World Bank AI report 2025: strengthening AI foundations
The World Bank AI report compares country groups by what AI needs to work: connectivity, computing, data and skills. Its headline number: high-income countries, home to 17% of the world’s people, produce 87% of notable AI models. Below are the figures on models, data centers, ChatGPT use and job ads, the method of the report and what they mean for a company in Russia.
Key figures
87%
of notable AI models, 86% of AI startups and 91% of the accumulated venture funding of AI startups are in high-income countries, home to 17% of the world’s people (as of July 2025)
77%
of commercial data center capacity (colocation) is in high-income countries, 18% in upper-middle-income countries, 5% in lower-middle-income countries, and less than 0.1% in low-income countries (June 2025)
50%
of the world’s secure internet servers in 2024 were in the US, 41% in other rich countries and 9% in the rest of the world
more than 40%
of global ChatGPT traffic in 2025 came from middle-income countries such as Brazil, India, Indonesia and Vietnam; low-income countries give less than 1%
24%
of internet users in rich countries used ChatGPT in April 2025; 5.8% in upper-middle-income countries, 4.7% in lower-middle-income countries and 0.7% in low-income countries
9 times
growth in the number of job ads that require generative AI skills from 2021 to 2024; in 2024 they were 0.2% of all online job ads
16%
growth in job ads with AI skills from 2021 to 2024 in upper-middle-income countries; 11% in lower-middle-income countries and 2% in rich countries
The report in brief
- Published by
- World Bank
- Published on
- 24 November 2025
- Original title
- Digital Progress and Trends Report 2025: Strengthening AI Foundations
- Length and language
- 136 pages, in English, license CC BY 3.0 IGO
- Checked
- 16 October 2026
- Next edition
- the series comes out every two years (2023 and 2025); there is no 2026 edition as of 16 October 2026
“Realizing AI’s potential hinges on four foundational investments: The “4Cs””World Bank, p. xiv (PDF 16), Main Messages
How the report was made
The report combines its own and third-party data by country income group: AI models, startups and venture investmentventure capital. Venture capital is money that funds and private investors put into young companies in exchange for a share, expecting a few of them to grow very large. as of July 2025 (p. xiii), data center capacity as of June 2025 (p. 39), ChatGPT web traffic in 2025 (p. xiv), online job ads for 2021–2024 (p. 71) and digital skills for 2023.
The analysis framework is called “4C”: connectivity (internet and energy), compute, context (data, models and applications) and competency (skills and talent) (p. 20, fig. 1.5). Separately, the report covers “small AI”, which is affordable and built for ordinary devices, with examples from agriculture, education and health care (p. xiii).
What the report does not measure
The report estimates people’s use of AI from web traffic, while adoption in companies is hardly measured outside a few countries, mainly the US through the BTOS survey. The effect on productivity and jobs is not measured.
The findings are given by country group, not for each country, so the position of a single country does not follow from them.
What this means for a company in Russia
Russia in the report: there are no separate figures for Russia in the AI indicators (models, startups, traffic, data centers). Russia is mentioned twice: 5% of the world’s ICT workforce in 2023, on a par with the United Kingdom (p. 82, PDF 104, next to fig. 5.5), and 6% of the world’s URLs in Russian in 2025 (p. 63, PDF 85, fig. 4.3).
- Almost all notable AI models come from rich countries: 87% (p. xiii). For a company in Russia, this means dependence on foreign providers and their access rules. Keep a fallback: some tasks on your own servers and some in an external service, as in a hybrid setup.
- Job ads that ask for 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. skills grew 9 times, but they are only 0.2% of online job ads (p. 71). Instead of hiring rare experts, train your own staff: corporate training built around your tasks fits here.
- In rich countries, 24% of internet users used ChatGPT (p. xiv), but this is web traffic, not business returns, and the figure has no data on Russia. Measure the effect on your own processes: hours, errors, lead times. The AI impact calculation shows how to do it.
What to compare it with
- Microsoft: 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. 2026: AI use by country, telemetry from one company, a different method
- IEA: key questions on energy and AI: energy and the data centers that computing depends on
- IMF on AI: annual report 2026: another international organization’s view of AI and the economy
Sources and dates
- World Bank, Digital Progress and Trends Report 2025: Strengthening AI Foundations, checked 16 October 2026worldbank.org
- World Bank Open Knowledge Repository, full text (PDF), checked 16 October 2026openknowledge.worldbank.org
- World Bank Open Knowledge Repository, permanent link to the report, checked 16 October 2026hdl.handle.net