Microsoft: global AI use hits 18.8% of working-age adults

Microsoft’s latest Global AI Diffusion Report says generative AI usage reached 18.8% of the worldwide working-age population in June 2026, up about one percentage point from the first quarter, with the UAE and Singapore still topping the country rankings.

FUTURE desk — one of the few recurring, method-stated maps of who is actually using generative AI, and it still shows a widening North–South split.

The measure is the share of people ages 15 to 64 who used a generative AI product in the period. The September post’s own wording is “ages of 15 and 64.” The May post in the same series says “between ages 15 and 64.” Same band. Microsoft builds the share from aggregated, anonymized telemetry — usage signals from its own products, with names stripped off and counts rolled up — then adjusts for operating-system and device market share, how many people are online, and population. The adjustment is there so a Windows-heavy sample is not mistaken for a whole country. Microsoft says the method notes live in its AI Diffusion technical paper. File the age band and those adjustments as Microsoft’s. Microsoft also says no single metric is perfect, and this one is no exception. This is a company usage map. It is not a United Nations census.

Who leads, still the September post. The United Arab Emirates is at 73.3% and Singapore at 64.3%. Microsoft says they continue to top the country list. South Korea again had the largest absolute rise, 3.5 percentage points. A percentage point is one step on the share itself: 20% to 23.5% is 3.5 points. Saudi Arabia climbed five places, from 30th to 25th, which Microsoft calls the largest rise in the rankings. Japan had the largest relative rise: up 2.2 percentage points, which Microsoft calls roughly 10% compared with Japan’s first-quarter usage of 22.5%. File those ranks and moves as Microsoft’s. This desk did not rebuild the leaderboard.

The same post says the gap between the Global North and the Global South kept widening. Those are Microsoft’s labels for groups of economies in this series. Usage is now 28.8% in the North and 16.2% in the South. Microsoft’s 7 May 2026 post, by the same author, put first-quarter worldwide usage at 17.8%, the North at 27.5%, and the South at 15.4%. June’s 18.8% sits one percentage point above that 17.8% print. That is the plain reading of “about 1%” on this share. From those two Microsoft prints, the North–South gap moved from 12.1 points to 12.6 points. The half point is this desk’s subtraction, so a reader can see what “widened” means. Microsoft’s September sentence is “continued widening.” It does not print 0.5. The May post had the UAE at 70.1%. The September post has the UAE at 73.3%, still first.

Microsoft says the quarter’s growth was broad: almost every economy saw higher usage. On how people use the tools, the post looks at consumer Microsoft Copilot — Microsoft’s own chat product — and says the clearest difference is a higher share of education and learning use in the Global South. File that as a Copilot insight. It is not a survey of every AI app. The post also says open-weight models are taking a growing share of token volume on application programming interfaces. An open-weight model is one whose learned numbers are public enough for others to run. An application programming interface, or API, is a paid pipe other software uses to call a model. A token is a small chunk of text the model is billed on. Microsoft says that shift could help the Global South by lowering access costs and by making it easier to adapt tools to local languages, while infrastructure, connectivity, and skills still block people. The next report, Microsoft says, will count more of the new tools that appeared this year, and Microsoft expects that wider net to raise the user share in almost every economy, with larger projected increases in China. Today’s 18.8% is this measure, before that expansion.

Do not treat Microsoft telemetry as a United Nations census. Do not invent a headcount, a dollar figure, or a claim that every new AI app sits inside the 18.8%. The next report’s wider tool list is a plan. It is not a China number published today. A country rank is Microsoft’s usage print. It is not a verdict on a government’s policy.

Plain English for the rest of the card: working-age population = people ages 15 to 64, Microsoft’s band for this report. generative AI = software that produces new text, pictures, or answers from a prompt. telemetry = usage signals from Microsoft’s own products, aggregated and anonymized. percentage point = one step on the share itself, so 17.8% to 18.8% is one point. Global North / Global South = Microsoft’s own split of economies in this series. open-weight model = a model whose learned numbers others can run. API = a paid pipe that lets other software call a model. token = a small chunk of text the model processes. This filing is Microsoft’s second-quarter 2026 diffusion print.

PRIMARY here: Microsoft On the Issues, 21 Sep 2026, “The continued state of global AI diffusion in 2026,” by Juan Lavista Ferres — Tier A PRIMARY company source, the originating record. The 7 May 2026 post in the same series is the earlier Microsoft print used for the 17.8%, 27.5%, 15.4%, and 70.1% baselines. The June 18.8% share, the about-1% line, the UAE 73.3% and Singapore 64.3% leads, South Korea’s 3.5-point rise, Saudi Arabia’s move from 30th to 25th, Japan’s 2.2-point and roughly-10% rise off a 22.5% first-quarter base, the North 28.8% and South 16.2% figures, the 15-to-64 telemetry method, the Copilot education-and-learning line, the open-weight token-volume line, and the next-report tool expansion are Microsoft-attributed on the September post. The half-point gap change is this desk’s subtraction of those two Microsoft prints. NOT claimed: a UN census, a headcount this desk took, that every new AI tool is inside 18.8%, a China increase already measured, that this desk saw the telemetry, a stock tip, or investment advice. Distinct from the already-filed gates-ai-language-partnership, microsoft-india-south-central-ai, and un-ai-panel-agents.