As you may already know, AI companies love throwing around big numbers: billions of tokens per minute, trillions per month, etc. But how does that actually compare to us? How much does humanity speak?
Let’s find out.
First: what’s a token?
AI models don’t read words like we do. They break text into smaller chunks called tokens: sometimes a full word, sometimes just a piece of one. Kinda like syllables. How text is split into tokens also depends on the model.
The rough conversion: 1000 tokens ≈ 750 words.
Estimating human speech (in tokens)
The goal here is just to get an order of magnitude. Nothing precise, just coherent enough to get a rough idea.
Words spoken per person per day: ~15,000 (decent global average)
World population: ~8 billion
Total words per year:
Converting to tokens (dividing by 0.75):
That’s ~60 quadrillion tokens per year. Sixty million billion!
Now: how does AI compare?
Numbers from the biggest labs, late 2025:
OpenAI (Dev Day in October 2025): processing over 6 billion tokens/minute → 3.1 quadrillion tokens/year → ~5% of human speech.
Google (CEO earnings remarks from October 2025): over 1.3 quadrillion tokens/month → 15.6 quadrillion tokens/year → ~26% of human speech.
Just two companies are now processing the equivalent of one-third of all words spoken by humanity.
We tend to think of “human scale” as some unreachable ceiling. It’s not. AI is getting close to out-talking us; and these numbers are already hard to wrap your head around. Curious to see where this goes.