The number of inference tokens produced per day still seems to be almost laughably low today. I asked Google Gemini to research about today’s inference situation, and got a reality check of how early we still are in the adoption of AI by consumers and businesses.

How Many Tokens Do We Consume Per Day?

According to my quick Google Gemini Deep Research, the world consumes between 5-8 trillion LLM inference tokens per day. At a output token price of $2 (which is the output token price per 1m tokens of GPT-4.1), this translates to only $16m in output token revenue per day for all inference providers combined.

This already includes self-hosted token production!

As input tokens are also charged, and as these typically make up about 2x of the output token cost, the total inference revenue would be $16m for output and $32m for input tokens, totaling a meagre $48m in global daily inference revenue per day.

Let’s assume one chat request leads to 1000 generated tokens. I probably use LLMs 20 times per day. That would result in 20’000 tokens per day for me as a heavy user. Then in my professional and academic life as an AI engineer and AI researcher, I probably spend and average of 10 million output tokens.

NVIDIA GPU demand.

Contact

Contact me on X at @janspoerer or email me at jan.spoerer@whu.edu if you want to discuss your thoughts about the future of agentic AI and OS integrations.