AI research output worldwide

Annual scholarly publications on artificial intelligence across 169 countries, shown both as totals and per million people.

Hover a country

Source: Stanford AI Index Report, via Our World in Data. Counts are annual, not cumulative. Publications are attributed by author affiliation, so internationally co-authored work counts for every country involved. What qualifies as an AI publication is defined by keyword and subject classification, which has widened as the field has, so growth partly reflects definitional drift as well as real expansion.

Download CSV
Complete annual series, 2016 to 2024 · every year 2016–2024 observations. The map above shows selected years; this file contains every year.
Free to use. Please cite Stanford AI Index Report, via Our World in Data as the original source, and M. Ikram, drmuhammadikram.com for the compilation.
Scale and intensity give different answers

The map opens on the per-capita view deliberately, because the total view is largely a map of population.

On volume, China leads decisively with over 116,000 AI publications a year, ahead of India and the United States. That ordering is widely reported and reflects the scale of those research systems.

On publications per million people, the ranking transforms. Singapore reaches 765 per million — roughly nine times China’s rate. Small, concentrated research systems dominate a measure that scale cannot buy.

Morocco produces around 1,540 AI publications a year, which places it modestly on the volume map. At 40 per million it sits closer to China’s 82 than the totals suggest, which is a more informative comparison for a country of its size.

Neither measure is the right one. Volume tells you where the field’s centre of mass sits; intensity tells you how deeply AI research has penetrated a given research system.

What this does and does not tell you

Publication counts are not capability. A country can publish extensively on AI without having the compute infrastructure, industrial base or talent retention to translate that into applied capacity. The relationship between academic output and deployed technology is loose.

The definition of AI research has widened. What counts is determined by keyword and subject classification, and as the field has expanded, so has the boundary. Some of the growth shown here is real expansion; some is reclassification of work that would previously have been labelled statistics, optimisation or control theory.

Co-authorship inflates every count. A paper with authors in three countries counts fully for all three. The more internationally collaborative a country’s AI research, the more this raises its figure — which is not a flaw so much as a property worth knowing.

Growth is steep everywhere. Use the year buttons. Almost every country has multiplied its output since 2016, so a rising figure is unremarkable; the question is whether a country is rising faster or slower than the field as a whole.

Read alongside

My own work on artificial intelligence and sustainable development:

Data from the Stanford AI Index Report, via Our World in Data.

← Back to all datasets