Here’s a number that deserves a double take: OpenAI’s heaviest internal token user now consumes roughly 100 billion tokens every single month. Six years ago, that same benchmark sat at about 100,000 tokens per month. That’s a million-fold increase in demand for AI inference.

And Sam Altman, OpenAI’s CEO, doesn’t think the ride is over. Speaking during a livestream on June 2, Altman suggested the trajectory could repeat itself, projecting another million-fold expansion in token usage ahead.

The numbers behind the noise

Altman revealed that at least one external customer is consuming even more tokens than the company’s own top user. The identity of that customer wasn’t disclosed.

Altman also flagged that token costs have become what he described as “all of a sudden a huge issue” for enterprise customers. It’s apparently the second most common concern businesses raise with OpenAI, trailing only their desire for more streamlined workflows.

The ‘tokenmaxxing’ era

There’s a cultural shift happening inside organizations that use AI heavily. The trend has been dubbed “tokenmaxxing,” a term describing companies actively encouraging employees to measure and maximize their AI token consumption.

Altman pointed to “constant running proactive AI” as the next phase: AI systems that don’t wait for you to type a prompt, running continuously in the background, monitoring data, flagging anomalies, drafting reports, and taking actions autonomously. Each of those background processes consumes tokens, often at scale that dwarfs interactive chat sessions.

What this means for investors

The risk is that Altman’s projection is aspirational, not guaranteed. A million-fold increase from the current baseline would imply individual users or systems consuming on the order of 100 quadrillion tokens per month. Reaching that scale requires not just demand growth but fundamental advances in chip efficiency, energy availability, and model architecture.

Still, the directional signal is hard to ignore. Six years ago, 100,000 tokens per month seemed like a lot. Today, 100 billion barely raises eyebrows inside OpenAI.