Nvidia VP Bryan Catanzaro has a message for every company cutting jobs for AI: You will not save money by layoffs for AI as the cost of …

Nvidia senior executive Bryan Catanzaro believes that companies that are betting on artificial intelligence (AI) to reduce workforce costs may not be on the ‘right path’. The vice president of applied deep learning at the US-based chip-making giant noted that AI is currently more expensive to run than employing people, and companies that are betting on it are misreading the technology’s economics. In an interview with Axios, Catanzaro said, “For my team, the cost of compute is far beyond the costs of the employees,” pushing back against the growing assumption that layoffs tied to AI adoption will immediately improve company finances. His remarks come at a time when several major technology companies, including Meta and Microsoft, have announced job cuts or buyouts alongside increased investments in AI infrastructure.

How Meta, Microsoft and other companies are choosing AI tools over human workforce

Recent workforce reductions across the tech sector suggest a shift toward automation. Meta said it plans to cut 10% of its workforce, affecting about 8,000 employees, and to cancel thousands of open roles. Microsoft, meanwhile, has offered one of its largest voluntary buyout programs.At the same time, companies are spending more and more on AI. Data from Morgan Stanley shows that Big Tech companies have already spent $740 billion on capital expenditures this year, a 69% increase compared to 2025.The spending surge has come alongside more than 92,000 layoffs so far in 2026, Layoffs.fyi, demonstrating a disconnect between cost-cutting and increased investments in AI systems.Research suggests that AI has yet to become a cost-effective replacement for human workers in many roles. A 2024 study from Massachusetts Institute of Technology found that AI automation made economic sense in only 23% of jobs involving visual tasks. In the remaining 77%, human labour remained the least expensive option.There are also operational risks tied to AI deployment. In one instance, an engineer reported that an AI agent damaged his database and network due to what he described as “overuse,” underlining concerns about reliability and oversight.According to Yale Budget Lab, there is still no broad evidence that AI adoption has led to widespread job displacement, even as companies continue to invest heavily in the technology.Experts say the current imbalance stems from the high cost of infrastructure and energy required to run AI systems. Keith Lee, an AI and finance professor at the Swiss Institute of Artificial Intelligence’s Gordon School of Business, described the situation as temporary.“What we’re seeing is a short-term mismatch,” Lee told Fortune.He noted that AI-related spending could reach $5.2 trillion by 2030, driven by investments in data centres and IT equipment, with the potential to rise further to $7.9 trillion. Meanwhile, the cost of AI software has increased between 20% and 37% over the past year.“As a result, some firms are beginning to reevaluate AI not as a clear cost-saving substitute for labour, but as a complementary tool—at least until the cost structure stabilises,” he said.The economics may shift as technology improves and costs decline. Lee pointed out that the cost of running large AI models could drop by more than 90% over the next four years, driven by improvements in hardware, infrastructure, and model efficiency, as Gartner projects.Pricing models could also shift from flat subscriptions to a usage-based system, which would help providers better align revenue with operating costs, he said.But cost alone will not determine the long-term viability of AI. The technology needs to be consistent, have fewer errors, and integrate into business workflows with minimal supervision.“It’s not just about AI becoming cheaper than humans. It’s about becoming both cheaper and more predictable at scale,” Lee added. For now, Catanzaro’s message remains clear: replacing workers with AI does not automatically translate into savings, as the cost of compute continues to outweigh labour in many real-world scenarios.