‘AI fired’ because of high costs

In March, AI was mentioned as the leading cause of layoffs in the US, accounting for a quarter of more than 60,000 announced layoffs. Globally, according to a survey by Duke University and the US Federal Reserve with 750 financial managers, the number of dismissals attributed to AI in 2026 is forecast to be 9 times higher than in 2025.

Enhancing AI is expected to help businesses improve operational efficiency, save time and streamline structure, but does not mean saving costs.

At the end of April, users Iusuallydrop shared on the social network Reddit that AI is increasingly expensive, forcing his company to adjust its recruitment plan. “We just canceled 5 AI subscription packages and hired two mid-level programmers to replace them,” this person wrote.

His company tested two programmers with the famous question: “I need to wash my car. The store is 100 meters away, should I walk or drive?”. According to normal logic, the driver of course has to drive because the car needs to be at the shop to be cleaned. However, many AI models failed the test when answering that they should walk and gave a series of reasons such as saving time, saving fuel, being good for health and the environment because “the distance is only 100 m”.

Theo Iusuallydroptwo programmers simply chose the driving option, completely without experiencing “hallucination” – the phenomenon of AI creating information that sounds reasonable but is actually inaccurate or misleading.

“New employees answer every question without turning on the ‘7.5 times increase in token usage’ warning. They also know how to make jokes and make the office more fun. The performance to cost ratio is really great. The downside is that their ‘coffee computing’ costs are a bit high,” he said happily.

The post attracted great interest on Reddit with thousands of likes and comments about the issue that AI can also be “fired” because it is too expensive compared to employees.

Previously, many technology leaders shared that the amount of investment in AI was very large, making it difficult for businesses to both use new tools and balance their budget.

Praveen Neppalli Naga, chief technology officer at Uber, said last month that increased use of AI programming tools, especially Anthropic’s Claude Code, caused the company to burn through its entire 2026 AI budget in a matter of months. “I have to start planning again from scratch, because the amount I thought was enough is now exhausted,” he said, but did not reveal the specific number.

Chamath Palihapitiya, CEO of software startup AI 8090, also warned about spending on artificial intelligence. “My expenses triple every three months, while my revenue does not increase proportionally,” he said on the show. All-In Podcast in March.

Bryan Catanzaro, Nvidia’s Vice President of Deep Learning, said above Axios April: “For my team, computing costs far exceed employee costs.”

A survey conducted by the Massachusetts Institute of Technology (MIT) supported Catanzaro’s comments. Specifically, the research team analyzed the necessary technical requirements of AI models to perform work at a level equivalent to humans and found that AI automation is only economically feasible in 23% of positions where vision plays a key role. In the remaining 77% of cases, it is less expensive for humans to do the work.

Theo Futurismsome companies also encountered difficulties after laying off employees to focus on artificial intelligence. Sebastian Siemiatkowski, CEO of buy-now-pay-later company Klarna in Sweden, once declared that AI can “do what humans can do” by 2024. They cooperated with OpenAI to develop virtual assistants, cut staff and use AI to take on the work of 700 customer service staff.

However, he changed his opinion somewhat when customers lost patience with “novice” bots. During the AI ​​adoption period, Klarna’s first quarter loss doubled compared to the same period last year. In May 2025, Siemiatkowski announced plans to recruit more employees for customer care operations.

However, many technology leaders still believe in the effectiveness of investing in AI. In early April, Amos Bar-Joseph, CEO of startup Swan AI, posted on LinkedIn a screenshot of Anthropic’s invoice for a total of 113,421.87 USD, double the previous month’s invoice. Notably, this is the cost of AI for 4 employees, or about 28,000 USD per person per month, exceeding the salary of many people.

Bar-Joseph said he has “never been more proud of a bill”, and stated that Swan AI has always invested a lot of money in artificial intelligence. “I want to clarify the real meaning: this is not an action against humans. On the contrary, we will recruit when we reach the limit of what AI can do. We have not reached that limit, even far away,” he told Business Insider.

Although AI usage fees are expensive, Keith Lee, professor of AI and finance at Gordon Business School – Swiss Institute of Artificial Intelligence, predicts things are likely to change in the future, when prices drop sharply.

According to a March report by analyst firm Gartner, the inference process of large language models with a trillion parameters is estimated to decline by more than 90% in the next four years. AI infrastructure, model design and hardware supply are also expected to improve. AI companies could change the way they price their tools, moving from flat subscriptions to pay-as-you-go pricing.

However, the economic feasibility of AI in the future depends on whether the technology proves its value. According to Mr. Lee, AI needs to demonstrate reliability, reduce illusions, reduce the need for human supervision, and effectively integrate into the business structure.