LLMs have already taken over the world. Dont be left behind.
You’ve probably already read about how Large Language Models (LLMs) like ChatGPT, Gemini, and others are growing in popularity. Many say they are “taking over the world,” but the truth is even more striking—their impact is already massive, though in ways we often cannot easily quantify.
Understanding Tokens
At the core of this impact is the concept of the token. A token is a unit of text measurement used by companies to calculate usage and costs. Roughly speaking, one token equals about three-fourths of a word. For example:
-
A page of text contains about 225 words, which is nearly 300 tokens.
-
Across the world, there are an estimated 130–190 million books, which equals a total size of 15–20 trillion tokens of text content.
Now compare that to social media: every day, around 500 million tweets are posted. In one year, that’s roughly 200 billion tweets, amounting to about 5 trillion tokens.
Beyond Books and Tweets: Other Human Text Reservoirs
Books and tweets are only two examples of how we measure human text production. To understand the scale of LLMs, it helps to compare them with other massive sources of text:
-
Wikipedia: ~60 million articles across languages, ≈ 40–50 billion tokens.
-
Scientific Papers (PubMed, arXiv, etc.): ≈ a few trillion tokens in total.
-
News Articles: ~100,000 published daily worldwide, ≈ 2–3 trillion tokens per year.
-
Emails: ~300 billion sent daily; even if only 10% contain text, that’s tens of trillions of tokens annually.
-
Messaging Apps (WhatsApp, WeChat, Telegram, etc.): WhatsApp alone sees ~100 billion messages daily, producing several trillion tokens per year.
-
Entire Indexed Web (Google): ~50 billion web pages, together holding hundreds of trillions of tokens.
Taken together, all these sources represent the full spectrum of human knowledge and communication—curated, factual, casual, and real-time. And yet, even when combined, they are now rivaled or exceeded by what LLMs produce.
The Scale of LLM Text Generation
To put costs in perspective, take GPT-5, one of the most advanced (and expensive) models. The price is about $10 per million output tokens. This means you could generate 1 trillion tokens with GPT-5 for only about $10 million.
Given that most text generation happens on cheaper LLMs, and considering the revenues of OpenAI and Anthropic, it can be estimated that each produces on the order of 5,000 trillion tokens per year. Google has publicly mentioned that Gemini alone generates 480 trillion tokens per month. Altogether, this suggests that nearly 20,000 trillion tokens are generated every year by LLMs—an almost unimaginable figure.
But Not All Tokens Become Final Text
It’s important to note that these raw numbers don’t directly translate to published content. A large share of LLM token generation consists of intermediate output—for example, system reasoning steps, multiple drafts, or trial-and-error runs that never reach the end user. Just as you might write several versions of an essay but only submit one, LLMs often produce far more text than what actually gets saved, shared, or published. This means the visible impact in documents, articles, and apps is somewhat smaller than the generation totals suggest—though still enormous.
Why LLM Usage is Surging
Despite their costs, usage of LLMs has exploded because of their usefulness in white-collar work—from drafting documents to coding, research, and creative tasks. And accessibility is not a barrier: even if you don’t pay for ChatGPT, there are many alternatives offering generous free usage. In fact, you can switch between platforms once you hit one’s free tier.
Here are some you can explore:
Closing Thoughts
The numbers make one thing clear: LLMs are no longer futuristic experiments—they are already producing text on a scale that dwarfs the entire written history of humanity. From books and research papers to tweets, emails, and news, their output rivals and exceeds our collective knowledge production, all while driving billions in value for the companies behind them. Whether you use ChatGPT, Gemini, Claude, or any of the many alternatives, the sheer accessibility of these tools explains their rapid rise. The true impact of LLMs may be hard to quantify, but one thing is undeniable—they are reshaping how we create, consume, and think about information faster than any technology before.
.png)
Excellent analysis!
ReplyDelete