The director from Nvidia shocked with the statement: Artificial intelligence costs us more than employees

Ra&ccaronica is poor

Bryan Catanzaro, vice president of applied deep learning at Nvidia, summed up the problem in a single sentence that set off an avalanche of discussion.

​- For my team, the cost of computing power is far greater than the cost of employees – Catanzaro told Axios.

His claim is not an isolated case. Uber’s chief technology officer, Praveen Neppalli Naga, has admitted that he has to return to the negotiating table because the artificial intelligence budget he predicted for the whole year was “already blown”. Equally illustrative is the example of Swan AI CEO Amos Bar-Joseph, who posted on LinkedIn the $113,000 bill he received from Anthropic for using their AI model – for a team of just four people.

Such experiences from the field are also supported by a study by the prestigious MIT from 2024. The researchers concluded that the automation of jobs with the help of artificial intelligence is financially profitable in only 23 percent of cases. In the remaining 77 percent of jobs, human workers are still the cheaper option.

The paradox of mass layoffs

Despiteč with these staggering costs, technology companies continue with mass layoffs, often citing AI as the reason for restructuring. According to Layoffs.fyi, more than 92,000 workers in the technology sector were laid off this year alone. At the same time, these same companies announced investments of approximately $740 billion in AI infrastructure, a 69 percent increase over last year.

Thus, a paradox was created in which companies lay off people in order to allegedly save money, while at the same time investing hundreds of billions in technology that, at least for now, is proving to be more expensive than the workers it replaces.

A long-term bet or a costly mistake?

Many CEOs, however, don’t see high AI bills as a problem, but rather; as a sign of progress. The vision is that the investment will pay off in the long term, when the processes are fully automated. The CEO of Nvidia himself, Jensen Huang, believes that the productivity of his engineers increases with spending, and he expects that an engineer with a salary of $500,000 a year will spend at least $250,000 on AI tokens.

Experts like Keith Lee, a professor at the Gordon School of Business, call this a “short-term mismatch.” According to him, the costs of AI will remain high due to the price of hardware and energy, but it is predicted that in the next four years, the cost of using large language models will drop by more than 90 percent. However, price is not the only factor. The examples of companies like Klarna and Duolingo, which after replacing workers with artificial intelligence saw a decline in service quality and were forced to re-employ people, show that technology must also prove its reliability. It is not enough for AI to become cheaper than humans, it must also become more predictable and reliable in large systems.