Microsoft think tank AI Economy Institute on Sept. 28 released findings from its report, "Global AI Diffusion: Trends and Insights for the Second Quarter of 2026," on global AI usage, regional patterns of use and growth in open-weight models.

The report used anonymised Microsoft telemetry data to measure the share of the working-age population, aged 15 to 64, using generative AI products as an AI diffusion indicator. It also reflected operating system and device market share, internet penetration and differences in population size by country in calculating the indicator.

As of June 2026, the global generative AI usage rate among the working-age population rose 1.0 percentage point to 18.8 percent from 17.8 percent in the first quarter. South Korea posted 40.6 percent, up 3.5 percentage points, ranking 12th worldwide. It recorded the biggest increase among major economies for a third consecutive time.

Differences also emerged in the level of AI diffusion and how it is used by region. The gap in AI use between the advanced-economy-focused Global North and the emerging-economy-focused Global South widened to 12.6 percentage points, and the Global South had a relatively higher share of usage for education and learning purposes.

Use of open weights is expanding rapidly. Open-weight models disclose trained weights so developers can modify them to meet their needs or run them in their own environments, making customisation and deployment easier than closed models.

The performance gap with closed models is also narrowing quickly. An independent benchmark study showed a repeated pattern in which leading open-weight models came close within months to levels previously set by new closed models after they set a new best performance. The report said this does not mean open-weight models have reached performance equal to the latest closed frontier models.

Performance gains led to wider actual use. Data released by OpenRouter after analysing more than 100 trillion tokens showed open-weight models accounted for about one-third of token usage in the dataset at the end of 2025. Increases in usage after the release of major models such as DeepSeek, Kimi, Qwen and GPT-OSS also persisted rather than proving temporary.

A recent analysis also confirmed expanding use of open-weight models. Microsoft’s additional analysis of OpenRouter’s public data found open-weight models accounted for 75 percent of total token usage on OpenRouter. The report said OpenRouter is a developer-focused platform, so the figure cannot be interpreted as global market share.

The report said AI application development activity is also rising rapidly. Hugging Face Spaces, which provides runnable AI applications and demos, rose to about 1.46 million as of the end of August 2026 from about 124,000 at the end of 2023. That was an increase of about 83 percent since the first AI diffusion report was published in November 2025.

Microsoft AI Economy Institute plans to broaden the scope of its measurements from the next report to include newly emerging AI tools.