AI is the technology that has so far reduced its costs the fastest, surpassing others with great transformative capacity such as DNA or classical computing. This is the conclusion of an analysis by Epoch AI, in which the basis of the measurement is the technical cost of solving a specific task with a fixed quality model, not the price of the token or the price of a subscription.
The study by the organisation shows how since the year 2023, the cost of achieving that specific performance has fallen by 47% each quarter. This means that the same capability is 13 times cheaper after a year.
In comparison with other technologies, AI is becoming four times cheaper than DNA sequencing, 6 times faster than classical computing, 18 times compared to lithium batteries, and 54 times compared to electricity.
The reasons for this reduction are basically three. Firstly, models are more efficient and therefore require less computation to achieve similar results. Also, a better distribution of capacity makes it possible that not all queries activate the same level of computation. Finally, new reasoning techniques increase precision even though they consume more tokens.
Furthermore, Epoch AI highlights a paradox: while the technical cost decreases, in practice, many organisations are clearly spending more on AI, not less. This happens because Artificial Intelligence is increasingly applied to more processes, agents multiply consumption, and the infrastructure is becoming more demanding. Investment in AI includes servers, storage, support, margins, and advanced functions.
