One hundred billion tokens per month. That’s how much OpenAI’s single biggest API customer is burning through, according to CEO Sam Altman. To put that in perspective, a single token is roughly three-quarters of a word. So we’re talking about the equivalent of approximately 75 billion words generated monthly by just one organization.

And here’s the kicker: Altman says that customer isn’t even the world’s highest consumer of AI tokens. Someone, somewhere, is using even more.

The scale of AI consumption is getting absurd

OpenAI now issues physical plaques, which it calls “Tokens of Appreciation,” to organizations that hit specific consumption milestones. The thresholds are 10 billion, 100 billion, and 1 trillion tokens used through its API.

McKinsey & Company, the consulting giant, recently celebrated crossing the 100 billion token threshold.

But McKinsey’s usage looks modest compared to some developer projects. OpenClaw, a developer initiative, reportedly logged 603 billion tokens in a single month. The cost? North of $1.3 million. OpenAI ended up covering those expenses after the creator joined the company.

Earlier data from February 2024 indicated that OpenAI was generating roughly 100 billion words daily across its entire platform. Using the standard conversion of approximately 0.75 words per token, that translates to about 4 trillion tokens per month across all users. The fact that a single top customer now accounts for 100 billion of those tokens, about 2.5% of the platform’s total output from that earlier benchmark, illustrates just how concentrated heavy usage has become.

Cost concerns are the elephant in the server room

Altman didn’t just drop the 100 billion token figure as a fun fact. He flagged cost concerns as a major issue.

The OpenClaw example is instructive. At $1.3 million per month for a single project, you can see why cost optimization is becoming an existential question for both AI providers and their customers. Most organizations don’t have OpenAI picking up their tab.