What if AI turns out not to be as cheap for businesses as it seems?

In the minds of many executives and shareholders, replacing a human with AI is often seen as saving on a salary. But it seems the numbers are a little more complex.

Firstly, because the cost of AI tools is starting to be higher than anticipated. For example, Uber used up its annual budget for AI tools in just four months this year.

Secondly, the cost looks set to continue rising significantly. Goldman Sachs estimates that token consumption driven by agentic AI will increase 24-fold by 2030. And although the cost per unit of computing power is falling, total consumption is growing faster than that price drop.

Thirdly, despite the growing sophistication of the models, employees have to spend a considerable amount of time fixing faulty code written by AI.

Finally – although the list could be longer – current licence prices may actually be cheaper than they ought to be. The race among large companies to increase their market share is leading to what may be unsustainable ‘dumping’, financed for the time being by investments running into the billions.

“The cost of AI on my team already exceeds that of the staff,” says an executive at Nvidia.

In many companies, incentives are being created to push employees to adopt AI at a rapid pace, as at Amazon, where teams are ranked according to their intensity of AI use.

It is often said that the hype surrounding AI helps to attract investment and inflate valuations, as if the word ‘AI’ were a magic wand that irrationally draws in any investor. A similar logic applies in human resources, and it seems this is beginning to take its effect.