Sam Altman stated in Intelligence at Work that the cost of use AI has evolved into «huge issue» for OpenAI customers.
The company's CEO said that businesses are now concerned about rising spending on tokens, as OpenAI seeks ways to make its models more efficient and offer greater value with lower consumption.
According to Altman, the issue appeared unexpectedly in customer discussions. At the beginning of the year, companies did not raise a cost issue and seemed satisfied with the amounts they spent. Now, as he said, the request is clear: customers want more effective use, because in some cases the first quarter spending ran out of budgets throughout 2026.
The pressure is linked to the practice described as tokenmaxxing, i.e. the intense and extensive use of AI tokens with the expectation of increasing productivity. Some company executives felt that greater use of AI by employees could lead to more revenue. Jensen Huang of Nvidia had stated features that he would be seriously concerned if the company's engineers did not utilize artificial intelligence tools to such an extent that the value of the AI tokens they consume would correspond to at least half of their annual salary.
There are already examples of very high consumption. His creator OpenClaw, Peter Steinberger, saw his team spend $1.3 million on OpenAI API tokens within a month, with a total use of 603 billion tokens.
However, intensive use begins to cause problems in some companies. Amazon employees admitted that they were using AI agents for unnecessary tasks to climb the internal AI ranking table. Microsoft reportedly limited the availability of Claude Code licenses due to increased operating costs. At the same time, Uber's CEO admitted that no clear correlation has been established between the extensive adoption of AI tools and the more effective delivery of products.
Despite concerns, Altman estimates that the tokens use will continue to increase. He said that six and a half years ago the largest consumer tokens in OpenAI used 100,000 tokens a month. Today, according to him, this corresponds to the world average per capita monthly use, while the leading OpenAI user reaches about 100 billion tokens per month. He also added, with apparent embarrassment, that someone else uses even more.
This perspective is based on the assumption that tokens prices will decrease faster than their use increase. At the moment, however, some find that AI models can be more expensive than hiring people.
Jevons' paradox describes that the cheaper a resource becomes, the more it is used. In the case of AI, the rise of the most complex AI systems exponentially increases consumption tokens, seemingly faster than efficiency gains in education and model execution.
-
2
-
2
