For two years, the dominant narrative in Silicon Valley was clear: artificial intelligence (AI) improves efficiency, cuts costs and replaces expensive human labor. But recent reports reveal a far more complicated reality.
Contrary to expectations, data from international research firms shows that companies investing more financial resources in AI than in employee salaries are not actually saving money. Instead, their IT budgets are rising.
Human labor proves more cost-effective
Axios cites senior figures in the U.S. tech industry who say that, in the end, human labor is likely more cost-effective. According to research firm Gartner, global IT spending is expected to reach $6.31 trillion in 2026 — a 13.5% increase from 2025. The rise is driven by continued AI momentum in infrastructure, software and cloud services, covering everything from development to subscription costs.
“For my team,” Bryan Catanzaro, vice president of applied deep learning at Nvidia, told Axios, “the cost of compute is far beyond the costs of the employees.”
Uber Chief Technology Officer Praveen Neppalli Naga also acknowledged this month in an interview with The Information that the company has already exhausted its entire 2026 AI budget simply on coding tools, particularly Anthropic’s Claude.
“I went back to the drawing board because the budget I thought I would need has already evaporated,” Naga said. Although Uber strongly encourages employees to use AI as much as possible, he said he is not considering slowing software engineering hiring “at this stage.”
According to Axios, more shareholders and board members are now demanding that executives prove that massive AI spending actually pays off over time — meaning it must deliver productivity gains or clear returns.
A groundbreaking study from MIT’s Computer Science and Artificial Intelligence Laboratory examined whether it is economically worthwhile to replace workers with AI in computer vision tasks. The results were striking: in only 23% of the tasks studied was AI cheaper than hiring a human. In the remaining 77%, implementation, maintenance and hardware costs significantly exceeded human wages.
“In many cases, the cost of purchasing an AI system, adapting it to organizational needs and maintaining it simply does not justify replacing the worker,” the report stated.
Massive spending on tokens
Several reasons explain why AI is “burning” budgets. Unlike human employees, whose salaries are relatively fixed, every AI query or action carries a cost. Compute expenses and the need for advanced chips such as those produced by Nvidia continue to rise, making each interaction a variable and difficult-to-predict expense. For companies handling millions of requests, “tokens” — the units of information processed by AI models — accumulate into enormous sums.
In addition, because AI systems still suffer from factual errors, companies are forced to invest in an additional layer of skilled workers tasked with reviewing outputs. Instead of reducing headcount, firms often end up with both expensive software and human quality control staff.
In customer service, for example, a human representative can resolve complex issues in minutes, while an AI model requires costly training on company data and may still mislead customers, potentially causing reputational and legal damage. Installing a computer vision system with 8K cameras, local servers and complex algorithms can cost five times more than hiring a skilled quality control worker for three years.
Furthermore, transitioning to AI often requires companies to reorganize their data, sometimes at a cost of millions of dollars. Without properly structured data, AI systems are ineffective, turning the investment into a financial sinkhole.