More users, a wider divide

Microsoft's latest map of generative AI use has two findings that belong together. Its researchers estimate that 18.8% of the world's working-age population used a generative AI product in June 2026, up from 17.8% in the first quarter. Use rose in almost every economy the company measured. Yet the estimated gap between the Global North and Global South widened again.

The North reached 28.8%, up from 27.5%. The South reached 16.2%, up from 15.4%. Both moved forward; the distance between them grew from 12.1 to 12.6 percentage points. That is a gap in estimated user share, not a measure of who benefits most from AI or how well any tool works.

What the index can see

The September 21 report comes from Microsoft's AI Economy Institute. It defines an AI user as someone aged 15 to 64 who used a generative AI product during the period. The estimate starts with aggregated, anonymized Microsoft telemetry and adjusts for differences in device and operating-system share, internet access and population. Its earlier technical paper explains the method across 147 economies.

The country rankings give the trend some shape. Microsoft estimates 73.3% for the United Arab Emirates and 64.3% for Singapore in June. South Korea rose 3.5 percentage points from the first quarter, to 40.6%, the largest absolute gain among the top 25. These are modelled estimates from Microsoft's data, not headcounts collected door to door.

Why use differs

Microsoft also classified a subset of consumer Copilot conversations by purpose. In the Global South, self-directed learning occupied a share 29.2% above its global share; schoolwork and academic support were 32.3% above. Those figures compare the mix of tasks, not the number of people studying with AI. The report notes that the South's younger population may be one reason for the pattern.

It is tempting to read the larger learning share as evidence that AI is closing an education gap. The data cannot show that. It describes early adopters of one product, not what students learned, whether answers were correct or who was unable to get online in the first place. The practical questions remain access, language, cost, connectivity and skills.

A moving measurement window

There is another reason to treat the precise percentages with care. Microsoft's index does not yet include every popular new AI tool. The company says it will widen coverage in the next report and expects that change to lift measured use in nearly every country, particularly China. Much of that lift would be a better count of existing activity, not a sudden wave of new users.

Open-weight models create an additional blind spot when an organization runs them on its own machines or through another service. The report explores developer-platform usage as a separate clue, but says that platform's customers are not representative of all AI users. The index remains useful for following direction over time if its boundaries are kept visible. It cannot, on its own, tell us whether access is fair or whether AI is improving people's lives.

Sources

  1. Microsoft AI Economy Institute: Global AI Diffusion, Q2 2026 Trends and InsightsOriginal 17-page September 2026 report. Executive summary, regional series, country estimates, Copilot task mix, limitations and planned measure expansion.
  2. Microsoft: The continued state of global AI diffusion in 2026September 21 announcement by Juan Lavista Ferres with definition and data source.
  3. Misra et al.: Measuring AI DiffusionOriginal technical method, submitted November 2025. Explains population adjustment, scope and Microsoft-telemetry bias.