Microsoft AI diffusion report, Q2 2026: key figures
The Microsoft AI diffusion report estimates every quarter what share of people aged 15–64 use generative AI, based on anonymized telemetry from its products. In June 2026 it was 18.8% worldwide, 1 percentage point more than in Q1. Below are the key figures by country, the data on Russia with caveats and the lessons for a company.
Key figures
18,8%
of the world population aged 15–64 used generative AI in June 2026; in Q1 it was 17.8%, a rise of 1 percentage point
73,3%
share in the UAE; together with Singapore (64.3%) it leads by a wide margin, followed by Ireland at 49.9% and France at 49.6%
12.6 pp
gap between the Global North (28.8%) and the Global South (16.2%); in the second half of 2025 it was 10.6 percentage points
33,0%
share in the United States, 21st place; Germany 32.3%, India 18.5%, China 17.5%
3.5 pp
gained by South Korea in a quarter, reaching 40.6%; over 12 months the rise is almost 15 percentage points
75%
of tokens on the OpenRouter platform go to open-weight models, up from about a third at the end of 2025; this is not a share of the global market
p. 7, Open-weight models are broadening AI Diffusion section
1/3 of the price
the open model GLM 5.3 Flash is comparable to GPT 5.6 Sol (high) at about a third of the cost, and companies increasingly split tasks between models by price
p. 8, chart Near-frontier intelligence, far lower cost per task
9,9%
share in Russia in Q2 2026; in Q1 it was 9.5%, a rise of 0.4 percentage points
The report in brief
- Published by
- Microsoft AI Economy Institute
- Published on
- 21 September 2026
- Original title
- Global AI Diffusion: Q2 2026 Trends and Insights
- Length and language
- 17 pages and an open data set on GitHub, in English
- Checked
- 16 October 2026
- Next edition
- quarterly; the Q3 2026 report had not come out as of 16 October 2026, and Microsoft promises to widen the list of AI tools it counts
“In June, AI usage was 18.8% of the working-age population worldwide”Microsoft, p. 2, Executive Summary
How the report was made
The report’s metric: the share of people aged 15–64 who used a generative AIgenerative AI. Generative AI is a family of neural networks that create new content: text, images, music, video and program code, from a description in plain language. product during the period (p. 2). Source: aggregated anonymized Microsoft telemetry, adjusted for the share of operating systems and devices, internet penetration and population size. The latest data is for June 2026, covering 147 economies in total (pp. 13–15, AI Diffusiondiffusion model. A diffusion model is a neural network that creates an image, a video or a sound by removing noise from a random picture, step by step. Data Source table).
Separately, Microsoft analyzes user intent from a sample of consumer Copilotcopilot. An AI copilot is an AI assistant built into a work program: it gives hints and prepares drafts, while the decision and the last action stay with a person. conversations and open data from OpenRouter and Hugging Face (p. 7). The table for all countries is open; you can download it and recalculate.
What the report does not measure
This is one company’s telemetry. Countries where Microsoft products and the covered AI services are less common may be undercounted; many new tools, including local Chinese ones, are not yet in the metric (p. 10).
The report does not measure how often or how deeply people use AI, AI use inside companies or the effect on productivity. OpenRouter data does not represent all AI users (p. 7).
What this means for a company in Russia
Russia in the report: Russia appears only in the table of the full report: 9.9% of people aged 15–64 in Q2 2026 against 9.5% in Q1 (p. 15); by the data setdataset. A dataset is a set of data collected and formatted to train or test a model: a table, a folder of pictures, or a body of texts in which every example is described in the same way. on GitHub, the 2025 figures were 7.6% and 8.0% by half-year. This is Microsoft telemetry, while in Russia some foreign AI services are unavailable and local ones are popular, so the figure may understate real use. The report states the undercount of local tools only for China (p. 10); for Russia this caveat is ours.
- The 9.9% figure for Russia (p. 15) should not be taken as a market estimate: it shows the share of users of services that Microsoft can see. It is more reliable to measure how many of your employees already work with AI and why the others have not started. Common reasons are covered on the page on why employees do not use AI.
- Open-weight models account for 75% of tokenstoken. A token is a piece of text that a language model works with: a word, part of a word, or a symbol. on OpenRouter (p. 7). This is not market share, but it is a sign that open modelsopen weights model. An open weights model is a neural network whose parameter files you can download and run on your own hardware. are fit for work tasks. Test them on your own data: how to choose and deploy them is described in the section open models on your own servers.
- An open model solves tasks at the level of a closed one for about a third of the cost (p. 8). Give simple tasks to a cheap model, keep complex ones for a strong model and compare the bill before and after. What makes up the budget is shown in the article how much AI implementation costs.
What to compare it with
- Anthropic Economic Index 2026: another company’s data on which tasks people use AI for
- NBER: how people use ChatGPT: use of one service, also from the company’s internal data
- World Bank: AI foundations 2025: AI use by country group, estimated from web traffic, a different method
Sources and dates
- Microsoft AI Economy Institute, Global AI Diffusion Report, checked 16 October 2026microsoft.com
- Microsoft AI Economy Institute, Global AI Diffusion: Q2 2026 Trends and Insights (PDF), checked 16 October 2026microsoft.com
- Microsoft, AI Diffusion Q2 2026 data set on GitHub, checked 16 October 2026github.com