Analysts monitor a glowing global network map with AI icons, charts, and real-time data dashboards.
Microsoft estimates that generative AI usage reached 18.8% of the world’s working-age population in June 2026, as adoption increased across almost every economy during the second quarter. Its Global AI Diffusion Report, released September 21, also finds a widening gap between the Global North and Global South. For IT leaders and developers, the findings provide context for AI adoption planning—but do not measure enterprise deployment, productivity gains, or Copilot market share.

The findings were announced by Microsoft Chief Data Scientist Juan Lavista Ferres on Microsoft’s On the Issues blog. They are Microsoft’s telemetry-based estimates, rather than independently confirmed counts of everyone using AI.

What changed in the second quarter​

Microsoft’s previous report put global usage at 17.8% in the first quarter of 2026, up from 16.3% in the preceding reporting period. That establishes the baseline for June’s 18.8% estimate.

The latest quarter therefore brought an increase of 1 percentage point, equivalent to roughly 5.6% growth relative to the first-quarter level. Microsoft’s September announcement describes the increase as “about 1%”; percentage points are the clearer expression because the comparison is between two population shares.

The country-level findings show several different kinds of movement:

EconomyMicrosoft’s second-quarter finding
United Arab EmiratesLed the adoption ranking at 73.3%.
SingaporeFollowed at 64.3%.
South KoreaRecorded the largest absolute increase, at 3.5 percentage points.
Saudi ArabiaRose from 30th to 25th, the largest climb in ranking position.
JapanRecorded what Microsoft describes as the largest relative increase, adding 2.2 percentage points from a first-quarter level of 22.5%.

These measures answer different questions. A percentage-point increase describes how much the estimated population share grew. Relative growth measures that increase against the previous level. A ranking change describes movement against other economies, without necessarily establishing a large increase in domestic usage.

Microsoft reports usage of 28.8% in the Global North and 16.2% in the Global South—a difference of 12.6 percentage points—and says the gap continued to widen. Broad-based growth can coexist with a widening divide when adoption increases faster in the already higher-usage group.

What Microsoft’s adoption measure captures​

Microsoft defines AI diffusion as the share of people aged 15–64 who used a generative AI product during the reporting period. The estimate comes from aggregated, anonymized Microsoft telemetry, adjusted for differences in operating-system and device market share, internet penetration, and population.

Those adjustments are intended to make comparisons across economies more meaningful than raw telemetry totals would be. The result remains an estimate built from Microsoft’s measurement coverage, rather than a population census.

The denominator is important: 18.8% refers to the working-age population, not all residents, employees, Windows users, or Microsoft customers. The measure also establishes usage, not how frequently people use AI or whether that use produces better work.

For enterprise planning, the practical implication is to keep national adoption statistics separate from organizational readiness. These findings do not establish how many employees have access to an approved AI service, which business processes use it, or whether an organization has achieved a measurable return on deployment.

Copilot conversations offer a narrower view of use​

Microsoft separately examined consumer Microsoft Copilot usage to understand differences in what people do with AI. It found a higher share of education and learning use in the Global South.

This analysis has a different scope from the worldwide adoption estimate. It describes activity within a particular consumer service; it should not be treated as a breakdown of all generative AI use or of Microsoft 365 Copilot activity in workplaces.

For developers considering education-focused products, that finding is a useful signal about the uses represented in Microsoft’s consumer data. It does not, by itself, establish demand across every service or demonstrate improved educational outcomes.

Open-weight models and the next measurement change​

The report also discusses open-weight AI models. Microsoft says these models have captured a growing share of token volume in API usage, and argues that lower access costs and adaptation to local languages and needs could particularly benefit the Global South.

Token volume measures model activity rather than population reach. Growth in that measure cannot be read directly as growth in the number of people using AI. Microsoft also identifies infrastructure, connectivity, and skills as continuing barriers, so greater model availability alone does not establish that access problems have been resolved.

The next report will introduce another important consideration: Microsoft plans to expand the tools covered by its adoption measurement. It expects that expansion to raise estimated usage across almost all economies, with larger projected increases in China.

That means a future increase could reflect both additional adoption and broader measurement coverage. Analysts tracking these figures should establish whether Microsoft provides a comparable historical baseline before interpreting the next jump as an acceleration in real-world usage. The September release measures second-quarter activity; the expanded tool coverage is a planned change for a subsequent report.