Microsoft's Global AI Diffusion Report, Explained With the Data

What the report is, in plain language

The Microsoft AI Economy Institute, part of Microsoft's research group, publishes an AI Diffusion report every few months. Each one answers one narrow question: out of everyone aged 15 to 64 in a country, what share used a generative AI product during the period? It reports the answer as an "AI user share", and publishes the underlying country table as open data. We use the Q2 2026 edition, released in September 2026, plus the earlier periods in the same table.

The figure comes from aggregated, anonymised Microsoft telemetry, not a survey. Because Microsoft's products are used differently around the world, the raw counts are adjusted for operating-system and device market share, internet penetration and population. Microsoft describes the method in a technical paper (see the sources), and says plainly that no single metric is perfect.

What a percentage does and does not mean

• It does: estimate how widely people have tried or used generative AI in a country.
• It does not: measure how often people use it, whether it is for work, how well it is used, or what it does to jobs or output.
• It is not a count of AI capability or investment. Microsoft describes the United States as leading in AI infrastructure and frontier models, yet it ranks 21st on this measure, at 33.0%.

Who is ahead

The leaders are small, highly digitised economies. The UAE and Singapore sit well clear of the rest, followed by Ireland, France and Norway at roughly 49% to 50%. The United States is 21st at 33.0%, and Germany 23rd at 32.3%. Japan is at 24.7% and India at 18.5%. Microsoft draws the lesson that leadership in infrastructure and frontier models does not by itself produce broad use.

Source: Microsoft AI Diffusion dataset, Q2 2026 update; ties on the published value share a rank; changes calculated by us from the dataset's earlier columns. Full table: ai_diffusion.json.

Who is moving fastest

Every economy in the chart rises in every period. South Korea is the standout: 25.9% in the first half of 2025, 30.7%, 37.1%, then 40.6%. Microsoft says it has posted the largest gain of any major economy for three reporting periods running. In Q2 2026 Japan recorded the largest relative increase, from 22.5% to 24.7%, and Saudi Arabia climbed five places to 25th, Microsoft reports.

Among the 96 economies with their own published values, the median gain over the year was 4.6 points. The biggest gains are not all in the poorest or richest places: they include South Korea, the UAE, France, Ireland and Japan.

The North-South gap

The Global North grew by 1.3 points in Q2 2026 and the Global South by 0.8. That widened the gap to 12.6 points, from 10.6 two periods earlier. Microsoft points to electricity, connectivity, skills and local-language support as the constraints in the South.

Microsoft also looked at how Copilot is used. Among its consumer Copilot conversations, self-directed learning is a 29.2% larger share of conversations in the Global South than globally, and schoolwork a 32.3% larger share. Microsoft cautions that this reflects early adopters, that the South is much younger so demographics probably explain part of it, and that the data does not say why.

Open-weight models and the developer picture

The report goes beyond country shares to open-weight models, which anyone can download and adapt. Microsoft's analysis of public OpenRouter data finds open-weight models at 75% of token usage there, up from roughly one third in late 2025. Microsoft warns this is not a global market share: OpenRouter users are unusually likely to try several providers, and models run on a company's own hardware are missing entirely. It also reports Hugging Face Spaces, a library of runnable AI apps, at about 1.46 million, roughly 83% higher than when Microsoft published its first report in late 2025.

What the measure leaves out

• Domestic tools. Microsoft says its current measure omits several widely used Chinese tools, so China's 17.5% is likely understated. Its H2 2025 report also says DeepSeek surged in Russia, Cuba and Belarus, where US services face restrictions. How much use that leaves out of the Russia, Cuba and Belarus figures is not quantified.
• Revisions are coming. Microsoft says its next report will count more tools and expects measured use to rise in almost all economies, most in China. Compare an economy with itself over time with care, and cross-country rankings with more.
• Shared values. In the Q2 table, 45 economies carry a value identical to at least two others, for example 10.4% for 13 economies. Those look like shared regional estimates, and the report does not explain them. We exclude them from our own gain and ranking claims. Microsoft's map also has an "insufficient data" category.
• Telemetry, not a census. The estimate depends on Microsoft's products and its adjustments.

How this connects to jobs

Use is not exposure. Our AI exposure scores describe how much of a country's work looks like what AI tools can do, using official employment data. They are model-estimated, and across 101 economies with their own value and a GDP figure, exposure and use correlate at 0.66, but that falls to 0.02 once national income is held fixed. We examine that in AI exposure vs actual AI use, and the same question at state level in US states: who uses AI more than their job mix predicts. For how exposure ranks across the largest economies, see the G7 ranking.

What to watch next

Microsoft updates the data every three to six months. The next edition should show whether the expanded tool list changes the picture in China and in economies where non-Western tools are popular, and whether the North-South gap keeps widening. We will refresh this page and our data file when it lands.

Spotted an error or know a better source? Report it and we will check it against the source data.