Turkey: average usage and a court-ordered chatbot block

Turkey's usage almost exactly matches the world average, at 18.9% against 18.8%, and it is growing quickly. Its 1.5-point gain from Q1 to Q2 2026 is about 8.6% in relative terms, the fastest relative rise among the region's measured countries (Microsoft Q2 2026; the relative rate is a calculation). Its IMF score of 0.540 sits between Saudi Arabia's and South Africa's (IMF). The one regulatory act on record is a court-ordered block of xAI's Grok in July 2025 after the chatbot insulted President Erdoğan and Atatürk (Wikipedia: Grok (chatbot), Wikipedia-only, citing Politico). Turkey is not in Microsoft's earlier technical-paper table; Anthropic's June 2026 release gives it a per-capita index of 0.68 (April) and 0.64 (May 2026) (Anthropic Economic Index data). OECD.AI's navigator lists a National AI Strategy (start year 2021) and 2024 guidance on generative AI and personal data (secondary). The research found no AI law or investment programme, so beyond these items the data is thin. The Grok document covers the other national blocks of that product.

Egypt: Africa's readiness leader and a sovereign AI factory

Egypt's government scores well above its population's measured use. Oxford Insights' 2025 Government AI Readiness Index places Egypt 51st in the world and first in Africa, with a score of 57.5, up from 65th of 188 in 2024. It shares the top Policy Capacity score of 100 with the UK, Serbia and Australia (Egypt Today, secondary). Measured usage is only 15.5%, below the world average, and the IMF score of 0.394 is below every other country profiled here except Nigeria and Ethiopia (Microsoft Q2 2026; IMF). The National AI Strategy 2025-2030 rests on six pillars: governance, technology, data, infrastructure, ecosystem and talent. The targets listed in the table above come from a search summary of the OECD page and were not verified on it (OECD.AI). ITIDA promoted the strategy at AI Everything MEA Egypt in February 2026 (ITIDA).

Two facts matter most for businesses that work with or in Egypt. First, Egypt now has local GPU capacity. On 25 September 2026, Vodafone Business and Cassava launched what they call Egypt's first sovereign AI factory, running Nvidia technology through Cassava, which is described as Africa's first Nvidia cloud partner (Engineering News, 25 Sep 2026, headline dated 25 Sep 2026; article paywalled, not re-read on 12 Oct 2026). Second, Egypt's AI use is heavily technical: in Anthropic's November 2025 data, 39.2% of Egyptian Claude tasks were software-related, against 42.1% for Vietnam and 45.2% for India, the highest of any country (Anthropic India brief). OECD.AI lists national generative-AI, trustworthy-AI and AI-governance guidelines from 2026, but no AI statute was found (OECD.AI, Egypt).

Morocco: "AI Made in Morocco" on top of a 2021 AI-ecosystem strategy

Morocco treats AI as a question of economic sovereignty. OECD.AI's navigator lists a Strategy for Developing an AI Ecosystem (2021) and the Al-Khwarizmi programme (2019) (secondary; OECD.AI, Morocco). The newer "AI Made in Morocco" strategy targets a 100bn-dirham contribution to GDP and 50,000 jobs by 2030, and the minister wants Morocco in the global top 50 by then (Middle East AI News; Hespress, figures from a search summary). Nvidia has confirmed two computing projects in Morocco. The larger is a planned 500 MW Nexus AI Factory near Casablanca on Nvidia Blackwell, led by Nexus Core Systems rather than Cassava (Agenzia Nova, undated; single source; not re-read on 12 Oct 2026). That is planned capacity, not built capacity. Cassava also lists Morocco in its own rollout pipeline (Connecting Africa). Measured use is low for a country with these ambitions: 12.0% of working-age people in June 2026, below Egypt's 15.5%, with an IMF score of 0.429 (Microsoft Q2 2026; IMF; both read 11 Oct 2026), though its Claude per-capita index (0.86 in April, 0.90 in May 2026) is the highest of the African values extracted (Anthropic Economic Index data). No Moroccan AI regulation was found. Elsewhere in North Africa the research has only index values: Libya 15.5%, Tunisia 13.9% and Algeria 13.8% on usage, and Tunisia 0.465 and Algeria 0.370 on the IMF index (same sources).

Jordan and the rest of the Levant: index values only

The research did not cover strategy or compute in Jordan, Lebanon or Syria, but Microsoft's table includes all three. Jordan's 31.2% is just below Saudi Arabia's 31.4% and level with the Czech Republic. Lebanon is at 27.7%. Syria, at 7.4%, has the lowest value of any Middle Eastern or African economy in the table and the only fall in the region (down 0.1 point). Their IMF scores are 0.483 for Jordan, 0.418 for Lebanon and 0.298 for Syria (Microsoft Q2 2026; IMF; both read 11 Oct 2026). Beyond these values, the coverage gap says nothing about the countries' actual activity.

Sub-Saharan Africa and the African Union: strategies outpace compute

Sub-Saharan Africa has more AI policy documents than AI infrastructure. Usage is low, and every profiled country's IMF readiness score is below 0.5 on the 0-to-1 scale. Only Seychelles (0.531) and Mauritius (0.525) score above that line in the subregion (IMF). The main compute build-out is private.

Nigeria: N-ATLaS and a strategy without a law

Nigeria is the clearest African example of a government-backed language model. Press coverage describes N-ATLaS (Nigeria Automatic Transcription and Language Systems) as the first open-source multilingual, multimodal LLM for Yoruba, Hausa, Igbo and Nigerian-accented English (Technology Times, undated). The model card is more precise. The release is one text LLM, a Llama-3 8B fine-tune on about 392 million tokens of instruction data with an 8,092-token context, plus four speech-recognition models, built by Awarri Technologies with NITDA. The repository dates from 19 September 2025, which fits reports of a launch at UNGA80. "Open source" overstates the terms: use is free for research, but organisations above 1,000 users, and any commercial use, need a separate licence from Awarri and the ministry. The card's own evaluation reports weaker results in Yoruba than in Hausa (Hugging Face: NCAIR1/N-ATLaS, read 11 Oct 2026). The government is now trying to turn the model into products. The National AI Innovation Challenge 2026, run by NCAIR with ONDI, asks teams to build on N-ATLaS. Applications close on 12 October 2026, the day after this document's cut-off, and the finals run from 8 to 10 November (MENTERPRISE). The National AI Strategy 2025-2029 provides the policy frame, with three objectives and five pillars (OECD.AI). Nigeria has no standalone AI law, and several AI bills are under deliberation (Digital Policy Alert; attribution inferred from a search summary).

Nigeria's numbers point in opposite directions. Only 10.4% of working-age Nigerians used generative AI in June 2026, up from 8.7% in H1 2025, though Microsoft gives exactly the same values to Ghana and ten other West African economies. Its Claude per-capita index is 0.21 (April) and 0.24 (May 2026), and its IMF score of 0.336 is the second-lowest of the countries profiled here, above only Ethiopia (Microsoft Q2 2026; Anthropic Economic Index data; IMF). Yet Stanford reports that more than 80% of Nigerian employees use AI at work regularly or semi-regularly, more than the 58% global figure (Stanford AI Index 2026, Public Opinion). Both can be true. One measures an employee survey and the other the whole working-age population, so a small, connected, formally employed workforce can be heavy AI users while most of the country is offline (inference). Cassava plans an AI factory in Nigeria, with no date confirmed (Connecting Africa).

Kenya: a Sh152bn plan and single-digit usage

Kenya has a costed national strategy and single-digit usage. The National AI Strategy 2025-2030 is a five-year plan costed at about Sh152bn. It covers governance and talent, and it targets adoption in health, agriculture, education and public services. It also names a skills shortage and a financing gap (The Standard, undated). The financing gap means the Sh152bn is a plan cost, not a budget appropriation, and the research did not convert it to US dollars. Usage of 8.9% is about one-eighth of the UAE's 73.3% (calculation). It grew by 0.2 points over the quarter, the smallest gain among the countries profiled here, and Ethiopia and Rwanda now sit only about a point below it (Microsoft Q2 2026). Readiness is middling. Kenya's IMF score is 0.445, ahead of Egypt and Nigeria, and Oxford Insights places it 65th (IMF; cadeproject, secondary). Cassava names Kenya as its next AI-factory site after South Africa (Connecting Africa). Microsoft's planned $1 billion data centre in Kenya was reported in May 2026 to have stalled over power capacity (headlines, two outlets: Windows Central; Tom's Hardware). Kenya's AI Bill 2026 (Senate Bill No. 4, sponsored by Senator Karen Nyamu) would set four risk tiers and create an Office of the AI Commissioner and an AI Authority (HapaKenya summary, 16 Mar 2026; secondary). Its parliamentary status is unconfirmed: secondary reports conflict on whether it has been tabled, and the Senate order paper was not checked.

South Africa: the continent's first AI factory and an unsettled policy

South Africa leads Sub-Saharan Africa on usage and supercomputing. Usage is 23.9%, above the world average and India's 18.5%. Its IMF score of 0.497 is the best of the larger African economies, behind only Seychelles and Mauritius, and it holds the only TOP500 top-100 system on the continent (Microsoft Q2 2026; IMF; TOP500). Stanford lists it with the UAE and Chile as the places where AI engineering skills are growing fastest (Stanford AI Index 2026). It was also the first country to receive Cassava's Nvidia-powered AI factory. That rollout has begun, after missing its original June 2025 target, and Nigeria, Kenya, Egypt and Morocco are next (Connecting Africa; ITWeb). Its policy position is weaker than its metrics. Cabinet approved a Draft National AI Policy for publication on 25 March 2026, and it appeared as Government Gazette Notice 3880 on 10 April 2026 (dates per Wikipedia, citing SAnews and the Gazette). Communications Minister Solly Malatsi withdrew it on 26 April 2026, after a News24 report led an internal review to confirm that its reference list contained fictitious sources, likely AI hallucinations "included without proper verification". No replacement timetable was found (Daily Maverick, 27 Apr 2026, secondary; TimesLive, 27 Apr 2026; Connecting Africa, 28 Apr 2026, both read 11 Oct 2026). Before the withdrawal it proposed a sector-by-sector, multi-regulator model rather than a single AI regulator (Baker McKenzie, 26 Feb 2026, secondary). Oxford Insights' 67th place puts South Africa behind Egypt and Kenya on government readiness (cadeproject, secondary). That fits a country whose technical base runs ahead of its policy framework (inference).

Ghana: Africa's largest new public compute pledge, from one source

Ghana made Africa's largest new public AI compute commitment of 2026, if a single report is confirmed. The strategy launched in Accra in late April 2026 carries $270m in funding, $250m of it for an AI Computing Centre, and presents Ghana as a continental AI hub (Further Africa, 29 Apr 2026, search summary, single source, not checked against an official source). OECD.AI's navigator lists a National AI Strategy 2023-2033, and how the two documents relate is unclear. Ghana's IMF score is 0.425, between Kenya's and Nigeria's, and its usage of 10.4% is identical to Nigeria's in both quarters (IMF; Microsoft Q2 2026; both read 11 Oct 2026).

Rwanda: "Responsible AI Champion" ambitions, little data

Rwanda's National AI Policy (2023, per OECD.AI's listing) aims to make the country Africa's "AI Lab" and "Responsible AI Champion", with emphasis on skills, data ecosystems and public-sector transformation (allAfrica, 25 Apr 2025, search summary). Oxford Insights ranks it 75th (cadeproject, secondary). The indices show the gap between ambition and use. Rwanda's usage of 7.6% is the lowest of the countries profiled here, while its IMF score of 0.437 is close to Kenya's (Microsoft Q2 2026; IMF; both read 11 Oct 2026). No compute figure was found.

Ethiopia: a 2024 AI policy and index values

OECD.AI's navigator lists a National AI Policy (2024) (secondary; text not read; OECD.AI, Ethiopia). The research found no 2026 policy or compute project. The indices give 8.0% usage in June 2026, a value Microsoft also gives seven other East African economies, and an IMF score of 0.254, the lowest of the countries profiled here (Microsoft Q2 2026; IMF; both read 11 Oct 2026). Ethiopia's only other appearance is as a news venue: the Addis Ababa outlet Capital Ethiopia reported the adoption of the African Union's continental strategy (Capital Ethiopia, 17 Jun 2024). This is a gap in coverage and says nothing about activity on the ground.

African Union: a continental framework with conflicting phase dates

The Continental AI Strategy is the reference framework for African national strategies. Ministers adopted it in June 2024, and it runs for a five-year implementation period from 2025 to 2030 (Capital Ethiopia, 17 Jun 2024; OECD.AI). The sources disagree on its phases. One gives a 2025-2026 governance phase, another 2027-2030, and a third a second phase from 2028 after a 2027 review. In February 2025 the AU signed the Paris AI Action Summit declaration alongside 62 countries and the EU, while the US and UK declined to sign (Wikipedia: 2025 in artificial intelligence, Wikipedia-only). The research found no AU announcement in 2026 and no continental compute or language-model programme. A frequently cited count of national strategies already adopted could not be traced to a source and is not used here.

Cassava and Masakhane: private compute and community language models

Africa's compute build-out is mostly private. Cassava Technologies, founded by Strive Masiyiwa, chose Nvidia in March 2025 to build what it called Africa's first AI factory (ITWeb; Dabafinance). It plans to invest up to $720m and deploy 12,000 GPUs across the continent over several years (Africa.com, single source; not re-read on 12 Oct 2026). Cassava says it has now acquired the 12,000 Nvidia GPUs (self-reported, via Connecting Africa). Its AI factory is deployed at South African data centres with CSIR NICIS as a partner, and Nigeria, Kenya, Egypt and Morocco are planned next (Connecting Africa, secondary). Its stated aim is to keep African data in Africa and to support local-language models, starting with Swahili, then Zulu and Afrikaans (Connecting Africa). The rollout began in South Africa and reached Egypt in September 2026 (Engineering News, 25 Sep 2026, headline only). African-language work also comes from the research community. Vukosi Marivate co-founded the Deep Learning Indaba and the Masakhane Research Foundation, an African NLP community (Wikipedia: Vukosi Marivate, search snippet only). The research could not verify the status of other African models, such as Lelapa AI's InkubaLM.

Cross-regional themes

Usage follows income and connectivity, and the gap is widening in absolute terms

The UAE's 73.3% is about eight times Kenya's 8.9% and nearly ten times Rwanda's 7.6%. The gap widened in absolute terms between Q1 and Q2 2026: the UAE gained 3.2 points, Saudi Arabia 2.0 and Qatar 1.7, against 0.4 for Ethiopia and Rwanda, 0.3 for Nigeria and 0.2 for Kenya (Microsoft Q2 2026; calculations). Usage tracks income more closely than readiness. Israel has the region's highest IMF score but trails the UAE and Qatar on reach, and Egypt leads Africa on government readiness while sitting below the world average on use. Yet more than 80% of employees in the UAE, Saudi Arabia, Egypt and Nigeria say they use AI at work (Stanford AI Index 2026, Public Opinion). The divide is therefore between connected formal-sector workers and everyone else, not between workplaces (inference).

Washington's licences set the pace of Gulf compute

Every large Gulf compute project depends on US export permission. The UAE's sequence of authorisation, assurance frameworks, first deliveries and "A5" status is described above. Qatar's Qai and Saudi Arabia's HUMAIN are both reported to need US licences, and the research shows no comparable upgrade for either (Qatar Tribune; TechFinitive). Export control now reaches models too. On 12 June 2026, a US export-control directive (attributed to the Commerce Department in press reports) suspended all access to Anthropic's Fable 5 and Mythos 5 by foreign nationals. Anthropic had to disable both models for every customer, including users in allied countries, until access returned on 1 July (Anthropic, 12 Jun 2026; Al Jazeera, 1 Jul 2026). Two weeks later, on 26 June, OpenAI said the US government would vet users of GPT-5.6 Sol (Washington Post, headline only). A Gulf state that buys US chips and US models therefore takes on US policy risk at both the hardware and the model layer. See the Timeline and Governance documents for the full sequence.

The Gulf sits between Washington and Beijing, but the research mostly shows one side

The evidence shows the Gulf choosing US partners for its core AI stack. That includes Nvidia, AMD, Qualcomm and Cerebras for chips, AWS, Microsoft and Oracle for cloud, and OpenAI for the flagship cluster. European partners are a second layer: MGX's France campus with Mistral, and HUMAIN's Mistral partnership (Wikipedia: MGX; Mistral news). The capital also runs from the Gulf to the US, through MGX's stakes in Stargate LLC and Anthropic, so the alignment rests on mutual stakes, not only on purchases (Wikipedia: Stargate LLC; Anthropic). The only Chinese AI deployment the research found is Baidu's driverless Apollo Go service in Dubai and its planned Abu Dhabi fleet (Wikipedia: Apollo Go, Wikipedia-only). Huawei, Alibaba and other Chinese partnerships in the region were not researched, so the research cannot show how far the Gulf still hedges toward China. The assurance frameworks G42 signed for US-origin chips show that US access comes with conditions on how the capacity is governed. Their terms were not read, so the research cannot say what they require (inference).

Announced gigawatts dwarf verified megawatts

Stargate UAE (1 GW), HUMAIN (about 1.9 GW by 2030 per Bloomberg, and 6 GW by 2034, on different bases), SDAIA's Hexagon (480 MW), Morocco's Nexus (500 MW, single source) and Ghana's $250m centre add up to a very large announced pipeline. As of 12 October 2026, the research confirms no operating megawatt figure for any of them. Stargate UAE's two targets ("by 2026" and Mubadala's Q3 2026) have produced no completion or operating notice. HUMAIN's AMD systems are live and serving customers, according to AMD, but the release gives no capacity. Cassava's factory in South Africa and its September 2026 Egypt launch are the other deliveries (The National, 5 Dec 2025; AMD press release, 31 Aug 2026; Connecting Africa; Engineering News). The two halves of the region also work at different scales: HUMAIN's first Nvidia phase of 18,000 GB300s is 1.5 times Cassava's 12,000-GPU plan for the whole of Africa (Cassava's own figure; it says it has acquired the 12,000 GPUs) (The National, 14 May 2025; Africa.com; calculation). The useful test is a handover or operating notice with a capacity figure, which none of the large projects has yet issued.

Sovereign language models split into Arabic builders and African state and community projects

The region follows three model strategies. Saudi Arabia builds its own Arabic-first model, ALLaM, and puts it in a free consumer app and an operating system (Wikipedia: Humain). The UAE publishes Falcon models under Apache-2.0 and custom licences, and in 2026 its new ones have been small or specialised (Hugging Face API, tiiuae, newest first). Qatar declines to build at all and focuses on commercialising existing models and agents (Qatar Tribune). In Africa, the efforts are a state-backed fine-tune of an open-weight model with a restricted licence (N-ATLaS in Nigeria), a private compute provider's language plans (Cassava's Swahili, Zulu and Afrikaans work) and a research community (Masakhane) (Hugging Face: NCAIR1/N-ATLaS; Connecting Africa). Apart from the self-reported scores on N-ATLaS's model card, the research found no usage figures, independent benchmarks or API terms for any of them. The models signal sovereignty more than they show measured performance.

Strategies are everywhere; binding AI laws are almost nowhere

Across the countries covered here and the AU, the research verified no AI-specific statute in force. The Gulf governs AI through ministers, state companies and US assurance agreements. Turkey acts against specific products in court (Wikipedia: Grok (chatbot)). Nigeria has no standalone AI law, according to one policy tracker (Digital Policy Alert). South Africa withdrew its draft policy in April 2026 (TimesLive, 27 Apr 2026), and Kenya's risk-based AI Bill 2026 is still a bill, with its parliamentary status unclear. The Asia-Pacific document describes binding AI laws in parts of that region for comparison. In practice, the compliance obligations a business deploying AI faces in this region come mostly from contracts, data-hosting arrangements and US export rules, not from local AI statutes (inference).

What this means for your business

The points below are inferences for businesses that operate in, sell to or host workloads in the region. They draw on the facts above; none is a finding of a cited source.

In the Gulf, government programmes set the pace, and Arabic matters. The UAE's two-year target of agentic AI in half of government operations, Qatar's plan to commercialise existing models and build agents, and HUMAIN's work on an OS and apps around ALLaM all point to integration, agent-building and process-redesign work. HUMAIN now also sells GPU capacity as a service. Your users are likely to be experienced already: 73.3% of working-age people in the UAE use generative AI, and more than 80% of employees in the UAE and Saudi Arabia use AI at work (Middle East AI News; Microsoft Q2 2026; Stanford, Public Opinion). The model vendors are building their own delivery teams there. OpenAI hires forward-deployed engineers in Abu Dhabi, and in May 2026 it launched OpenAI Deployment Company, a majority-owned deployment and integration business with more than $4bn of initial investment (OpenAI job board API; Wikipedia: OpenAI, secondary). Sierra, the customer-service agent platform, lists Arabic-language versions of its agent-development roles (Sierra job board API). Your shortlist of implementation partners may therefore include vendor-owned delivery arms. Weigh Arabic-language capability and evidence of government-grade security above model expertise. The research holds no data on Gulf procurement rules, rates or local-partner requirements, so check them directly before you contract. The Companies, OpenAI and Playbook documents give more detail.

Treat US export policy as a continuity risk for anything hosted in the Gulf. UAE capacity now operates inside a "regulated technology environment" defined by US assurance agreements. Saudi and Qatari capacity depends on licences the research cannot confirm. US model controls can also switch off a frontier model for every customer at short notice, as the 19-day suspension of Fable 5 and Mythos 5 showed (see above; Anthropic, 12 Jun 2026). If you operate in the Gulf, or your US data might be processed in Gulf regions, design for portability. Abstract the model layer, keep a tested fallback (a self-hostable open-weight model is the cheapest), pin model versions, and record in contracts where each workload runs. The IRGC threat adds a plain disaster-recovery reason to keep a second region. The Tool stack, Frameworks and Governance documents give the controls.

Use the region's national models selectively and check each licence. ALLaM reaches users through Humain Chat, but the research found no API terms for it. Falcon licences differ by model: the 2023 falcon-7b models and the 2026 Falcon-OCR and Falcon-Perception use Apache-2.0, while Falcon3, Falcon-H1 and Falcon-H1-Tiny use a custom licence (Hugging Face API, tiiuae). N-ATLaS is free only for research-scale use. A deployment above 1,000 users, or any commercial product, needs a separate licence from Awarri and Nigeria's communications ministry (Hugging Face: NCAIR1/N-ATLaS). For Arabic or Nigerian-language automation, test these models against general-purpose frontier and open-weight models on your own tasks before committing. "Sovereign" is a procurement preference, not a quality guarantee. Price and capability comparisons are in Other models and Top models.

In Africa, design for first-time users and government-led programmes. Nigeria's National AI Innovation Challenge, with finals in November 2026, asks builders to work on N-ATLaS, so any product that comes out of it will also need the model's commercial licence. Kenya's strategy sets out where public money is meant to go, namely health, agriculture, education and public services, though it acknowledges a financing gap. Ghana's computing centre, if funded, would add local capacity (MENTERPRISE; The Standard; Further Africa). With population usage of 9-10% in Kenya and Nigeria, most users are first-time adopters, so automation has to suit them and their local languages; much of the funding will come through government programmes, not private enterprise budgets, which the research does not quantify. Cassava's GPUs in South Africa and Egypt give an option for in-continent hosting if your data must stay in Africa.

If you source software work offshore, Egypt is an option. Egypt's Claude use is among the most software-heavy recorded (39.2% of tasks, against 45.2% for India, the highest of any country). It leads Africa on government AI readiness, and it now has a local Nvidia-based AI factory (Anthropic India brief; Egypt Today; Engineering News). For US businesses comparing offshore suppliers of AI-assisted software, Egypt is a plausible alternative to India, alongside Vietnam (see Asia-Pacific and India).

Do not plan around Gulf GPU capacity until its size is known. HUMAIN's AMD capacity is live and sold as GPU-as-a-service, which makes Saudi-resident hosting possible now if your data must stay in Saudi Arabia. The release gives no capacity figure, and Stargate UAE's first 200 MW has no completion notice. Check availability and price directly before you commit, budget from capacity that exists today, and expect short quote validity (see Hardware and Costs).

Questions to ask any vendor or implementation partner: Where will each workload run, and does the contract record it? What is the fallback outside the Gulf, and when was it last tested? Which licence covers any national model used (ALLaM, Falcon, N-ATLaS), and do your user numbers or commercial use trigger a separate one? How was the system tested in Arabic or your users' local languages? What evidence of government-grade security can the team show?

Open questions and unverified items

Sources

1. Microsoft AI Diffusion Report Q2 2026 (PDF)
2. Microsoft AI Diffusion Report Q1 2026 (PDF)
3. arXiv:2511.02781, Microsoft AI diffusion technical paper
4. MediaPost, 22 Sep 2026: Microsoft tracks AI adoption
5. Middle East AI News, 25 Sep 2026: UAE retains top world AI use ranking
6. WAM: UAE ranks first globally in AI adoption
7. IMF DataMapper, AI Preparedness Index (AI_PI)
8. Anthropic Economic Index, September 2025 report
9. Anthropic Economic Index data, June 2026 release (CSV) (country file read)
10. Anthropic Economic Index, India country brief
11. NRIGlobe, March 2026: Israel per-capita Claude usage (different series; not comparable with the June 2026 values)
12. Stanford HAI, AI Index 2026 (landing page)
13. Stanford HAI, AI Index 2026, Economy chapter
14. Stanford HAI, AI Index 2026, Public Opinion chapter
15. TOP500, June 2026 list
16. Wikipedia: Omar Sultan Al Olama
17. The National, 19 Jul 2026: US decision to raise UAE export status
18. Gulf News, 16 Oct 2025 (last updated): Stargate UAE first phase of 200 MW to complete by 2026
19. The National, 5 Dec 2025: Stargate UAE's first phase to be completed in third quarter of 2026
20. The Verge, 6 Apr 2026: Iran threatens OpenAI's Stargate UAE data centre
21. Wikipedia: MGX (company)
22. Wikipedia: G42 (company)
23. Hugging Face API: TII (tiiuae) models by downloads
24. Hugging Face API: TII (tiiuae) models, newest first (read 11 Oct 2026)
25. Wikipedia: Stargate UAE (Stargate LLC article)
26. OpenAI, "Introducing Stargate UAE"
27. Enterprise AM, 23 May 2025: G42 to develop Stargate UAE
28. MIT Sloan Management Review Middle East: UAE Stargate campus targets 200 MW by 2026
29. MEA Tech Watch, 17 Oct 2025: Stargate UAE nears 2026 milestone
30. Hacker News listing: Stargate 2026 headlines
31. Epoch AI: AI data centers data
32. Gigazine, 15 Jan 2026: H200 and MI325X export review for China
33. The Media Leader (French): France AI pledge commitments
34. Wikipedia: Stargate LLC
35. Anthropic, 12 Feb 2026: Series G funding
36. Wikipedia: Cerebras
37. OpenAI job board API
38. Wikipedia: Apollo Go
39. Al Riyadh Daily: Saudi Arabia in Oxford Insights readiness index
40. Wikipedia: Humain
41. MEA Tech Watch, 2 Jan 2026: SDAIA starts work on Riyadh government data centre
42. The National, 14 May 2025: HUMAIN deals with Nvidia, AMD and AWS
43. TechFinitive: HUMAIN signs AI partnerships with AMD, AWS and Nvidia
44. Seeking Alpha: AWS expands deal with HUMAIN
45. Mistral AI news
46. AMD press release, 31 Aug 2026: AMD, Cisco and HUMAIN expand Saudi Arabia's AI infrastructure as AMD Instinct systems go live
47. CryptoBriefing: HUMAIN $3bn data-centre fund
48. trade.gov: Saudi Arabia new data center strategy
49. MIT Sloan Management Review Middle East: 480 MW government data centre
50. OGN News: SDAIA Hexagon data centre
51. Gulf Times: Qai launch a strategic step for AI in Qatar
52. Qatar Tribune: Qai launch, chairman's comments
53. MEA Tech Watch: QIA and Qai
54. Israeldesks: Israeli defence-tech startup funding reaches $3bn
55. VC Cafe: Israel's defence-tech surge
56. Wikipedia: Grok (chatbot)
57. OECD.AI: Egypt National AI Strategy
58. Egypt Today: Egypt tops Africa in Oxford Insights index
59. Engineering News, 25 Sep 2026: Egypt's first national AI factory launched (headline; paywalled, not re-read on 12 Oct 2026)
60. ITIDA: AI Everything MEA Egypt 2026 opening
61. Middle East AI News: AI Made in Morocco
62. Hespress: Morocco AI strategy targets
63. Agenzia Nova: Nvidia confirms two computing projects in Morocco (single source; not re-read on 12 Oct 2026)
64. Connecting Africa: Cassava deploys AI factory in South Africa
65. TimesLive, 27 Apr 2026: Minister Malatsi withdraws draft AI policy
66. Connecting Africa, 28 Apr 2026: SA withdraws AI policy after "fictitious sources" scandal
67. OECD.AI: Nigeria National AI Strategy
68. Technology Times: Nigeria's N-ATLAS
69. Hugging Face: NCAIR1/N-ATLaS model card and licence (read 11 Oct 2026)
70. MENTERPRISE: National AI Innovation Challenge 2026
71. Digital Policy Alert: Nigeria AI governance approach
72. Oxford Insights: Government AI Readiness Index
73. Let's Data Science: Nigeria rises to 72nd in AI readiness
74. The Standard: Kenya bets Sh152 billion on AI
75. cadeproject: Oxford Insights 2025 ranks
76. ITWeb: Cassava to equip data centres with Nvidia tech
77. Further Africa, 29 Apr 2026: Ghana launches $270m AI strategy
78. allAfrica, 25 Apr 2025: Rwanda National AI Policy
79. Capital Ethiopia, 17 Jun 2024: African ministers adopt Continental AI Strategy
80. OECD.AI: Continental Artificial Intelligence Strategy
81. Wikipedia: 2025 in artificial intelligence
82. Dabafinance: Cassava taps Nvidia for Africa's first AI factory
83. Africa.com: Cassava investment and GPU plan (single source; not re-read on 12 Oct 2026)
84. Wikipedia: Vukosi Marivate
85. Anthropic, 12 Jun 2026: Fable and Mythos access
86. Al Jazeera, 1 Jul 2026: US lifts restrictions on Fable and Mythos
87. Washington Post, 26 Jun 2026: US government to vet users of OpenAI's latest model (headline only)
88. Wikipedia: OpenAI
89. Sierra job board API
90. Anthropic Economic Index dataset, Hugging Face (June 2026 release)
91. TechRadar: OpenAI halts Stargate UK project (secondary)
92. silicon.co.uk: Saudi AI data centres (citing Bloomberg) (secondary)
93. cairoscene: HUMAIN to import 18,000 Nvidia AI chips for new data centres (secondary)
94. OECD.AI: Israel national AI policies (OECD.AI listing, secondary)
95. OECD.AI: Türkiye national AI policies (OECD.AI listing, secondary)
96. OECD.AI: Egypt national AI policies (OECD.AI listing, secondary)
97. OECD.AI: Morocco national AI policies (OECD.AI listing, secondary)
98. OECD.AI: Nigeria national AI policies (OECD.AI listing, secondary)
99. OECD.AI: Kenya national AI policies (OECD.AI listing, secondary)
100. OECD.AI: Ghana national AI policies (OECD.AI listing, secondary)
101. OECD.AI: Rwanda national AI policies (OECD.AI listing, secondary)
102. OECD.AI: Ethiopia national AI policies (OECD.AI listing, secondary)
103. HapaKenya, 16 Mar 2026: a summary of the Artificial Intelligence Bill 2026 (secondary)
104. Windows Central: Kenya's president warns on Microsoft's $1 billion AI data centre (headline only)
105. Tom's Hardware: Microsoft's $1 billion Kenya data centre stalls over power capacity (headline only)
106. Daily Maverick, 27 Apr 2026: Malatsi withdraws draft AI policy after hallucination revelations (secondary)
107. Baker McKenzie, 26 Feb 2026: South African AI policy moves towards approval (secondary)
108. Wikipedia: Draft South Africa National Artificial Intelligence (AI) Policy 2026 (Wikipedia-only for the 25 Mar, Notice 3880 and 27 Apr details)