Countries Leading in AI: Models, Research & Investment by Country

Executive summary

This report ranks the global AI race across three dimensions that are usually reported in isolation: the AI models a country's organisations build (Epoch AI), the AI research it produces (scholarly publications, Our World in Data via OpenAlex) and the private capital it attracts (Stanford AI Index). Read together they tell a sharper story than any single league table. United States dominates deployed models (394, ~55% of the 712 worldwide) and private investment ($240.6bn); China leads research volume (116,434 publications) yet ships fewer frontier systems. Below the duopoly, the UK, France, South Korea, Canada and Germany compete on specialisation and openness. Every figure is sourced and dated; the full multi-source ranking is downloadable as CSV and JSON and free to embed under CC BY 4.0.

Key findings

The ranking — top countries by notable AI models

Notable AI models built since 2020 by organisation country (Sweden highlighted where shown). The chart is capped at the top 15 — the full multi-source table below covers all 23 model-building countries plus research and R&D.

What the data reveals

The United States is not merely ahead — it is the field. With 394 notable AI models since 2020 it accounts for about 55% of the 712 tracked worldwide, roughly 2.6× second-placed China (149) and about 98× the median country. Of every metric Affärslivet tracks, deployed AI-model output is the most concentrated.

Concentration is extreme: the top five countries (United States, China, United Kingdom, France, South Korea) produce roughly 88% of all notable models. Below the US–China duopoly a second tier — the United Kingdom, France, South Korea, Canada, Germany — competes at the frontier's edges.

Research volume and deployed models are two different races — and different countries win them. China leads the world in AI scholarly publications (116,434 in 2024, OWID/OpenAlex) ahead of India (52,054) and the United States (50,810). Yet the US builds far more notable and frontier models. The gap between publishing research and shipping frontier systems is the single most revealing pattern in the data: papers are necessary but not sufficient — compute, capital and elite labs convert research into deployed capability, and those cluster in the US.

Capital tells the same story, louder. US private AI investment reached $240.6bn in 2025 against China's $10.4bn — a roughly 23× gap (OWID, via the Stanford AI Index). Frontier AI is now as much a capital race as a talent race, and the US has both the deepest capital markets and the labs to absorb it.

Open vs closed splits the field on strategy, not just scale. China releases 60.4% of its models with open weights against the United States' 39.6%; France, Germany and Singapore lean furthest open (above 75%). Openness is a deliberate bet — ecosystems and adoption over secrecy — and Europe has made it most decisively.

Sweden ranks 16 of 23 on model output (2 notable models since 2020), with 2,519 AI publications. Its strength is applied AI, research depth and adoption rather than frontier model-building — the Nordic pattern of punching above its weight on quality while leaving the compute-heavy frontier to the giants.

The money: AI investment by country

Private AI investment is the most lopsided metric of all. In 2025 the United States attracted $240.6bn against China's $10.4bn — a roughly 23-to-1 gap (Stanford AI Index, via Our World in Data). Global corporate AI investment surpassed $580 billion in 2025, with generative AI alone capturing nearly half.

Capital is the flywheel: it buys the compute clusters and the salaries that turn research into frontier models, which is why the investment map and the frontier-model map look almost identical.

The research: AI scholarly publications by country

AI research output tells a different story from deployment. Measured by scholarly publications (2024, OWID via OpenAlex), China leads decisively, followed by India and the United States — the volume race and the frontier race have different winners.

The latest frontier models

Frontier models are the largest, most capable systems at the technical edge. The most recent, by publication date:

Who builds the models — leading organisations

The organisations behind the most notable models since 2020. A handful of US and Chinese labs — OpenAI, Alibaba, Google DeepMind, Meta AI — set the pace.

How model output has grown

Notable AI models published per year. The recent plateau in count reflects a shift from many small models to fewer, far larger and more expensive ones.

Full ranking — 23 countries (models, research & R&D)

Every model-building country, cross-referenced with AI research output (OWID) and R&D intensity and researcher density (World Bank). One table, four authoritative sources.

Scoreboard (machine-readable data)

Every headline indicator with its value, period, source and confidence. Free to reuse under CC BY 4.0.

↓ CSV · ↓ JSON

Methodology & verification

Four authoritative, freely-licensed sources, cross-referenced by country. Models: Epoch AI's notable-models database (CC BY 4.0) — models judged notable on objective criteria, counted since 2020 by the building organisation's country (a model spanning several countries is counted once per country); 'frontier' uses Epoch's flag; 'open %' is the share released with open weights. Research: AI scholarly publications from Our World in Data via OpenAlex. Investment: private AI investment from the Stanford HAI AI Index (via OWID). R&D intensity and researchers per million: the World Bank. Nothing is estimated by Affärslivet — every value carries its source and year. Confidence: High.

Data dictionary

Frequently asked questions

Glossary

Notable AI model

A machine-learning model judged significant by Epoch AI on objective criteria — state-of-the-art performance, high citations or historical importance. ↗

Frontier model

One of the largest and most capable AI systems at the leading edge, typically requiring the most training compute. ↗

Open-weights model

A model whose trained parameters are publicly released, letting anyone run, study or fine-tune it. ↗

AI scholarly publication

A peer-reviewed paper on artificial intelligence, indexed here via OpenAlex — a proxy for a country's AI research output. ↗

Private AI investment

Capital invested into AI companies by private investors in a given year — a measure of commercial momentum, tracked by the Stanford AI Index. ↗

R&D intensity

Gross expenditure on research and development as a share of GDP — how much of an economy is devoted to research. ↗

Sources

• Epoch AI — Notable AI Models database (CC BY 4.0)
• Our World in Data — AI publications & investment (CC BY)
• World Bank — R&D expenditure & researchers
• Stanford HAI — AI Index Report 2026
• OECD.AI — AI policy & data observatory

Related reports & data

• AI & Tech Readiness Index
• Innovation Index
• R&D Spending by Country
• High-Tech Exports by Country