MIT Technology Review Brasil asks whether intelligence is getting cheaper
🎧 Listen to this story
MIT Technology Review Brasil examines what it calls the marginal cost paradox of intelligence: although the price of tokens—the unit billed by generative AI providers—has fallen sharply, companies’ AI bills continue to rise. The article says the reduction has been 90% or more in less than a decade, with models that cost dozens of dollars per million tokens in 2020 now costing a few dollars or, in some cases, cents, depending on the capacity selected.
The text connects this dynamic to the Jevons paradox, observed by economist William Stanley Jevons in 1865: efficiency gains can make a resource cheaper while increasing its total consumption. In the analysis, cheaper tokens stop being scarce and make larger, more complex and continuous tasks viable, such as processing entire documents, knowledge bases and long conversation histories. The article therefore asks whether intelligence is becoming cheaper, or whether only the information used to produce it is.
Why it matters · editorial interpretation
The distinction helps companies understand why falling token prices can coexist with higher bills: broader, more complex and continuous tasks can increase total consumption. For budgeting, tracking the price per token is not enough; organizations also need to consider how much and in what ways AI is used.