Which Countries Use AI Most? The Economics Behind the Map

The single best predictor of how much a country uses AI is how rich it is. Anthropic put a number on it: a 1% higher GDP per capita is associated with roughly a 0.7% higher AI Usage Index. That one relationship explains most of the global map of AI use. The countries worth studying are the ones it does not explain.

Israel uses Claude seven times as much as its share of the world’s workforce would predict, the highest intensity of any country, on an income that is middling by rich-world standards. Singapore and South Korea also use AI far beyond what their wealth alone would suggest. Meanwhile some of the richest countries on earth, Luxembourg foremost, use it less than their income predicts. The gap between the trend line and the outliers is where the interesting economics live.

Wealth sets the baseline

Plotting AI usage intensity against income gives the clearest picture of the relationship. The vertical axis is the Anthropic AI Usage Index: a country’s share of Claude.ai usage divided by its share of the global working-age population, where 1.0 means exactly proportional. The horizontal axis is World Bank GDP per capita for 2024.

AI usage intensity versus GDP per capita, by country

Item	GDP per capita (US$, 2024)	AI Usage Index (×)
Israel	$54k	7.0×
Singapore	$91k	4.6×
Australia	$65k	4.1×
New Zealand	$49k	4.0×
South Korea	$36k	3.7×
United States	$85k	3.6×
Canada	$54k	2.9×
Switzerland	$104k	2.8×
Luxembourg	$138k	2.7×
United Kingdom	$53k	2.7×
Netherlands	$68k	2.6×
Denmark	$71k	2.3×
Norway	$87k	2.3×
Ireland	$113k	2.3×
Sweden	$57k	2.2×
France	$46k	1.9×
Finland	$53k	1.9×
Belgium	$57k	1.9×
Japan	$32k	1.9×
Germany	$56k	1.8×
Austria	$58k	1.7×

The upward slope is real and it is the main story. Richer countries have the broadband, the devices, the knowledge-work jobs, and the disposable income that AI tools assume. The 2026 Stanford HAI AI Index, drawing on data from the Microsoft AI Economy Institute, found the same correlation in its population adoption figures: generative-AI adoption climbs with GDP per capita almost everywhere.

But a correlation that explains most of the variance still leaves a lot unexplained. The residual, the vertical distance between a country and the trend line, is where national strategy, industry mix, and language show up.

The overperformers: who beats their income

Israel is the cleanest example of an economy that uses AI more than its wealth predicts. Its usage intensity tops the table.

AI usage intensity by country (Anthropic AI Usage Index, August 2025)

Item	Value
Israel	7.0×
Singapore	4.6×
Australia	4.1×
New Zealand	4.0×
South Korea	3.7×
United States	3.6×
Canada	2.9×
Switzerland	2.8×
Luxembourg	2.7×
United Kingdom	2.7×
Netherlands	2.6×
Denmark	2.3×

Three things the overperformers tend to share, none of which is captured by GDP per capita alone:

1. A dense technology-export sector. Israel’s economy is unusually concentrated in software, cybersecurity, and R&D. When a large share of the workforce already writes code or builds products, a tool that accelerates exactly that work spreads fast. Singapore and South Korea have the same engine in different forms: Singapore as a regional tech and finance hub, South Korea through its hardware and software giants.

2. Language access. The leading models were trained on and perform best in English, and they reason most fluently in a handful of high-resource languages. Countries where the working population operates comfortably in English (Israel, Singapore, Ireland, the Nordics) clear a friction barrier that much of the world still faces.

3. Scale and coordination. Small, dense, high-trust economies can move together. Singapore pairs near-universal connectivity with active state direction of AI in government and education. The United Arab Emirates, which leads the Stanford HAI population adoption ranking at 64%, has made AI adoption explicit national policy. State direction does not show up in income, but it shows up in usage.

The underperformers: wealth without usage

The other side of the residual is just as instructive. Luxembourg has the highest GDP per capita in the dataset, yet its usage intensity is mid-pack, well below the level its income would predict. Much of continental Europe sits slightly under the trend too. Income is there; the usage is not, at least not yet.

The likely reasons are the mirror image of the overperformer traits. Economies weighted toward sectors that AI touches less directly (heavy industry, public administration, finance under tight regulation) have fewer obvious daily uses for a coding or writing assistant. Non-English working languages add friction. And a more cautious regulatory posture, while sensible on its own terms, slows the casual experimentation that drives usage upward.

This is the part of the map that should interest anyone thinking about where AI demand grows next. The underperformers are not poor. They have the infrastructure and the income. What they lack is the pull, the concentration of work that AI makes cheaper. As models get better at non-English languages and at the specific tasks these economies run, the residual should shrink.

What this means for the economics of AI

For a country or a policymaker, the lesson is that raw GDP is the baseline, not the lever. The things that move a country above its income trend are buildable: connectivity and devices, a workforce with the skills and language access to use the tools, and a concentration of knowledge work that gives people a daily reason to reach for them. State direction, as the UAE and Singapore show, can pull a country up the curve faster than income alone would.

For anyone selling or building on AI, the map says demand is not uniform even among rich countries, and the highest-intensity markets are not always the largest economies. A small, English-fluent, tech-dense country can out-use a much larger one. The cost side of that demand, what the underlying models actually cost to run per task, is its own question: the AI model tracker and the cost-per-task calculator cover the per-token economics behind the usage on this map.

The single line that fits the data best is still the income line. But the countries worth watching are the ones drawn above it, because they show that AI usage is something a country can choose to grow, not just something that arrives with wealth.

Frequently asked questions

Which country uses AI the most?

By Claude usage intensity relative to workforce size (Anthropic, August 2025), Israel leads at 7.0 times, then Singapore at 4.6 times and Australia at 4.1 times. By population-level generative-AI adoption rate (Stanford HAI / Microsoft, second half 2025), the United Arab Emirates leads at 64%, then Singapore at 60.9%.

Does being a rich country mean using more AI?

Usually, but not always. Anthropic found roughly a 0.7% higher usage index per 1% higher GDP per capita, so income predicts most of the pattern. But Israel, Singapore, and South Korea use AI far more than their income predicts, while some of the wealthiest economies, such as Luxembourg, use it less.

Why does the US rank 24th on AI adoption but high on usage?

The two metrics measure different things. The US is near the top on usage intensity (how heavily its workforce uses AI) but 24th on the share of its whole population that uses generative AI. Smaller, wealthy, often English-speaking or state-directed economies reach a high population adoption rate faster than a large, diverse one.

Sources

• Anthropic (2025). Anthropic Economic Index (September 2025 release; data Aug 4-11, 2025). anthropic.com/research/economic-index-geography
• Stanford HAI (2026). 2026 AI Index Report, Chapter 4: Economy (underlying data: Microsoft AI Economy Institute, 2025). hai.stanford.edu/ai-index/2026-ai-index-report/economy
• International Monetary Fund (2024). AI Preparedness Index (reference year 2023). imf.org/external/datamapper/AI_PI@AIPI
• World Bank (2024). GDP per capita (current US$), indicator NY.GDP.PCAP.CD. data.worldbank.org/indicator/NY.GDP.PCAP.CD