Is AI use growing, is it getting cheaper, and how fast are people and businesses adopting it? Token volume and prices from OpenRouter; adoption from Microsoft, the US Census, Gallup and Pew.
All three lines start at 100 so they share one axis. If volume rises faster than price falls, spend grows. Only traffic that goes through OpenRouter is counted, so read this as a signal of the direction things are moving.
Weekly tokens · average price per 1M tokens across tracked models · implied spend (volume × price)
12-week rolling slope of Δln(tokens) on Δln(price). Below −1: price cuts grow total spend. Shows correlation, not cause.
Week-over-week token growth (bars) and 4-week growth of the moving average (line), %
Average price reflects both price cuts and the shift to cheaper models. The same-model index keeps last week's model mix fixed, so it moves only with real price changes.
Share of weekly tokens, % — shifts here explain moves in the average price
The solid line is the world: the share of working-age people who used generative AI. Dashed lines are US industries: the share of businesses that used AI in the last two weeks.
World: Microsoft AI Diffusion (% of people aged 15–64). US: Census BTOS (% of businesses using AI in last 2 weeks).
Current use vs expected use in 6 months, with a fitted S-curve projection (dotted)
% of working-age population using generative AI (Microsoft AI Diffusion, via Our World in Data)
Having tried AI is not the same as using it every day. These survey series split users by how often they use it.
% of US employees · daily · a few times a week or more · a few times a year or more
% of US adults who use chatbots, and the share who use them daily