Sam Altman calls AI tokens a ‘kind of fake unit’ for measuring business value

2 weeks ago 20

OpenAI's CEO says token counts miss what companies actually get from AI, even as customers keep choosing token-based pricing

The CEO of the company selling the most famous AI tokens on Earth just told everyone not to take them too seriously.

OpenAI CEO Sam Altman described AI tokens as a “kind of fake unit” when it comes to business value. The comments came during a recent podcast appearance on Cheeky Pint alongside John Collison, dated October 31, 2026, in the research findings.

What Altman actually said

First, a quick primer. In AI, a token is a small chunk of text or data that a model reads or writes. Providers meter usage in tokens and bill accordingly, much like a utility meters electricity in kilowatt-hours.

Altman’s argument is that the meter and the value are two different things. A token count tells a business how much computation happened. It does not tell that business what it got out of the exchange.

Altman did not dismiss tokens entirely. He acknowledged they are technically valid. He also suggested token counts may approach the minimum amount needed to carry information once models strip out waste.

According to Altman, many customers would rather have stable, predictable token-based pricing, even if it means accepting inflated token counts. The alternative, outcome-based pricing, ties cost to results. That sounds fairer in theory but is far harder to forecast.

The rise of tokenmaxxing

Altman’s remarks land in the middle of a 2026 trend that has been dubbed “tokenmaxxing.”

The idea is simple. Companies began treating raw token consumption as a signal of productivity and engagement. More tokens meant more AI adoption, which meant a more forward-thinking workforce, or so the logic went.

The scale of consumption has grown enormously. OpenAI’s top internal user went from approximately 100,000 tokens per month in late 2019 to about 100 billion tokens per month in 2026, according to the research findings.

Amazon and Uber pump the brakes

Big enterprise users have reached a similar conclusion, and they have the budget overruns to prove it.

Amazon and Uber have criticized token-consumption metrics as “deeply flawed” performance indicators. Both are now setting caps on token usage after exhausting budgets earlier in the year, with the caps reported in October 2026.

The underlying complaint is that high token volumes are better at measuring computational cost than productive output. As a result, firms are increasingly treating AI token spend as a cost center rather than a growth driver.

Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

Read Entire Article