Which countries are adopting AI most? It’s not just about GDP
You don’t need data to notice that AI adoption is increasing fast around us - both in how much we use AI tools, and in the complexity of the tools we use.
But how does that vary by country?
There is no perfect dataset that covers this:
Polling data can be comprehensive, but relies on self-reported data and users remembering and reporting accurately
App download or usage data can be noisy, and doesn’t give any visibility on what users are actually using AI tools for
Proprietary data released by one of the AI labs such as Anthropic, Microsoft or OpenAI is often closest to real usage - but even when it is released comprehensively, we know that different models release different audiences
In order to try and bring some of these datasets together, we created an initial composite usage index drawing on data from SensorTower, GitHub, Anthropic and Microsoft.
The most obvious finding is just how closely connected overall usage is to overall economic develop. ChatGPT and Gemini might be freely available for use, but we see a much larger delta here between more and less advanced economies.
That said, GDP per capita doesn’t explain everything.
In our data, we do see a few clear patterns of countries that are ahead of the curve: English-speaking tech hubs like Singapore, Israel or New Zealand; countries that have focussed on a clear national strategy in AI like the UAE, Singapore (again) or France; or the wider South Asian developer ecosystem (Pakistan or India).
By contrast, our data can be noisier for countries that are under performing. It is hard to distinguish whether a country like China or Russia really are lagging behind - or just not showing up in the Western stack that most of our data draws on. Similarly, it is less surprising that we see lags for other countries whose GDP per capita is largely based on oil or resource wealth such as Guyana or Bahrain. Neither of those excuses though really explain why Japan (-7.5 below what we would expect) or Turkiye (-11.0 below what we would expect) lag.
Anthropic’s data for Claude also allows us to get a view at how task sophistication varies across countries - at least among a group of what are likely to be relatively power users to start with.
What we see here is a more complex relationship with GDP:
Higher GDP countries are more likely to use AI for work and in their leisure time, while lower GDP countries are more likely to use it for education
Higher GDP countries tend to use AI in a more collaborative fashion, rather than delegate to it entirely
That said, there is no overall correlation between the average complexity of the task and GDP per capita - suggesting that when people do use AI, they just are just as likely to use it for more complex work
