Just three years after ChatGPT’s public launch, artificial intelligence has gone from novelty to everyday infrastructure for hundreds of millions of people. But “AI and humanity” is not one story – it is at least three. There is a divide in who uses generative AI day to day. There is a much starker divide in who builds the large-scale AI systems everyone else relies on. And there is a quieter, slower-moving divide in who researches AI – who gets to shape the field’s next decade.
This piece uses three real, publicly downloadable datasets from Our World in Data to trace all three divides, with Australia as a recurring data point in each. The first is estimates of generative AI use by country in 2025, based on anonymised usage data from Microsoft platforms, scaled to each country’s working-age population. The second is a count of large-scale AI systems (notable AI models requiring significant compute to train) built by country, 2019-2025, from Epoch AI. The third is annual scholarly publications on AI by country and region, 2016-2024, from the Center for Security and Emerging Technology (CSET).
genai <- read_csv("genai_usage_by_country_2025.csv")
sys <- read_csv("ai_systems_by_country_2019_2025.csv")
pubs <- read_csv("ai_publications_by_region_2016_2024.csv")In 2025, an estimated 64% of working-age adults in the United Arab Emirates had used a generative AI tool – compared with under 6% in Afghanistan. Australia sits well above the global picture, with an estimated 36.9% of working-age adults using GenAI.
chart1_countries <- c("United Arab Emirates", "Singapore", "Norway", "Ireland",
"France", "Australia", "United Kingdom", "Japan",
"Brazil", "India", "Nigeria", "Ethiopia", "Afghanistan")
c1 <- genai %>%
filter(entity %in% chart1_countries) %>%
mutate(
entity = fct_reorder(entity, ai_user_share),
highlight = if_else(entity == "Australia", "Australia", owid_region)
)
p1 <- ggplot(c1, aes(x = ai_user_share, y = entity, fill = highlight,
text = paste0(entity, ": ", ai_user_share,
"% of working-age adults used\n",
"generative AI in 2025 (", owid_region, ")"))) +
geom_col(width = 0.7) +
scale_fill_manual(values = c(
"Australia" = brand$terracotta,
"Africa" = brand$grey,
"Asia" = brand$gold,
"Europe" = brand$navy,
"North America" = "#6E8898",
"South America" = "#A6A57A"
)) +
labs(x = "Share of working-age adults using generative AI (%)", y = NULL,
fill = NULL,
title = "Generative AI use varies hugely between countries",
caption = "Source: Our World in Data (2025), Estimated share of working-age\nadults who use generative AI") +
theme_minimal(base_size = 11) +
theme(legend.position = "right")
ggplotly(p1, tooltip = "text") %>%
layout(margin = list(l = 10, r = 10, b = 60))| Country | Region | GenAI users (% of working-age adults, 2025) |
|---|---|---|
| United Arab Emirates | Asia | 64.0 |
| Singapore | Asia | 60.9 |
| Norway | Europe | 46.4 |
| Ireland | Europe | 44.6 |
| France | Europe | 44.0 |
| United Kingdom | Europe | 38.9 |
| Australia | Oceania | 36.9 |
| Japan | Asia | 19.1 |
| Brazil | South America | 17.1 |
| India | Asia | 15.7 |
| Nigeria | Africa | 9.3 |
| Ethiopia | Africa | 6.8 |
| Afghanistan | Asia | 5.6 |
Averaging across all 165 countries in the dataset, a clear regional pattern emerges. Europe, North America and Oceania – the world’s wealthiest regions – have roughly double the generative AI uptake of Africa and parts of Asia.
regional <- genai %>%
group_by(owid_region) %>%
summarise(
mean_share = mean(ai_user_share),
min_share = min(ai_user_share),
max_share = max(ai_user_share),
n_countries = n(),
.groups = "drop"
) %>%
mutate(owid_region = fct_reorder(owid_region, mean_share))
p2 <- ggplot(regional, aes(x = mean_share, y = owid_region,
text = paste0(owid_region, "\n",
"Average: ", round(mean_share, 1), "%\n",
"Range: ", min_share, "% – ", max_share, "%\n",
"Countries: ", n_countries))) +
geom_segment(aes(x = min_share, xend = max_share, yend = owid_region),
colour = "grey80", linewidth = 3) +
geom_point(size = 4, colour = brand$navy) +
labs(x = "Share of working-age adults using generative AI (%)", y = NULL,
title = "Average generative AI use by world region (2025)",
subtitle = "Dot = regional average. Grey bar = range across countries in the region.",
caption = "Source: Our World in Data (2025), Estimated share of working-age\nadults who use generative AI") +
theme_minimal(base_size = 11)
ggplotly(p2, tooltip = "text") %>%
layout(margin = list(l = 10, r = 10, b = 60))| Region | Average (%) | Minimum (%) | Maximum (%) | Countries |
|---|---|---|---|---|
| Oceania | 28.2 | 7.3 | 40.5 | 3 |
| Europe | 27.2 | 8.0 | 46.4 | 32 |
| North America | 18.6 | 6.1 | 35.0 | 13 |
| Asia | 17.9 | 5.1 | 64.0 | 40 |
| South America | 14.8 | 9.0 | 22.5 | 13 |
| Africa | 9.7 | 6.3 | 21.1 | 46 |
Within the group of large English-speaking economies often used as Australia’s comparators, Australia’s generative AI uptake (36.9%) sits behind New Zealand but ahead of the United Kingdom, Canada, the United States – and well above the global average across all 165 countries.
peers <- c("Australia", "New Zealand", "United Kingdom", "Canada", "United States")
global_avg <- mean(genai$ai_user_share)
c3 <- genai %>%
filter(entity %in% peers) %>%
mutate(group = if_else(entity == "Australia", "Australia", "Peer economy")) %>%
bind_rows(
tibble(entity = "Global average (165 countries)", code = NA_character_,
year = 2025, ai_user_share = global_avg, owid_region = NA_character_,
group = "Global average")
) %>%
mutate(entity = fct_reorder(entity, ai_user_share))
p3 <- ggplot(c3, aes(x = ai_user_share, y = entity, fill = group,
text = paste0(entity, ": ", round(ai_user_share, 1),
"% of working-age adults"))) +
geom_col(width = 0.6) +
scale_fill_manual(values = c("Australia" = brand$terracotta,
"Peer economy" = brand$navy,
"Global average" = brand$grey)) +
labs(x = "Share of working-age adults using generative AI (%)", y = NULL,
fill = NULL,
title = "Australia's generative AI uptake vs. comparable economies",
caption = "Source: Our World in Data (2025), Estimated share of working-age\nadults who use generative AI") +
theme_minimal(base_size = 11)
ggplotly(p3, tooltip = "text") %>%
layout(margin = list(l = 10, r = 10, b = 60))| Country / group | Category | GenAI users (% of working-age adults, 2025) |
|---|---|---|
| New Zealand | Peer economy | 40.5 |
| United Kingdom | Peer economy | 38.9 |
| Australia | Australia | 36.9 |
| Canada | Peer economy | 35.0 |
| United States | Peer economy | 28.3 |
| Global average (165 countries) | Global average | 17.4 |
Generative AI tools run on a small number of “large-scale” AI systems – notable models that took huge amounts of computing power to train. Epoch AI’s count of these systems shows that, by 2025, 496 had been built worldwide. The United States and China between them account for 415 of those 496 – about 84% – and the gap between them and everyone else has widened every year since 2021. Australia, by contrast, had built precisely one by the end of 2025.
world_total <- sys %>%
filter(entity == "All large-scale AI systems") %>%
select(year, total = cumulative_count)
c4 <- sys %>%
filter(entity %in% c("United States", "China")) %>%
select(year, entity, cumulative_count) %>%
pivot_wider(names_from = entity, values_from = cumulative_count) %>%
left_join(world_total, by = "year") %>%
mutate(`Rest of the world` = total - `United States` - China) %>%
select(-total) %>%
pivot_longer(-year, names_to = "entity", values_to = "cumulative_count") %>%
mutate(entity = factor(entity, levels = c("United States", "China", "Rest of the world")))
p4 <- ggplot(c4, aes(x = year, y = cumulative_count, colour = entity,
text = paste0(entity, ", ", year, ": ", cumulative_count,
" large-scale AI systems (cumulative)"))) +
geom_line(linewidth = 1) +
geom_point(size = 1.6) +
scale_colour_manual(values = c("United States" = brand$terracotta,
"China" = brand$navy,
"Rest of the world" = brand$grey)) +
labs(x = NULL, y = "Cumulative large-scale AI systems built", colour = NULL,
title = "The US and China dominate large-scale AI development",
subtitle = "Cumulative count of notable AI systems built, by country/region, 2019-2025",
caption = "Source: Epoch AI, via Our World in Data (2025), Cumulative number of\nlarge-scale AI systems by country") +
theme_minimal(base_size = 11)
ggplotly(p4, tooltip = "text") %>%
layout(margin = list(l = 10, r = 10, b = 60))| Year | China | United States | Rest of the world |
|---|---|---|---|
| 2019 | 0 | 0 | 1 |
| 2020 | 0 | 2 | 1 |
| 2021 | 3 | 4 | 6 |
| 2022 | 5 | 26 | 12 |
| 2023 | 52 | 74 | 38 |
| 2024 | 112 | 155 | 65 |
| 2025 | 192 | 223 | 81 |
The third divide is about who studies AI. Drawing on annual counts of scholarly AI publications, Asia’s share of the world’s AI research output has climbed from 41% in 2016 to 58% in 2024, while Europe’s share has fallen by about a third (32% to 21%) and North America’s by about a quarter (20% to 14%) over the same period. Australia’s own AI publication count more than doubled (3,060 to 7,605 a year) – but because the rest of the world, and Asia in particular, grew even faster, Australia’s share of global AI research slipped slightly, from 2.2% to 1.8%.
pubs_wide <- pubs %>%
select(entity, year, num_articles) %>%
pivot_wider(names_from = entity, values_from = num_articles)
c5 <- pubs_wide %>%
mutate(
`Rest of the world` = Africa + `South America` + Oceania,
Asia = Asia / World * 100,
Europe = Europe / World * 100,
`North America` = `North America` / World * 100,
`Rest of the world` = `Rest of the world` / World * 100
) %>%
select(year, Asia, Europe, `North America`, `Rest of the world`) %>%
pivot_longer(-year, names_to = "region", values_to = "share") %>%
mutate(region = factor(region, levels = c("Asia", "Europe", "North America", "Rest of the world")))
p5 <- ggplot(c5, aes(x = year, y = share, colour = region,
text = paste0(region, ", ", year, ": ", round(share, 1),
"% of world AI publications"))) +
geom_line(linewidth = 1) +
geom_point(size = 1.6) +
scale_colour_manual(values = c("Asia" = brand$terracotta,
"Europe" = brand$navy,
"North America" = "#6E8898",
"Rest of the world" = brand$grey)) +
labs(x = NULL, y = "Share of global AI publications (%)", colour = NULL,
title = "Asia's share of AI research keeps growing",
subtitle = "Annual scholarly AI publications by region, as a share of the world total, 2016-2024",
caption = "Source: CSET, via Our World in Data (2025), Annual scholarly publications\non artificial intelligence") +
theme_minimal(base_size = 11)
ggplotly(p5, tooltip = "text") %>%
layout(margin = list(l = 10, r = 10, b = 60))| Year | Asia | Europe | North America | Rest of the world |
|---|---|---|---|---|
| 2016 | 40.9 | 32.3 | 19.6 | 7.3 |
| 2017 | 42.4 | 30.9 | 19.7 | 7.1 |
| 2018 | 43.8 | 29.3 | 19.9 | 7.0 |
| 2019 | 44.9 | 28.2 | 20.3 | 6.7 |
| 2020 | 45.1 | 27.7 | 20.4 | 6.8 |
| 2021 | 48.7 | 26.3 | 18.5 | 6.5 |
| 2022 | 53.0 | 24.9 | 15.8 | 6.3 |
| 2023 | 55.3 | 23.5 | 15.1 | 6.2 |
| 2024 | 58.4 | 21.1 | 14.4 | 6.0 |
The other narrative around generative AI revolves around its potential as a great equalizer, with access to AI as a tutor, translator, and research assistant possible to all those connected to the internet. Indeed, as evident from the first graph below, even the mere access to such capabilities is not distributed equally, following the same wealth divisions in the world, nearly identically. Luckily enough, Australia scores significantly above the global median on this metric.
However, the use of AI is only the tip of the iceberg. As far as the development and research related to this technology are concerned, the differences become truly shocking. Firstly, two countries – US and China – have developed 84% of all large-scale AI systems in the world, whereas Australia alone has built only one such system. Secondly, the base of Asian AI researchers has grown to the dominant position compared to the rest of the world, leaving behind Australia which continues to grow but not fast enough to keep up with the global rate of expansion in this domain.
It is important to note that none of the above implies that the whole world, including Australia, will not benefit from generative AI technologies. Usage of the latter in Australia exceeds the world average by several metrics. However, if the trend related to the development of these systems and scientific work stays unchanged, it would mean that “AI and humanity” would become the story of just a few countries.
Our World in Data. (2025). Annual scholarly publications on artificial intelligence [Data set]. Center for Security and Emerging Technology (CSET). https://ourworldindata.org/grapher/annual-scholarly-publications-on-artificial-intelligence
Our World in Data. (2025). Cumulative number of large-scale AI systems by country [Data set]. Epoch AI. https://ourworldindata.org/grapher/cumulative-number-of-large-scale-ai-systems-by-country
Our World in Data. (2025). Estimated share of working-age adults who use generative AI [Data set]. https://ourworldindata.org/grapher/estimated-share-people-generative-ai