Where AI Research Actually Happens, in Six Charts

· 5 min read ·
ai uk sovereignty research talent data

Compute Isn’t the Constraint argued that the UK’s sovereign AI fund is buying the wrong input. That post was prose. This one is the numbers behind it, drawn so you can check them, with the source data under every chart as CSV. The sources are at the end.

Who publishes

Share of global AI publications, 2000 to 2025Three lines. China rises from 5% to 36%. The EU falls from 30% to 12%. The US falls from 28% to 12%.0%10%20%30%40%200020052010201520202025China, 2000: 4.9%China, 2005: 18.3%China, 2010: 28.7%China, 2015: 17.0%China, 2020: 20.6%China, 2025: 35.9%EU27, 2000: 29.5%EU27, 2005: 22.3%EU27, 2010: 23.0%EU27, 2015: 23.2%EU27, 2020: 17.3%EU27, 2025: 12.4%United States, 2000: 27.6%United States, 2005: 22.9%United States, 2010: 14.9%United States, 2015: 15.6%United States, 2020: 18.5%United States, 2025: 12.0%China 35.9%EU27 12.4%US 12.0%ChinaEU27United States
Share of global AI publications by fractional count, OpenAlex via OECD.AI. China’s dip between 2010 and 2015 is in the source data. Download the data (CSV)

The United States and the EU together produced over half of the world’s AI papers in 2000. They produce under a quarter now. Volume moved east and it is not coming back.

Who gets cited

Papers in the 100 most-cited AI papers of the year, by countryGrouped columns for 2021, 2022 and 2024. The US leads with 46, 68 and 64. China has 33, 27 and 41. The UK has 16 in 2022, the only year with a UK figure.0204060802021United States, 2021: 46 papers46China, 2021: 33 papers332022United States, 2022: 68 papers68China, 2022: 27 papers27United Kingdom, 2022: 16 papers162024United States, 2024: 64 papers64China, 2024: 41 papers41United StatesChinaUnited Kingdom
Papers in the year’s 100 most-cited AI papers, by country of the authors’ institutions, Zeta Alpha. The UK figure is only available for 2022. Download the data (CSV)

Volume and influence are different things. The US has never had fewer than 46 of the hundred most-cited papers in a year, and in 2022 had 68. The UK’s 16 in 2022 is a respectable third place, and 11 of those 16 came from one laboratory, Google DeepMind.

Where the people are trained, and where they work

Where elite AI researchers were educated versus where they workPaired horizontal bars for four countries. The US educates 24% of top researchers and employs 59%. China educates 38% and employs 11%. The UK educates 4% and employs 6%. Germany educates 3% and employs 4%.0%20%40%60%United StatesUnited States: 24% of elite researchers did their undergraduate degree here24%United States: 59% of elite researchers work here59%ChinaChina: 38% of elite researchers did their undergraduate degree here38%China: 11% of elite researchers work here11%United KingdomUnited Kingdom: 4% of elite researchers did their undergraduate degree here4%United Kingdom: 6% of elite researchers work here6%GermanyGermany: 3% of elite researchers did their undergraduate degree here3%Germany: 4% of elite researchers work here4%Undergraduate degreeCurrent employer
Elite AI researchers, defined as authors at NeurIPS, ICML and ICLR, by country of undergraduate degree and country of current employer. MacroPolo Global AI Talent Tracker 3.0. Download the data (CSV)

The American lead is imported. US universities produce a quarter of the world’s top AI researchers and US institutions employ six in ten of them. China produces the largest share of anyone and keeps about a tenth.

Which way they move

Net flow of elite AI researchersDiverging bars. The US gains 1,619 researchers net. China loses 1,248. The UK gains 92 and Germany 46.-1,500-1,000-5000+500+1,000+1,500United StatesUnited States: net gain of 1,619 elite researchers+1,619ChinaChina: net loss of 1,248 elite researchers-1,248United KingdomUnited Kingdom: net gain of 92 elite researchers+92GermanyGermany: net gain of 46 elite researchers+46
Net flow of elite AI researchers, employer country minus undergraduate country. Same source as above. Download the data (CSV)

The UK is on the right side of this chart by 92 people. It is the smallest positive number that still counts as a win, and the 65% retention rate behind it is the figure a fund should be trying to move.

What the money looks like

Private AI investment in 2025, billions of US dollarsThree bars. United States 285.9 billion dollars. China 12.4 billion. Europe 4.7 billion, generative AI only.$0bn$100bn$200bn$300bnUnited StatesUnited States: $285.9bn, all private AI$285.9bnChinaChina: $12.4bn, all private AI$12.4bnEuropeEurope: $4.7bn, generative AI only$4.7bn (generative AI only)
Private AI investment in 2025, Stanford AI Index 2026. The Europe figure covers generative AI only and is the comparable number the Index reports. Download the data (CSV)

This is the gravity. A lab funded from a £500m pot competes for people against an industry that raised two hundred and eighty-six billion dollars last year. The researchers it trains are worth more to the companies on the left of the chart than to the state that paid for them.

How far behind the published frontier sits

How far the best open-weight models trail the closed frontier, in monthsRange bars. Epoch Capabilities Index 3 to 4 months. Public benchmarks 4 to 6. Private benchmarks 8 to 10. Training compute 15 to 16.0369121518months behindEpoch Capabilities IndexEpoch Capabilities Index: 3 to 4 months3 to 4 monthsPublic benchmarksPublic benchmarks: 4 to 6 months4 to 6 monthsPrivate benchmarksPrivate benchmarks: 8 to 10 months8 to 10 monthsTraining computeTraining compute: 15 to 16 months15 to 16 months
Lag between the best open-weight models and the closed frontier, by how you measure it. Epoch AI, 2025 and 2026. Download the data (CSV)

Pick the measure that flatters you and open weights are four months behind. Pick the one you cannot game and it is closer to ten. Measure by the compute actually deployed and it is sixteen. A team that trains on the public literature is a generation behind by construction, which is the whole argument of the earlier post in one bar.

The data

Every chart above has a CSV link. The charts and the CSV files come out of the same script, so the numbers in the pictures and the numbers in the files cannot drift apart.

Sources

28 October 2026