AI vs. workers: at Nvidia they admit that “the cost of computing far exceeds the costs of employees”

“For my team, the computing cost is much higher than the cost of the employees,” he explained. Bryan Catanzarovice president of applied deep learning at Nvidia. His approach calls into question a widespread idea: that AI is already, today, a cheaper alternative to human labor.

The context seems to go in the opposite direction. Meta announced cuts of 10% of its workforce, some 8,000 employees, and froze thousands of open positions. Microsoft, for its part, launched the largest voluntary retirement program in its history. In parallel, the sector accumulates more than 92,000 layoffs in 2026, according to Layoffs.fyi, with a rate that already exceeds that of the previous year.

That combination fuels the narrative of a direct replacement by AI. But when looking in detail, the cuts respond to multiple factors: internal restructuring, pressure for profitability after years of accelerated growth and, above all, the need to finance increasingly large investments in artificial intelligence.

In recent months, several technology companies have reorganized teams while redirecting resources toward AI. Google reduced non-priority areas while strengthening its strategy in generative models, and Amazon cut 14 thousand positions across units while accelerating the development of AI-based services for the cloud.

These movements are often read as a direct replacement of workers with algorithms. The available data show another dynamic.

A 2024 Massachusetts Institute of Technology study analyzed tasks where computer vision could replace human labor and found that automation was only economically viable in 23% of cases. Elsewhere, the cost of implementing and operating AI systems exceeded that of maintaining human workers.

The central problem is in the cost structure. Training and running advanced models requires specialized data centers, high-performance chips, and significant power consumption. According to estimates by McKinsey & Company, total spending on AI could exceed 5 trillion dollars by 2030driven primarily by infrastructure and hardware.

Added to this is a less visible factor: the cost of continuous use. Technology expense management firms found that the price of AI tools increased between 20% and 37% in the last year, in part because current subscriptions do not cover the intensive use that companies make.

Even within companies, the budget impact is already generating tensions. Uber CTO Praveen Neppalli Naga acknowledged that the costs associated with scheduling tools with AI exceeded expectations and forced financial plans to be rethought.

There are also technical limits that affect the equation. The models still require human oversight, can make errors, and in some cases lead to operational failures. Engineers reported incidents where automated agents deleted databases or affected networks due to incorrect decisions, adding recovery and control costs.

Experts agree that what is observed today is a temporary mismatch between investment and return. Companies are betting heavily on a technology that still cannot efficiently replace the human labor in economic terms.

Economist Keith Lee, from the Swiss Institute of Artificial Intelligence, suggests that AI currently functions as a complementary tool rather than as a direct substitute for work. In that logic, its value lies in increasing the productivity of existing equipment, not in eliminating it completely.

That scenario could change in the coming years. Gartner projections indicate that the cost of running large-scale models could fall more than 90% in the next four years, as chips improve, the efficiency of the models and the availability of infrastructure.

A change in the business model is also expected. Many companies today offer AI at flat rates that do not reflect the true cost of heavy usage. A migration towards consumption schemes could adjust prices and make the operation more sustainable.

Another key factor will be reliability. To displace large-scale tasks, AI needs to reduce errors, minimize hallucinations, and better adapt to companies’ internal systems. Without that predictability, the cost of human oversight remains a relevant component.

Adoption data show sustained growth, although still partial. By the end of 2025, around 18% of companies were already using AI tools, with an accelerated increase in just a few months. The trend indicates expansion, but also a stage of experimentation where the real impact is still being evaluated.

The turning point will come when two variables converge: a significant drop in costs and a clear improvement in reliability. Until then, the contrast remains. While companies adjust workforces and multiply investments in artificial intelligence, the technology they promote continues, in many cases, to bemore expensive than the workers it is supposedly coming to replace.