Where AI researchers are, and when they come free

What is being measured

Where are AI researchers, country by country?

A country here means where the institution sits. It is a proxy, and it errs consistently in one direction: it undercounts people who have already moved and have not yet published from their new address.

The practical consequence is the one nobody plans for. A subject search run without a region filter returns a list dominated by countries most companies cannot recruit from, and the researcher who could have been hired sits several pages down.

How many AI researchers finish each year?

The end of a doctorate is when a strong researcher becomes reachable, and for several years afterwards they are not. The date is public and scattered across manuscript records, thesis registries, ORCID entries and lab pages. Scattered enough that the field treats it as unknowable.

Read this as a floor. It counts profiles for which we found a dated source, not everyone who is finishing.

One report per country and per subject

The table above is the overview. From 18 September 2026 the index publishes one report a day, first by country and then by subject, each written around what sets that country or subject apart: specialties against the world, rising subjects, newcomers by year, share in industry, the largest institutions.

• The reports, by country and by subject, in the order they are published.
• The interactive atlas, to compare any two countries in a few clicks.
• The aggregates in CSV and JSON, free to reuse with a citation.
• The method: what counts as a researcher, the sources, and the limits.

Method, and the limits of these numbers

The index is built from arXiv, OpenAlex, ORCID and DBLP, and it grows every night, and that has consequences for how these numbers read.

• Counts move as the corpus grows. A rise between two weekly figures can mean more people in the field, or more of them found. Each table states the number of profiles behind it for that reason.
• Shares are computed on what we examined, never on the whole index, and the denominator sits in the caption.
• This measures published research. Engineers who build AI systems without publishing are absent by construction, and they are found through public code instead: how to find and hire AI engineers.

Common questions

Which country has the most AI researchers?

By volume of published AI research, China and the United States are well ahead of every other country, and the distance between them and the country in third place is larger than the distance between third and tenth. Volume and hireable pool are different things, which is what the table below is for.

How many AI PhDs finish each year?

The number rises every year and concentrates in a small set of countries. The figure here counts only profiles for which we found a dated end of doctorate in a public source, so read it as a floor rather than a total.

Why does a researcher's country matter when hiring?

Because a large share of any unfiltered candidate list cannot practically be recruited: wrong timezone, no work authorisation, no intention of relocating. Applying a region filter first usually separates a list you can work from a list you cannot.