The New AI Metric: Token Consumption
Business Insider recently reported on what has become one of the more surreal metrics in modern software engineering: token consumption.
During an OpenAI enterprise event, CEO Sam Altman reflected on how dramatically AI usage has changed over the past six and a half years.
“Six and a half years ago, OpenAI’s top token burner went through 100,000 a month. That was very likely the token leader in the world.”
Today, Altman said:
“The token leader at OpenAI uses about 100 billion tokens a month.”
But the most interesting part of the story wasn’t the number.
It was the admission that someone outside OpenAI is spending even more—something Altman jokingly described as a personal “embarrassment.”
That single remark says more about the state of AI than the headline figure ever could.
For a Trillion-Dollar AI Company, 100 Billion Tokens Is Surprisingly Small
A hundred billion tokens sounds like an impossible amount of usage.
But for a company at the center of the AI revolution, with a valuation approaching the trillion-dollar range, it is actually a surprisingly modest number for the world’s top internal user.
The reason is simple: token consumption is no longer just a measure of chatbot usage. It is becoming an infrastructure metric.
Inside AI-native companies, developers are not just asking questions. They are running coding agents, automated workflows, evaluations, debugging systems, documentation pipelines, research assistants, and AI-powered operations continuously.
At our company, seeing one engineer consume 1–2 billion tokens every month is no longer remarkable.
Modern development workflows involve AI systems reviewing code, generating tests, writing documentation, debugging failures, and running background reasoning tasks. Much of this happens automatically.
Developers are not sitting in front of a chatbot all day typing prompts.
AI has become infrastructure.
Viewed through that lens, 100 billion tokens as the highest usage inside a trillion-dollar AI company feels less like a maximum and more like an early milestone.
OpenAI’s Token Culture Shows How Quickly AI Usage Has Changed
According to the Business Insider report, OpenAI has developed an internal culture around token usage.
The company reportedly maintains a token leaderboard, employees share enormous usage statistics publicly, and because OpenAI builds the models, employees have unusual access to experiment at scale.
The reported numbers are staggering.
Peter Steinberger, creator of OpenClaw, shared a screenshot showing 603 billion tokens consumed over 30 days. The New York Times previously reported that one OpenAI employee reached 210 billion tokens in a single week.
Those numbers make for great headlines.
But the bigger story is what they represent.
The AI industry has moved from humans chatting with models to software systems continuously communicating with other software systems.
The Biggest Token Consumers May Not Be AI Companies
The assumption is that the companies building AI models will also be the largest users.
That may no longer be true.
The largest token consumers of the future may be companies using AI everywhere: software development, customer service, research, analytics, operations, and autonomous workflows.
OpenAI discovering that its biggest token spender is outside the company is not just a funny anecdote.
It is a sign that AI adoption is spreading beyond the labs that created the technology.
The next phase of AI will not be defined by who experiments the most.
It will be defined by who integrates AI most deeply.
The Future May Not Be More Tokens—It May Be Cheaper Tokens
There is another possible future for AI that changes this entire conversation.
The first assumption is that AI progress means organizations consuming ever larger amounts of tokens at ever higher costs.
But a different race is emerging: making intelligence dramatically cheaper.
Chinese AI companies are aggressively competing on efficiency, cost, and accessibility. Companies such as MiniMax represent a different vision of the AI market: intelligence as an affordable commodity rather than an expensive premium product.
If users can access billions of tokens per month for the price of a low-cost subscription, token consumption stops being a major enterprise expense.
It becomes a utility.
When Tokens Become Cheap, Everything Changes
Today, consuming 1–2 billion tokens per month sounds like a serious usage pattern.
But if that level of usage costs only a few hundred dollars per year, it becomes ordinary.
The same way cloud storage, computing power, and internet bandwidth became cheaper over time, AI intelligence may follow the same path.
The question changes from:
“How many tokens can you afford to use?”
to:
“What can you build when tokens are effectively unlimited?”
The First AI Race Was About Intelligence. The Next Race Is About Economics.
The first phase of generative AI was a race for capability.
Who could build the smartest model?
The next phase may be a race for efficiency.
Who can deliver useful intelligence at the lowest cost?
Business Insider noted that while OpenAI celebrates extraordinary usage, other companies are becoming more cautious. Amazon reportedly shut down its token leaderboard, and Uber introduced token caps after executives questioned whether increasing AI spending was justified.
That contrast shows where the industry is heading.
The first AI era rewarded experimentation.
The second AI era will reward optimization.
Companies will ask:
- How much value does each token create?
- Which workflows actually improve productivity?
- Are expensive models being used where cheaper ones would work?
- Can AI become accessible to millions because costs collapse?
The Future Belongs to Cheap Intelligence
As AI becomes a utility rather than a novelty, raw token counts will matter less.
A company burning billions of tokens is not automatically winning.
It may simply be paying a bigger bill.
The real winners will be the companies that transform intelligence into an affordable resource.
The next competitive advantage in AI will not belong to whoever burns the most tokens.
It will belong to whoever makes intelligence cheap enough—and valuable enough—for everyone to use.